diff --git a/Cargo.lock b/Cargo.lock
index e2fd29b..0ce3965 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2,6 +2,3913 @@
 # It is not intended for manual editing.
 version = 4
 
+[[package]]
+name = "addr2line"
+version = "0.24.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dfbe277e56a376000877090da837660b4427aad530e3028d44e0bffe4f89a1c1"
+dependencies = [
+ "gimli",
+]
+
+[[package]]
+name = "adler2"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "320119579fcad9c21884f5c4861d16174d0e06250625266f50fe6898340abefa"
+
 [[package]]
 name = "aha"
 version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "base64 0.22.1",
+ "candle-core",
+ "candle-flash-attn",
+ "candle-nn",
+ "candle-transformers",
+ "ffmpeg-next",
+ "image",
+ "minijinja",
+ "num",
+ "openai_dive",
+ "reqwest",
+ "serde",
+ "serde_json",
+ "tokenizers",
+]
+
+[[package]]
+name = "ahash"
+version = "0.8.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5a15f179cd60c4584b8a8c596927aadc462e27f2ca70c04e0071964a73ba7a75"
+dependencies = [
+ "cfg-if",
+ "getrandom 0.3.3",
+ "once_cell",
+ "serde",
+ "version_check",
+ "zerocopy",
+]
+
+[[package]]
+name = "aho-corasick"
+version = "1.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e60d3430d3a69478ad0993f19238d2df97c507009a52b3c10addcd7f6bcb916"
+dependencies = [
+ "memchr",
+]
+
+[[package]]
+name = "aligned-vec"
+version = "0.6.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc890384c8602f339876ded803c97ad529f3842aba97f6392b3dba0dd171769b"
+dependencies = [
+ "equator",
+]
+
+[[package]]
+name = "anyhow"
+version = "1.0.100"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a23eb6b1614318a8071c9b2521f36b424b2c83db5eb3a0fead4a6c0809af6e61"
+dependencies = [
+ "backtrace",
+]
+
+[[package]]
+name = "arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3d036a3c4ab069c7b410a2ce876bd74808d2d0888a82667669f8e783a898bf1"
+dependencies = [
+ "derive_arbitrary",
+]
+
+[[package]]
+name = "arg_enum_proc_macro"
+version = "0.3.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0ae92a5119aa49cdbcf6b9f893fe4e1d98b04ccbf82ee0584ad948a44a734dea"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "arrayvec"
+version = "0.7.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c02d123df017efcdfbd739ef81735b36c5ba83ec3c59c80a9d7ecc718f92e50"
+
+[[package]]
+name = "atomic-waker"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1505bd5d3d116872e7271a6d4e16d81d0c8570876c8de68093a09ac269d8aac0"
+
+[[package]]
+name = "autocfg"
+version = "1.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c08606f8c3cbf4ce6ec8e28fb0014a2c086708fe954eaa885384a6165172e7e8"
+
+[[package]]
+name = "av1-grain"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4f3efb2ca85bc610acfa917b5aaa36f3fcbebed5b3182d7f877b02531c4b80c8"
+dependencies = [
+ "anyhow",
+ "arrayvec",
+ "log",
+ "nom",
+ "num-rational",
+ "v_frame",
+]
+
+[[package]]
+name = "avif-serialize"
+version = "0.8.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "47c8fbc0f831f4519fe8b810b6a7a91410ec83031b8233f730a0480029f6a23f"
+dependencies = [
+ "arrayvec",
+]
+
+[[package]]
+name = "backtrace"
+version = "0.3.75"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6806a6321ec58106fea15becdad98371e28d92ccbc7c8f1b3b6dd724fe8f1002"
+dependencies = [
+ "addr2line",
+ "cfg-if",
+ "libc",
+ "miniz_oxide",
+ "object",
+ "rustc-demangle",
+ "windows-targets 0.52.6",
+]
+
+[[package]]
+name = "base64"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9e1b586273c5702936fe7b7d6896644d8be71e6314cfe09d3167c95f712589e8"
+
+[[package]]
+name = "base64"
+version = "0.22.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
+
+[[package]]
+name = "bindgen"
+version = "0.72.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "993776b509cfb49c750f11b8f07a46fa23e0a1386ffc01fb1e7d343efc387895"
+dependencies = [
+ "bitflags 2.9.4",
+ "cexpr",
+ "clang-sys",
+ "itertools 0.13.0",
+ "proc-macro2",
+ "quote",
+ "regex",
+ "rustc-hash",
+ "shlex",
+ "syn",
+]
+
+[[package]]
+name = "bindgen_cuda"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1f8489af5b7d17a81bffe37e0f4d6e1e4de87c87329d05447f22c35d95a1227d"
+dependencies = [
+ "glob",
+ "num_cpus",
+ "rayon",
+]
+
+[[package]]
+name = "bit-set"
+version = "0.5.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0700ddab506f33b20a03b13996eccd309a48e5ff77d0d95926aa0210fb4e95f1"
+dependencies = [
+ "bit-vec",
+]
+
+[[package]]
+name = "bit-vec"
+version = "0.6.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "349f9b6a179ed607305526ca489b34ad0a41aed5f7980fa90eb03160b69598fb"
+
+[[package]]
+name = "bit_field"
+version = "0.10.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e4b40c7323adcfc0a41c4b88143ed58346ff65a288fc144329c5c45e05d70c6"
+
+[[package]]
+name = "bitflags"
+version = "1.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bef38d45163c2f1dde094a7dfd33ccf595c92905c8f8f4fdc18d06fb1037718a"
+
+[[package]]
+name = "bitflags"
+version = "2.9.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2261d10cca569e4643e526d8dc2e62e433cc8aba21ab764233731f8d369bf394"
+
+[[package]]
+name = "bitstream-io"
+version = "2.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6099cdc01846bc367c4e7dd630dc5966dccf36b652fae7a74e17b640411a91b2"
+
+[[package]]
+name = "built"
+version = "0.7.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "56ed6191a7e78c36abdb16ab65341eefd73d64d303fffccdbb00d51e4205967b"
+
+[[package]]
+name = "bumpalo"
+version = "3.19.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "46c5e41b57b8bba42a04676d81cb89e9ee8e859a1a66f80a5a72e1cb76b34d43"
+
+[[package]]
+name = "bytemuck"
+version = "1.23.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3995eaeebcdf32f91f980d360f78732ddc061097ab4e39991ae7a6ace9194677"
+dependencies = [
+ "bytemuck_derive",
+]
+
+[[package]]
+name = "bytemuck_derive"
+version = "1.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4f154e572231cb6ba2bd1176980827e3d5dc04cc183a75dea38109fbdd672d29"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "byteorder"
+version = "1.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1fd0f2584146f6f2ef48085050886acf353beff7305ebd1ae69500e27c67f64b"
+
+[[package]]
+name = "byteorder-lite"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f1fe948ff07f4bd06c30984e69f5b4899c516a3ef74f34df92a2df2ab535495"
+
+[[package]]
+name = "bytes"
+version = "1.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d71b6127be86fdcfddb610f7182ac57211d4b18a3e9c82eb2d17662f2227ad6a"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "candle-core"
+version = "0.9.1"
+source = "git+https://github.com/huggingface/candle.git#1a699fb2018d316dd5f6c1c179588e4cfcd7888e"
+dependencies = [
+ "byteorder",
+ "candle-kernels",
+ "cudarc",
+ "float8 0.4.2",
+ "gemm 0.17.1",
+ "half",
+ "memmap2",
+ "num-traits",
+ "num_cpus",
+ "rand 0.9.2",
+ "rand_distr",
+ "rayon",
+ "safetensors",
+ "thiserror 1.0.69",
+ "ug",
+ "ug-cuda",
+ "yoke 0.7.5",
+ "zip",
+]
+
+[[package]]
+name = "candle-flash-attn"
+version = "0.9.1"
+source = "git+https://github.com/huggingface/candle.git#1a699fb2018d316dd5f6c1c179588e4cfcd7888e"
+dependencies = [
+ "anyhow",
+ "bindgen_cuda",
+ "candle-core",
+ "half",
+]
+
+[[package]]
+name = "candle-kernels"
+version = "0.9.1"
+source = "git+https://github.com/huggingface/candle.git#1a699fb2018d316dd5f6c1c179588e4cfcd7888e"
+dependencies = [
+ "bindgen_cuda",
+]
+
+[[package]]
+name = "candle-nn"
+version = "0.9.1"
+source = "git+https://github.com/huggingface/candle.git#1a699fb2018d316dd5f6c1c179588e4cfcd7888e"
+dependencies = [
+ "candle-core",
+ "half",
+ "libc",
+ "num-traits",
+ "rayon",
+ "safetensors",
+ "serde",
+ "thiserror 1.0.69",
+]
+
+[[package]]
+name = "candle-transformers"
+version = "0.9.1"
+source = "git+https://github.com/huggingface/candle.git#1a699fb2018d316dd5f6c1c179588e4cfcd7888e"
+dependencies = [
+ "byteorder",
+ "candle-core",
+ "candle-nn",
+ "fancy-regex",
+ "num-traits",
+ "rand 0.9.2",
+ "rayon",
+ "serde",
+ "serde_json",
+ "serde_plain",
+ "tracing",
+]
+
+[[package]]
+name = "castaway"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dec551ab6e7578819132c713a93c022a05d60159dc86e7a7050223577484c55a"
+dependencies = [
+ "rustversion",
+]
+
+[[package]]
+name = "cc"
+version = "1.2.38"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "80f41ae168f955c12fb8960b057d70d0ca153fb83182b57d86380443527be7e9"
+dependencies = [
+ "find-msvc-tools",
+ "jobserver",
+ "libc",
+ "shlex",
+]
+
+[[package]]
+name = "cexpr"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6fac387a98bb7c37292057cffc56d62ecb629900026402633ae9160df93a8766"
+dependencies = [
+ "nom",
+]
+
+[[package]]
+name = "cfg-expr"
+version = "0.15.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d067ad48b8650848b989a59a86c6c36a995d02d2bf778d45c3c5d57bc2718f02"
+dependencies = [
+ "smallvec",
+ "target-lexicon",
+]
+
+[[package]]
+name = "cfg-if"
+version = "1.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2fd1289c04a9ea8cb22300a459a72a385d7c73d3259e2ed7dcb2af674838cfa9"
+
+[[package]]
+name = "clang-sys"
+version = "1.8.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b023947811758c97c59bf9d1c188fd619ad4718dcaa767947df1cadb14f39f4"
+dependencies = [
+ "glob",
+ "libc",
+ "libloading",
+]
+
+[[package]]
+name = "color_quant"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3d7b894f5411737b7867f4827955924d7c254fc9f4d91a6aad6b097804b1018b"
+
+[[package]]
+name = "compact_str"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3fdb1325a1cece981e8a296ab8f0f9b63ae357bd0784a9faaf548cc7b480707a"
+dependencies = [
+ "castaway",
+ "cfg-if",
+ "itoa",
+ "rustversion",
+ "ryu",
+ "serde",
+ "static_assertions",
+]
+
+[[package]]
+name = "console"
+version = "0.15.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "054ccb5b10f9f2cbf51eb355ca1d05c2d279ce1804688d0db74b4733a5aeafd8"
+dependencies = [
+ "encode_unicode",
+ "libc",
+ "once_cell",
+ "unicode-width",
+ "windows-sys 0.59.0",
+]
+
+[[package]]
+name = "core-foundation"
+version = "0.9.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91e195e091a93c46f7102ec7818a2aa394e1e1771c3ab4825963fa03e45afb8f"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
+[[package]]
+name = "core-foundation-sys"
+version = "0.8.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "773648b94d0e5d620f64f280777445740e61fe701025087ec8b57f45c791888b"
+
+[[package]]
+name = "crc32fast"
+version = "1.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9481c1c90cbf2ac953f07c8d4a58aa3945c425b7185c9154d67a65e4230da511"
+dependencies = [
+ "cfg-if",
+]
+
+[[package]]
+name = "crossbeam-deque"
+version = "0.8.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9dd111b7b7f7d55b72c0a6ae361660ee5853c9af73f70c3c2ef6858b950e2e51"
+dependencies = [
+ "crossbeam-epoch",
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-epoch"
+version = "0.9.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5b82ac4a3c2ca9c3460964f020e1402edd5753411d7737aa39c3714ad1b5420e"
+dependencies = [
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-utils"
+version = "0.8.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d0a5c400df2834b80a4c3327b3aad3a4c4cd4de0629063962b03235697506a28"
+
+[[package]]
+name = "crunchy"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "460fbee9c2c2f33933d720630a6a0bac33ba7053db5344fac858d4b8952d77d5"
+
+[[package]]
+name = "cudarc"
+version = "0.17.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "72ba848ae5c6f3cb36e71eab5f268763e3fabcabe3f7bc683e16f7fa3d46281e"
+dependencies = [
+ "float8 0.3.0",
+ "half",
+ "libloading",
+]
+
+[[package]]
+name = "darling"
+version = "0.20.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fc7f46116c46ff9ab3eb1597a45688b6715c6e628b5c133e288e709a29bcb4ee"
+dependencies = [
+ "darling_core",
+ "darling_macro",
+]
+
+[[package]]
+name = "darling_core"
+version = "0.20.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0d00b9596d185e565c2207a0b01f8bd1a135483d02d9b7b0a54b11da8d53412e"
+dependencies = [
+ "fnv",
+ "ident_case",
+ "proc-macro2",
+ "quote",
+ "strsim",
+ "syn",
+]
+
+[[package]]
+name = "darling_macro"
+version = "0.20.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fc34b93ccb385b40dc71c6fceac4b2ad23662c7eeb248cf10d529b7e055b6ead"
+dependencies = [
+ "darling_core",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "dary_heap"
+version = "0.3.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06d2e3287df1c007e74221c49ca10a95d557349e54b3a75dc2fb14712c751f04"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "derive_arbitrary"
+version = "1.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e567bd82dcff979e4b03460c307b3cdc9e96fde3d73bed1496d2bc75d9dd62a"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "derive_builder"
+version = "0.20.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "507dfb09ea8b7fa618fcf76e953f4f5e192547945816d5358edffe39f6f94947"
+dependencies = [
+ "derive_builder_macro",
+]
+
+[[package]]
+name = "derive_builder_core"
+version = "0.20.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2d5bcf7b024d6835cfb3d473887cd966994907effbe9227e8c8219824d06c4e8"
+dependencies = [
+ "darling",
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "derive_builder_macro"
+version = "0.20.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ab63b0e2bf4d5928aff72e83a7dace85d7bba5fe12dcc3c5a572d78caffd3f3c"
+dependencies = [
+ "derive_builder_core",
+ "syn",
+]
+
+[[package]]
+name = "displaydoc"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "97369cbbc041bc366949bc74d34658d6cda5621039731c6310521892a3a20ae0"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "dyn-stack"
+version = "0.10.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "56e53799688f5632f364f8fb387488dd05db9fe45db7011be066fc20e7027f8b"
+dependencies = [
+ "bytemuck",
+ "reborrow",
+]
+
+[[package]]
+name = "dyn-stack"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "490bd48eb68fffcfed519b4edbfd82c69cbe741d175b84f0e0cbe8c57cbe0bdd"
+dependencies = [
+ "bytemuck",
+]
+
+[[package]]
+name = "either"
+version = "1.15.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "48c757948c5ede0e46177b7add2e67155f70e33c07fea8284df6576da70b3719"
+
+[[package]]
+name = "encode_unicode"
+version = "1.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "34aa73646ffb006b8f5147f3dc182bd4bcb190227ce861fc4a4844bf8e3cb2c0"
+
+[[package]]
+name = "encoding_rs"
+version = "0.8.35"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "75030f3c4f45dafd7586dd6780965a8c7e8e285a5ecb86713e63a79c5b2766f3"
+dependencies = [
+ "cfg-if",
+]
+
+[[package]]
+name = "enum-as-inner"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a1e6a265c649f3f5979b601d26f1d05ada116434c87741c9493cb56218f76cbc"
+dependencies = [
+ "heck",
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "equator"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4711b213838dfee0117e3be6ac926007d7f433d7bbe33595975d4190cb07e6fc"
+dependencies = [
+ "equator-macro",
+]
+
+[[package]]
+name = "equator-macro"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "44f23cf4b44bfce11a86ace86f8a73ffdec849c9fd00a386a53d278bd9e81fb3"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "equivalent"
+version = "1.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f"
+
+[[package]]
+name = "errno"
+version = "0.3.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "39cab71617ae0d63f51a36d69f866391735b51691dbda63cf6f96d042b63efeb"
+dependencies = [
+ "libc",
+ "windows-sys 0.61.0",
+]
+
+[[package]]
+name = "esaxx-rs"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d817e038c30374a4bcb22f94d0a8a0e216958d4c3dcde369b1439fec4bdda6e6"
+dependencies = [
+ "cc",
+]
+
+[[package]]
+name = "exr"
+version = "1.73.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f83197f59927b46c04a183a619b7c29df34e63e63c7869320862268c0ef687e0"
+dependencies = [
+ "bit_field",
+ "half",
+ "lebe",
+ "miniz_oxide",
+ "rayon-core",
+ "smallvec",
+ "zune-inflate",
+]
+
+[[package]]
+name = "fancy-regex"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "531e46835a22af56d1e3b66f04844bed63158bc094a628bec1d321d9b4c44bf2"
+dependencies = [
+ "bit-set",
+ "regex-automata",
+ "regex-syntax",
+]
+
+[[package]]
+name = "fastrand"
+version = "2.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be"
+
+[[package]]
+name = "fax"
+version = "0.2.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f05de7d48f37cd6730705cbca900770cab77a89f413d23e100ad7fad7795a0ab"
+dependencies = [
+ "fax_derive",
+]
+
+[[package]]
+name = "fax_derive"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a0aca10fb742cb43f9e7bb8467c91aa9bcb8e3ffbc6a6f7389bb93ffc920577d"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "fdeflate"
+version = "0.3.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e6853b52649d4ac5c0bd02320cddc5ba956bdb407c4b75a2c6b75bf51500f8c"
+dependencies = [
+ "simd-adler32",
+]
+
+[[package]]
+name = "ffmpeg-next"
+version = "8.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d658424d233cbd993a972dd73a66ca733acd12a494c68995c9ac32ae1fe65b40"
+dependencies = [
+ "bitflags 2.9.4",
+ "ffmpeg-sys-next",
+ "libc",
+]
+
+[[package]]
+name = "ffmpeg-sys-next"
+version = "8.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9bca20aa4ee774fe384c2490096c122b0b23cf524a9910add0686691003d797b"
+dependencies = [
+ "bindgen",
+ "cc",
+ "libc",
+ "num_cpus",
+ "pkg-config",
+ "vcpkg",
+]
+
+[[package]]
+name = "find-msvc-tools"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1ced73b1dacfc750a6db6c0a0c3a3853c8b41997e2e2c563dc90804ae6867959"
+
+[[package]]
+name = "flate2"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4a3d7db9596fecd151c5f638c0ee5d5bd487b6e0ea232e5dc96d5250f6f94b1d"
+dependencies = [
+ "crc32fast",
+ "miniz_oxide",
+]
+
+[[package]]
+name = "float8"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0f498aec3b227cd892ce18967f4033d9d397d28a80a7ab67e9f6b0176a79654e"
+dependencies = [
+ "half",
+]
+
+[[package]]
+name = "float8"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4203231de188ebbdfb85c11f3c20ca2b063945710de04e7b59268731e728b462"
+dependencies = [
+ "cudarc",
+ "half",
+ "num-traits",
+ "rand 0.9.2",
+ "rand_distr",
+]
+
+[[package]]
+name = "fnv"
+version = "1.0.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3f9eec918d3f24069decb9af1554cad7c880e2da24a9afd88aca000531ab82c1"
+
+[[package]]
+name = "foreign-types"
+version = "0.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f6f339eb8adc052cd2ca78910fda869aefa38d22d5cb648e6485e4d3fc06f3b1"
+dependencies = [
+ "foreign-types-shared",
+]
+
+[[package]]
+name = "foreign-types-shared"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "00b0228411908ca8685dba7fc2cdd70ec9990a6e753e89b6ac91a84c40fbaf4b"
+
+[[package]]
+name = "form_urlencoded"
+version = "1.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cb4cb245038516f5f85277875cdaa4f7d2c9a0fa0468de06ed190163b1581fcf"
+dependencies = [
+ "percent-encoding",
+]
+
+[[package]]
+name = "futures-channel"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2dff15bf788c671c1934e366d07e30c1814a8ef514e1af724a602e8a2fbe1b10"
+dependencies = [
+ "futures-core",
+ "futures-sink",
+]
+
+[[package]]
+name = "futures-core"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05f29059c0c2090612e8d742178b0580d2dc940c837851ad723096f87af6663e"
+
+[[package]]
+name = "futures-io"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9e5c1b78ca4aae1ac06c48a526a655760685149f0d465d21f37abfe57ce075c6"
+
+[[package]]
+name = "futures-macro"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "162ee34ebcb7c64a8abebc059ce0fee27c2262618d7b60ed8faf72fef13c3650"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "futures-sink"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e575fab7d1e0dcb8d0c7bcf9a63ee213816ab51902e6d244a95819acacf1d4f7"
+
+[[package]]
+name = "futures-task"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f90f7dce0722e95104fcb095585910c0977252f286e354b5e3bd38902cd99988"
+
+[[package]]
+name = "futures-util"
+version = "0.3.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9fa08315bb612088cc391249efdc3bc77536f16c91f6cf495e6fbe85b20a4a81"
+dependencies = [
+ "futures-core",
+ "futures-io",
+ "futures-macro",
+ "futures-sink",
+ "futures-task",
+ "memchr",
+ "pin-project-lite",
+ "pin-utils",
+ "slab",
+]
+
+[[package]]
+name = "gemm"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6ab24cc62135b40090e31a76a9b2766a501979f3070fa27f689c27ec04377d32"
+dependencies = [
+ "dyn-stack 0.10.0",
+ "gemm-c32 0.17.1",
+ "gemm-c64 0.17.1",
+ "gemm-common 0.17.1",
+ "gemm-f16 0.17.1",
+ "gemm-f32 0.17.1",
+ "gemm-f64 0.17.1",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 10.7.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ab96b703d31950f1aeddded248bc95543c9efc7ac9c4a21fda8703a83ee35451"
+dependencies = [
+ "dyn-stack 0.13.0",
+ "gemm-c32 0.18.2",
+ "gemm-c64 0.18.2",
+ "gemm-common 0.18.2",
+ "gemm-f16 0.18.2",
+ "gemm-f32 0.18.2",
+ "gemm-f64 0.18.2",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 11.6.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-c32"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9c030d0b983d1e34a546b86e08f600c11696fde16199f971cd46c12e67512c0"
+dependencies = [
+ "dyn-stack 0.10.0",
+ "gemm-common 0.17.1",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 10.7.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-c32"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f6db9fd9f40421d00eea9dd0770045a5603b8d684654816637732463f4073847"
+dependencies = [
+ "dyn-stack 0.13.0",
+ "gemm-common 0.18.2",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 11.6.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-c64"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fbb5f2e79fefb9693d18e1066a557b4546cd334b226beadc68b11a8f9431852a"
+dependencies = [
+ "dyn-stack 0.10.0",
+ "gemm-common 0.17.1",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 10.7.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-c64"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dfcad8a3d35a43758330b635d02edad980c1e143dc2f21e6fd25f9e4eada8edf"
+dependencies = [
+ "dyn-stack 0.13.0",
+ "gemm-common 0.18.2",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 11.6.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-common"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a2e7ea062c987abcd8db95db917b4ffb4ecdfd0668471d8dc54734fdff2354e8"
+dependencies = [
+ "bytemuck",
+ "dyn-stack 0.10.0",
+ "half",
+ "num-complex",
+ "num-traits",
+ "once_cell",
+ "paste",
+ "pulp 0.18.22",
+ "raw-cpuid 10.7.0",
+ "rayon",
+ "seq-macro",
+ "sysctl 0.5.5",
+]
+
+[[package]]
+name = "gemm-common"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a352d4a69cbe938b9e2a9cb7a3a63b7e72f9349174a2752a558a8a563510d0f3"
+dependencies = [
+ "bytemuck",
+ "dyn-stack 0.13.0",
+ "half",
+ "libm",
+ "num-complex",
+ "num-traits",
+ "once_cell",
+ "paste",
+ "pulp 0.21.5",
+ "raw-cpuid 11.6.0",
+ "rayon",
+ "seq-macro",
+ "sysctl 0.6.0",
+]
+
+[[package]]
+name = "gemm-f16"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7ca4c06b9b11952071d317604acb332e924e817bd891bec8dfb494168c7cedd4"
+dependencies = [
+ "dyn-stack 0.10.0",
+ "gemm-common 0.17.1",
+ "gemm-f32 0.17.1",
+ "half",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 10.7.0",
+ "rayon",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-f16"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cff95ae3259432f3c3410eaa919033cd03791d81cebd18018393dc147952e109"
+dependencies = [
+ "dyn-stack 0.13.0",
+ "gemm-common 0.18.2",
+ "gemm-f32 0.18.2",
+ "half",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 11.6.0",
+ "rayon",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-f32"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e9a69f51aaefbd9cf12d18faf273d3e982d9d711f60775645ed5c8047b4ae113"
+dependencies = [
+ "dyn-stack 0.10.0",
+ "gemm-common 0.17.1",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 10.7.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-f32"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bc8d3d4385393304f407392f754cd2dc4b315d05063f62cf09f47b58de276864"
+dependencies = [
+ "dyn-stack 0.13.0",
+ "gemm-common 0.18.2",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 11.6.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-f64"
+version = "0.17.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "aa397a48544fadf0b81ec8741e5c0fba0043008113f71f2034def1935645d2b0"
+dependencies = [
+ "dyn-stack 0.10.0",
+ "gemm-common 0.17.1",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 10.7.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "gemm-f64"
+version = "0.18.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "35b2a4f76ce4b8b16eadc11ccf2e083252d8237c1b589558a49b0183545015bd"
+dependencies = [
+ "dyn-stack 0.13.0",
+ "gemm-common 0.18.2",
+ "num-complex",
+ "num-traits",
+ "paste",
+ "raw-cpuid 11.6.0",
+ "seq-macro",
+]
+
+[[package]]
+name = "getrandom"
+version = "0.2.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "335ff9f135e4384c8150d6f27c6daed433577f86b4750418338c01a1a2528592"
+dependencies = [
+ "cfg-if",
+ "libc",
+ "wasi 0.11.1+wasi-snapshot-preview1",
+]
+
+[[package]]
+name = "getrandom"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "26145e563e54f2cadc477553f1ec5ee650b00862f0a58bcd12cbdc5f0ea2d2f4"
+dependencies = [
+ "cfg-if",
+ "libc",
+ "r-efi",
+ "wasi 0.14.7+wasi-0.2.4",
+]
+
+[[package]]
+name = "gif"
+version = "0.13.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4ae047235e33e2829703574b54fdec96bfbad892062d97fed2f76022287de61b"
+dependencies = [
+ "color_quant",
+ "weezl",
+]
+
+[[package]]
+name = "gimli"
+version = "0.31.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "07e28edb80900c19c28f1072f2e8aeca7fa06b23cd4169cefe1af5aa3260783f"
+
+[[package]]
+name = "glob"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0cc23270f6e1808e30a928bdc84dea0b9b4136a8bc82338574f23baf47bbd280"
+
+[[package]]
+name = "h2"
+version = "0.4.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f3c0b69cfcb4e1b9f1bf2f53f95f766e4661169728ec61cd3fe5a0166f2d1386"
+dependencies = [
+ "atomic-waker",
+ "bytes",
+ "fnv",
+ "futures-core",
+ "futures-sink",
+ "http",
+ "indexmap",
+ "slab",
+ "tokio",
+ "tokio-util",
+ "tracing",
+]
+
+[[package]]
+name = "half"
+version = "2.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "459196ed295495a68f7d7fe1d84f6c4b7ff0e21fe3017b2f283c6fac3ad803c9"
+dependencies = [
+ "bytemuck",
+ "cfg-if",
+ "crunchy",
+ "num-traits",
+ "rand 0.9.2",
+ "rand_distr",
+]
+
+[[package]]
+name = "hashbrown"
+version = "0.16.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5419bdc4f6a9207fbeba6d11b604d481addf78ecd10c11ad51e76c2f6482748d"
+
+[[package]]
+name = "heck"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
+
+[[package]]
+name = "hermit-abi"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fc0fef456e4baa96da950455cd02c081ca953b141298e41db3fc7e36b1da849c"
+
+[[package]]
+name = "http"
+version = "1.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f4a85d31aea989eead29a3aaf9e1115a180df8282431156e533de47660892565"
+dependencies = [
+ "bytes",
+ "fnv",
+ "itoa",
+]
+
+[[package]]
+name = "http-body"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1efedce1fb8e6913f23e0c92de8e62cd5b772a67e7b3946df930a62566c93184"
+dependencies = [
+ "bytes",
+ "http",
+]
+
+[[package]]
+name = "http-body-util"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b021d93e26becf5dc7e1b75b1bed1fd93124b374ceb73f43d4d4eafec896a64a"
+dependencies = [
+ "bytes",
+ "futures-core",
+ "http",
+ "http-body",
+ "pin-project-lite",
+]
+
+[[package]]
+name = "httparse"
+version = "1.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6dbf3de79e51f3d586ab4cb9d5c3e2c14aa28ed23d180cf89b4df0454a69cc87"
+
+[[package]]
+name = "hyper"
+version = "1.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eb3aa54a13a0dfe7fbe3a59e0c76093041720fdc77b110cc0fc260fafb4dc51e"
+dependencies = [
+ "atomic-waker",
+ "bytes",
+ "futures-channel",
+ "futures-core",
+ "h2",
+ "http",
+ "http-body",
+ "httparse",
+ "itoa",
+ "pin-project-lite",
+ "pin-utils",
+ "smallvec",
+ "tokio",
+ "want",
+]
+
+[[package]]
+name = "hyper-rustls"
+version = "0.27.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e3c93eb611681b207e1fe55d5a71ecf91572ec8a6705cdb6857f7d8d5242cf58"
+dependencies = [
+ "http",
+ "hyper",
+ "hyper-util",
+ "rustls",
+ "rustls-pki-types",
+ "tokio",
+ "tokio-rustls",
+ "tower-service",
+]
+
+[[package]]
+name = "hyper-tls"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "70206fc6890eaca9fde8a0bf71caa2ddfc9fe045ac9e5c70df101a7dbde866e0"
+dependencies = [
+ "bytes",
+ "http-body-util",
+ "hyper",
+ "hyper-util",
+ "native-tls",
+ "tokio",
+ "tokio-native-tls",
+ "tower-service",
+]
+
+[[package]]
+name = "hyper-util"
+version = "0.1.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3c6995591a8f1380fcb4ba966a252a4b29188d51d2b89e3a252f5305be65aea8"
+dependencies = [
+ "base64 0.22.1",
+ "bytes",
+ "futures-channel",
+ "futures-core",
+ "futures-util",
+ "http",
+ "http-body",
+ "hyper",
+ "ipnet",
+ "libc",
+ "percent-encoding",
+ "pin-project-lite",
+ "socket2",
+ "system-configuration",
+ "tokio",
+ "tower-service",
+ "tracing",
+ "windows-registry",
+]
+
+[[package]]
+name = "icu_collections"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "200072f5d0e3614556f94a9930d5dc3e0662a652823904c3a75dc3b0af7fee47"
+dependencies = [
+ "displaydoc",
+ "potential_utf",
+ "yoke 0.8.0",
+ "zerofrom",
+ "zerovec",
+]
+
+[[package]]
+name = "icu_locale_core"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0cde2700ccaed3872079a65fb1a78f6c0a36c91570f28755dda67bc8f7d9f00a"
+dependencies = [
+ "displaydoc",
+ "litemap",
+ "tinystr",
+ "writeable",
+ "zerovec",
+]
+
+[[package]]
+name = "icu_normalizer"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "436880e8e18df4d7bbc06d58432329d6458cc84531f7ac5f024e93deadb37979"
+dependencies = [
+ "displaydoc",
+ "icu_collections",
+ "icu_normalizer_data",
+ "icu_properties",
+ "icu_provider",
+ "smallvec",
+ "zerovec",
+]
+
+[[package]]
+name = "icu_normalizer_data"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "00210d6893afc98edb752b664b8890f0ef174c8adbb8d0be9710fa66fbbf72d3"
+
+[[package]]
+name = "icu_properties"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "016c619c1eeb94efb86809b015c58f479963de65bdb6253345c1a1276f22e32b"
+dependencies = [
+ "displaydoc",
+ "icu_collections",
+ "icu_locale_core",
+ "icu_properties_data",
+ "icu_provider",
+ "potential_utf",
+ "zerotrie",
+ "zerovec",
+]
+
+[[package]]
+name = "icu_properties_data"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "298459143998310acd25ffe6810ed544932242d3f07083eee1084d83a71bd632"
+
+[[package]]
+name = "icu_provider"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "03c80da27b5f4187909049ee2d72f276f0d9f99a42c306bd0131ecfe04d8e5af"
+dependencies = [
+ "displaydoc",
+ "icu_locale_core",
+ "stable_deref_trait",
+ "tinystr",
+ "writeable",
+ "yoke 0.8.0",
+ "zerofrom",
+ "zerotrie",
+ "zerovec",
+]
+
+[[package]]
+name = "ident_case"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9e0384b61958566e926dc50660321d12159025e767c18e043daf26b70104c39"
+
+[[package]]
+name = "idna"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b0875f23caa03898994f6ddc501886a45c7d3d62d04d2d90788d47be1b1e4de"
+dependencies = [
+ "idna_adapter",
+ "smallvec",
+ "utf8_iter",
+]
+
+[[package]]
+name = "idna_adapter"
+version = "1.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3acae9609540aa318d1bc588455225fb2085b9ed0c4f6bd0d9d5bcd86f1a0344"
+dependencies = [
+ "icu_normalizer",
+ "icu_properties",
+]
+
+[[package]]
+name = "image"
+version = "0.25.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "529feb3e6769d234375c4cf1ee2ce713682b8e76538cb13f9fc23e1400a591e7"
+dependencies = [
+ "bytemuck",
+ "byteorder-lite",
+ "color_quant",
+ "exr",
+ "gif",
+ "image-webp",
+ "moxcms",
+ "num-traits",
+ "png",
+ "qoi",
+ "ravif",
+ "rayon",
+ "rgb",
+ "tiff",
+ "zune-core",
+ "zune-jpeg",
+]
+
+[[package]]
+name = "image-webp"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "525e9ff3e1a4be2fbea1fdf0e98686a6d98b4d8f937e1bf7402245af1909e8c3"
+dependencies = [
+ "byteorder-lite",
+ "quick-error",
+]
+
+[[package]]
+name = "imgref"
+version = "1.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d0263a3d970d5c054ed9312c0057b4f3bde9c0b33836d3637361d4a9e6e7a408"
+
+[[package]]
+name = "indexmap"
+version = "2.11.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4b0f83760fb341a774ed326568e19f5a863af4a952def8c39f9ab92fd95b88e5"
+dependencies = [
+ "equivalent",
+ "hashbrown",
+]
+
+[[package]]
+name = "indicatif"
+version = "0.17.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "183b3088984b400f4cfac3620d5e076c84da5364016b4f49473de574b2586235"
+dependencies = [
+ "console",
+ "number_prefix",
+ "portable-atomic",
+ "unicode-width",
+ "web-time",
+]
+
+[[package]]
+name = "interpolate_name"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c34819042dc3d3971c46c2190835914dfbe0c3c13f61449b2997f4e9722dfa60"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "io-uring"
+version = "0.7.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "046fa2d4d00aea763528b4950358d0ead425372445dc8ff86312b3c69ff7727b"
+dependencies = [
+ "bitflags 2.9.4",
+ "cfg-if",
+ "libc",
+]
+
+[[package]]
+name = "ipnet"
+version = "2.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "469fb0b9cefa57e3ef31275ee7cacb78f2fdca44e4765491884a2b119d4eb130"
+
+[[package]]
+name = "iri-string"
+version = "0.7.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dbc5ebe9c3a1a7a5127f920a418f7585e9e758e911d0466ed004f393b0e380b2"
+dependencies = [
+ "memchr",
+ "serde",
+]
+
+[[package]]
+name = "itertools"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba291022dbbd398a455acf126c1e341954079855bc60dfdda641363bd6922569"
+dependencies = [
+ "either",
+]
+
+[[package]]
+name = "itertools"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "413ee7dfc52ee1a4949ceeb7dbc8a33f2d6c088194d9f922fb8318faf1f01186"
+dependencies = [
+ "either",
+]
+
+[[package]]
+name = "itertools"
+version = "0.14.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2b192c782037fadd9cfa75548310488aabdbf3d2da73885b31bd0abd03351285"
+dependencies = [
+ "either",
+]
+
+[[package]]
+name = "itoa"
+version = "1.0.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4a5f13b858c8d314ee3e8f639011f7ccefe71f97f96e50151fb991f267928e2c"
+
+[[package]]
+name = "jobserver"
+version = "0.1.34"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9afb3de4395d6b3e67a780b6de64b51c978ecf11cb9a462c66be7d4ca9039d33"
+dependencies = [
+ "getrandom 0.3.3",
+ "libc",
+]
+
+[[package]]
+name = "js-sys"
+version = "0.3.80"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "852f13bec5eba4ba9afbeb93fd7c13fe56147f055939ae21c43a29a0ecb2702e"
+dependencies = [
+ "once_cell",
+ "wasm-bindgen",
+]
+
+[[package]]
+name = "lebe"
+version = "0.5.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7a79a3332a6609480d7d0c9eab957bca6b455b91bb84e66d19f5ff66294b85b8"
+
+[[package]]
+name = "libc"
+version = "0.2.175"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6a82ae493e598baaea5209805c49bbf2ea7de956d50d7da0da1164f9c6d28543"
+
+[[package]]
+name = "libfuzzer-sys"
+version = "0.4.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5037190e1f70cbeef565bd267599242926f724d3b8a9f510fd7e0b540cfa4404"
+dependencies = [
+ "arbitrary",
+ "cc",
+]
+
+[[package]]
+name = "libloading"
+version = "0.8.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "07033963ba89ebaf1584d767badaa2e8fcec21aedea6b8c0346d487d49c28667"
+dependencies = [
+ "cfg-if",
+ "windows-targets 0.53.3",
+]
+
+[[package]]
+name = "libm"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f9fbbcab51052fe104eb5e5d351cf728d30a5be1fe14d9be8a3b097481fb97de"
+
+[[package]]
+name = "linux-raw-sys"
+version = "0.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "df1d3c3b53da64cf5760482273a98e575c651a67eec7f77df96b5b642de8f039"
+
+[[package]]
+name = "litemap"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "241eaef5fd12c88705a01fc1066c48c4b36e0dd4377dcdc7ec3942cea7a69956"
+
+[[package]]
+name = "log"
+version = "0.4.28"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "34080505efa8e45a4b816c349525ebe327ceaa8559756f0356cba97ef3bf7432"
+
+[[package]]
+name = "loop9"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0fae87c125b03c1d2c0150c90365d7d6bcc53fb73a9acaef207d2d065860f062"
+dependencies = [
+ "imgref",
+]
+
+[[package]]
+name = "macro_rules_attribute"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "65049d7923698040cd0b1ddcced9b0eb14dd22c5f86ae59c3740eab64a676520"
+dependencies = [
+ "macro_rules_attribute-proc_macro",
+ "paste",
+]
+
+[[package]]
+name = "macro_rules_attribute-proc_macro"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "670fdfda89751bc4a84ac13eaa63e205cf0fd22b4c9a5fbfa085b63c1f1d3a30"
+
+[[package]]
+name = "maybe-rayon"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8ea1f30cedd69f0a2954655f7188c6a834246d2bcf1e315e2ac40c4b24dc9519"
+dependencies = [
+ "cfg-if",
+ "rayon",
+]
+
+[[package]]
+name = "memchr"
+version = "2.7.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32a282da65faaf38286cf3be983213fcf1d2e2a58700e808f83f4ea9a4804bc0"
+
+[[package]]
+name = "memmap2"
+version = "0.9.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "843a98750cd611cc2965a8213b53b43e715f13c37a9e096c6408e69990961db7"
+dependencies = [
+ "libc",
+ "stable_deref_trait",
+]
+
+[[package]]
+name = "mime"
+version = "0.3.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6877bb514081ee2a7ff5ef9de3281f14a4dd4bceac4c09388074a6b5df8a139a"
+
+[[package]]
+name = "mime_guess"
+version = "2.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f7c44f8e672c00fe5308fa235f821cb4198414e1c77935c1ab6948d3fd78550e"
+dependencies = [
+ "mime",
+ "unicase",
+]
+
+[[package]]
+name = "minijinja"
+version = "2.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a9f264d75233323f4b7d2f03aefe8a990690cdebfbfe26ea86bcbaec5e9ac990"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "minimal-lexical"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "68354c5c6bd36d73ff3feceb05efa59b6acb7626617f4962be322a825e61f79a"
+
+[[package]]
+name = "miniz_oxide"
+version = "0.8.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1fa76a2c86f704bdb222d66965fb3d63269ce38518b83cb0575fca855ebb6316"
+dependencies = [
+ "adler2",
+ "simd-adler32",
+]
+
+[[package]]
+name = "mio"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "78bed444cc8a2160f01cbcf811ef18cac863ad68ae8ca62092e8db51d51c761c"
+dependencies = [
+ "libc",
+ "wasi 0.11.1+wasi-snapshot-preview1",
+ "windows-sys 0.59.0",
+]
+
+[[package]]
+name = "monostate"
+version = "0.1.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5f766eeb5719df144c29802e56f3f7e0b3f29bd3ec8ab6c819aa1eaddec3f80c"
+dependencies = [
+ "monostate-impl",
+ "serde",
+ "serde_core",
+]
+
+[[package]]
+name = "monostate-impl"
+version = "0.1.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "88d4b3dcd6ce9277a52fb05f00bf3e45d475af8cce7376de2f8d6bd065fa4adb"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "moxcms"
+version = "0.7.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ddd32fa8935aeadb8a8a6b6b351e40225570a37c43de67690383d87ef170cd08"
+dependencies = [
+ "num-traits",
+ "pxfm",
+]
+
+[[package]]
+name = "native-tls"
+version = "0.2.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "87de3442987e9dbec73158d5c715e7ad9072fda936bb03d19d7fa10e00520f0e"
+dependencies = [
+ "libc",
+ "log",
+ "openssl",
+ "openssl-probe",
+ "openssl-sys",
+ "schannel",
+ "security-framework",
+ "security-framework-sys",
+ "tempfile",
+]
+
+[[package]]
+name = "new_debug_unreachable"
+version = "1.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "650eef8c711430f1a879fdd01d4745a7deea475becfb90269c06775983bbf086"
+
+[[package]]
+name = "nom"
+version = "7.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d273983c5a657a70a3e8f2a01329822f3b8c8172b73826411a55751e404a0a4a"
+dependencies = [
+ "memchr",
+ "minimal-lexical",
+]
+
+[[package]]
+name = "noop_proc_macro"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0676bb32a98c1a483ce53e500a81ad9c3d5b3f7c920c28c24e9cb0980d0b5bc8"
+
+[[package]]
+name = "num"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "35bd024e8b2ff75562e5f34e7f4905839deb4b22955ef5e73d2fea1b9813cb23"
+dependencies = [
+ "num-bigint",
+ "num-complex",
+ "num-integer",
+ "num-iter",
+ "num-rational",
+ "num-traits",
+]
+
+[[package]]
+name = "num-bigint"
+version = "0.4.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a5e44f723f1133c9deac646763579fdb3ac745e418f2a7af9cd0c431da1f20b9"
+dependencies = [
+ "num-integer",
+ "num-traits",
+]
+
+[[package]]
+name = "num-complex"
+version = "0.4.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "73f88a1307638156682bada9d7604135552957b7818057dcef22705b4d509495"
+dependencies = [
+ "bytemuck",
+ "num-traits",
+]
+
+[[package]]
+name = "num-derive"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ed3955f1a9c7c0c15e092f9c887db08b1fc683305fdf6eb6684f22555355e202"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "num-integer"
+version = "0.1.46"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7969661fd2958a5cb096e56c8e1ad0444ac2bbcd0061bd28660485a44879858f"
+dependencies = [
+ "num-traits",
+]
+
+[[package]]
+name = "num-iter"
+version = "0.1.45"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1429034a0490724d0075ebb2bc9e875d6503c3cf69e235a8941aa757d83ef5bf"
+dependencies = [
+ "autocfg",
+ "num-integer",
+ "num-traits",
+]
+
+[[package]]
+name = "num-rational"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f83d14da390562dca69fc84082e73e548e1ad308d24accdedd2720017cb37824"
+dependencies = [
+ "num-bigint",
+ "num-integer",
+ "num-traits",
+]
+
+[[package]]
+name = "num-traits"
+version = "0.2.19"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "071dfc062690e90b734c0b2273ce72ad0ffa95f0c74596bc250dcfd960262841"
+dependencies = [
+ "autocfg",
+ "libm",
+]
+
+[[package]]
+name = "num_cpus"
+version = "1.17.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91df4bbde75afed763b708b7eee1e8e7651e02d97f6d5dd763e89367e957b23b"
+dependencies = [
+ "hermit-abi",
+ "libc",
+]
+
+[[package]]
+name = "num_enum"
+version = "0.7.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a973b4e44ce6cad84ce69d797acf9a044532e4184c4f267913d1b546a0727b7a"
+dependencies = [
+ "num_enum_derive",
+ "rustversion",
+]
+
+[[package]]
+name = "num_enum_derive"
+version = "0.7.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "77e878c846a8abae00dd069496dbe8751b16ac1c3d6bd2a7283a938e8228f90d"
+dependencies = [
+ "proc-macro-crate",
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "number_prefix"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "830b246a0e5f20af87141b25c173cd1b609bd7779a4617d6ec582abaf90870f3"
+
+[[package]]
+name = "object"
+version = "0.36.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "62948e14d923ea95ea2c7c86c71013138b66525b86bdc08d2dcc262bdb497b87"
+dependencies = [
+ "memchr",
+]
+
+[[package]]
+name = "once_cell"
+version = "1.21.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "42f5e15c9953c5e4ccceeb2e7382a716482c34515315f7b03532b8b4e8393d2d"
+
+[[package]]
+name = "onig"
+version = "6.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "336b9c63443aceef14bea841b899035ae3abe89b7c486aaf4c5bd8aafedac3f0"
+dependencies = [
+ "bitflags 2.9.4",
+ "libc",
+ "once_cell",
+ "onig_sys",
+]
+
+[[package]]
+name = "onig_sys"
+version = "69.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c7f86c6eef3d6df15f23bcfb6af487cbd2fed4e5581d58d5bf1f5f8b7f6727dc"
+dependencies = [
+ "cc",
+ "pkg-config",
+]
+
+[[package]]
+name = "openai_dive"
+version = "1.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bac82ff995cdbe4e120fa87fe4d76bd452866dd7b0c9bcbdb562ba4be0f1786d"
+dependencies = [
+ "bytes",
+ "derive_builder",
+ "reqwest",
+ "serde",
+ "serde_html_form",
+ "serde_json",
+ "tokio",
+ "tokio-util",
+]
+
+[[package]]
+name = "openssl"
+version = "0.10.73"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8505734d46c8ab1e19a1dce3aef597ad87dcb4c37e7188231769bd6bd51cebf8"
+dependencies = [
+ "bitflags 2.9.4",
+ "cfg-if",
+ "foreign-types",
+ "libc",
+ "once_cell",
+ "openssl-macros",
+ "openssl-sys",
+]
+
+[[package]]
+name = "openssl-macros"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a948666b637a0f465e8564c73e89d4dde00d72d4d473cc972f390fc3dcee7d9c"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "openssl-probe"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d05e27ee213611ffe7d6348b942e8f942b37114c00cc03cec254295a4a17852e"
+
+[[package]]
+name = "openssl-sys"
+version = "0.9.109"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "90096e2e47630d78b7d1c20952dc621f957103f8bc2c8359ec81290d75238571"
+dependencies = [
+ "cc",
+ "libc",
+ "pkg-config",
+ "vcpkg",
+]
+
+[[package]]
+name = "paste"
+version = "1.0.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "57c0d7b74b563b49d38dae00a0c37d4d6de9b432382b2892f0574ddcae73fd0a"
+
+[[package]]
+name = "percent-encoding"
+version = "2.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9b4f627cb1b25917193a259e49bdad08f671f8d9708acfd5fe0a8c1455d87220"
+
+[[package]]
+name = "pin-project-lite"
+version = "0.2.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b3cff922bd51709b605d9ead9aa71031d81447142d828eb4a6eba76fe619f9b"
+
+[[package]]
+name = "pin-utils"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184"
+
+[[package]]
+name = "pkg-config"
+version = "0.3.32"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7edddbd0b52d732b21ad9a5fab5c704c14cd949e5e9a1ec5929a24fded1b904c"
+
+[[package]]
+name = "png"
+version = "0.18.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "97baced388464909d42d89643fe4361939af9b7ce7a31ee32a168f832a70f2a0"
+dependencies = [
+ "bitflags 2.9.4",
+ "crc32fast",
+ "fdeflate",
+ "flate2",
+ "miniz_oxide",
+]
+
+[[package]]
+name = "portable-atomic"
+version = "1.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f84267b20a16ea918e43c6a88433c2d54fa145c92a811b5b047ccbe153674483"
+
+[[package]]
+name = "potential_utf"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "84df19adbe5b5a0782edcab45899906947ab039ccf4573713735ee7de1e6b08a"
+dependencies = [
+ "zerovec",
+]
+
+[[package]]
+name = "ppv-lite86"
+version = "0.2.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "85eae3c4ed2f50dcfe72643da4befc30deadb458a9b590d720cde2f2b1e97da9"
+dependencies = [
+ "zerocopy",
+]
+
+[[package]]
+name = "proc-macro-crate"
+version = "3.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "219cb19e96be00ab2e37d6e299658a0cfa83e52429179969b0f0121b4ac46983"
+dependencies = [
+ "toml_edit 0.23.6",
+]
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.101"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "89ae43fd86e4158d6db51ad8e2b80f313af9cc74f5c0e03ccb87de09998732de"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "profiling"
+version = "1.0.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3eb8486b569e12e2c32ad3e204dbaba5e4b5b216e9367044f25f1dba42341773"
+dependencies = [
+ "profiling-procmacros",
+]
+
+[[package]]
+name = "profiling-procmacros"
+version = "1.0.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "52717f9a02b6965224f95ca2a81e2e0c5c43baacd28ca057577988930b6c3d5b"
+dependencies = [
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "pulp"
+version = "0.18.22"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a0a01a0dc67cf4558d279f0c25b0962bd08fc6dec0137699eae304103e882fe6"
+dependencies = [
+ "bytemuck",
+ "libm",
+ "num-complex",
+ "reborrow",
+]
+
+[[package]]
+name = "pulp"
+version = "0.21.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "96b86df24f0a7ddd5e4b95c94fc9ed8a98f1ca94d3b01bdce2824097e7835907"
+dependencies = [
+ "bytemuck",
+ "cfg-if",
+ "libm",
+ "num-complex",
+ "reborrow",
+ "version_check",
+]
+
+[[package]]
+name = "pxfm"
+version = "0.1.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "83f9b339b02259ada5c0f4a389b7fb472f933aa17ce176fd2ad98f28bb401fde"
+dependencies = [
+ "num-traits",
+]
+
+[[package]]
+name = "qoi"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7f6d64c71eb498fe9eae14ce4ec935c555749aef511cca85b5568910d6e48001"
+dependencies = [
+ "bytemuck",
+]
+
+[[package]]
+name = "quick-error"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a993555f31e5a609f617c12db6250dedcac1b0a85076912c436e6fc9b2c8e6a3"
+
+[[package]]
+name = "quote"
+version = "1.0.40"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1885c039570dc00dcb4ff087a89e185fd56bae234ddc7f056a945bf36467248d"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "r-efi"
+version = "5.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "69cdb34c158ceb288df11e18b4bd39de994f6657d83847bdffdbd7f346754b0f"
+
+[[package]]
+name = "rand"
+version = "0.8.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "34af8d1a0e25924bc5b7c43c079c942339d8f0a8b57c39049bef581b46327404"
+dependencies = [
+ "libc",
+ "rand_chacha 0.3.1",
+ "rand_core 0.6.4",
+]
+
+[[package]]
+name = "rand"
+version = "0.9.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6db2770f06117d490610c7488547d543617b21bfa07796d7a12f6f1bd53850d1"
+dependencies = [
+ "rand_chacha 0.9.0",
+ "rand_core 0.9.3",
+]
+
+[[package]]
+name = "rand_chacha"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e6c10a63a0fa32252be49d21e7709d4d4baf8d231c2dbce1eaa8141b9b127d88"
+dependencies = [
+ "ppv-lite86",
+ "rand_core 0.6.4",
+]
+
+[[package]]
+name = "rand_chacha"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d3022b5f1df60f26e1ffddd6c66e8aa15de382ae63b3a0c1bfc0e4d3e3f325cb"
+dependencies = [
+ "ppv-lite86",
+ "rand_core 0.9.3",
+]
+
+[[package]]
+name = "rand_core"
+version = "0.6.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ec0be4795e2f6a28069bec0b5ff3e2ac9bafc99e6a9a7dc3547996c5c816922c"
+dependencies = [
+ "getrandom 0.2.16",
+]
+
+[[package]]
+name = "rand_core"
+version = "0.9.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "99d9a13982dcf210057a8a78572b2217b667c3beacbf3a0d8b454f6f82837d38"
+dependencies = [
+ "getrandom 0.3.3",
+]
+
+[[package]]
+name = "rand_distr"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6a8615d50dcf34fa31f7ab52692afec947c4dd0ab803cc87cb3b0b4570ff7463"
+dependencies = [
+ "num-traits",
+ "rand 0.9.2",
+]
+
+[[package]]
+name = "rav1e"
+version = "0.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cd87ce80a7665b1cce111f8a16c1f3929f6547ce91ade6addf4ec86a8dda5ce9"
+dependencies = [
+ "arbitrary",
+ "arg_enum_proc_macro",
+ "arrayvec",
+ "av1-grain",
+ "bitstream-io",
+ "built",
+ "cfg-if",
+ "interpolate_name",
+ "itertools 0.12.1",
+ "libc",
+ "libfuzzer-sys",
+ "log",
+ "maybe-rayon",
+ "new_debug_unreachable",
+ "noop_proc_macro",
+ "num-derive",
+ "num-traits",
+ "once_cell",
+ "paste",
+ "profiling",
+ "rand 0.8.5",
+ "rand_chacha 0.3.1",
+ "simd_helpers",
+ "system-deps",
+ "thiserror 1.0.69",
+ "v_frame",
+ "wasm-bindgen",
+]
+
+[[package]]
+name = "ravif"
+version = "0.11.20"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5825c26fddd16ab9f515930d49028a630efec172e903483c94796cfe31893e6b"
+dependencies = [
+ "avif-serialize",
+ "imgref",
+ "loop9",
+ "quick-error",
+ "rav1e",
+ "rayon",
+ "rgb",
+]
+
+[[package]]
+name = "raw-cpuid"
+version = "10.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c297679cb867470fa8c9f67dbba74a78d78e3e98d7cf2b08d6d71540f797332"
+dependencies = [
+ "bitflags 1.3.2",
+]
+
+[[package]]
+name = "raw-cpuid"
+version = "11.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "498cd0dc59d73224351ee52a95fee0f1a617a2eae0e7d9d720cc622c73a54186"
+dependencies = [
+ "bitflags 2.9.4",
+]
+
+[[package]]
+name = "rayon"
+version = "1.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "368f01d005bf8fd9b1206fb6fa653e6c4a81ceb1466406b81792d87c5677a58f"
+dependencies = [
+ "either",
+ "rayon-core",
+]
+
+[[package]]
+name = "rayon-cond"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2964d0cf57a3e7a06e8183d14a8b527195c706b7983549cd5462d5aa3747438f"
+dependencies = [
+ "either",
+ "itertools 0.14.0",
+ "rayon",
+]
+
+[[package]]
+name = "rayon-core"
+version = "1.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "22e18b0f0062d30d4230b2e85ff77fdfe4326feb054b9783a3460d8435c8ab91"
+dependencies = [
+ "crossbeam-deque",
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "reborrow"
+version = "0.5.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "03251193000f4bd3b042892be858ee50e8b3719f2b08e5833ac4353724632430"
+
+[[package]]
+name = "regex"
+version = "1.11.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "23d7fd106d8c02486a8d64e778353d1cffe08ce79ac2e82f540c86d0facf6912"
+dependencies = [
+ "aho-corasick",
+ "memchr",
+ "regex-automata",
+ "regex-syntax",
+]
+
+[[package]]
+name = "regex-automata"
+version = "0.4.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6b9458fa0bfeeac22b5ca447c63aaf45f28439a709ccd244698632f9aa6394d6"
+dependencies = [
+ "aho-corasick",
+ "memchr",
+ "regex-syntax",
+]
+
+[[package]]
+name = "regex-syntax"
+version = "0.8.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "caf4aa5b0f434c91fe5c7f1ecb6a5ece2130b02ad2a590589dda5146df959001"
+
+[[package]]
+name = "reqwest"
+version = "0.12.23"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d429f34c8092b2d42c7c93cec323bb4adeb7c67698f70839adec842ec10c7ceb"
+dependencies = [
+ "base64 0.22.1",
+ "bytes",
+ "encoding_rs",
+ "futures-channel",
+ "futures-core",
+ "futures-util",
+ "h2",
+ "http",
+ "http-body",
+ "http-body-util",
+ "hyper",
+ "hyper-rustls",
+ "hyper-tls",
+ "hyper-util",
+ "js-sys",
+ "log",
+ "mime",
+ "mime_guess",
+ "native-tls",
+ "percent-encoding",
+ "pin-project-lite",
+ "rustls-pki-types",
+ "serde",
+ "serde_json",
+ "serde_urlencoded",
+ "sync_wrapper",
+ "tokio",
+ "tokio-native-tls",
+ "tokio-util",
+ "tower",
+ "tower-http",
+ "tower-service",
+ "url",
+ "wasm-bindgen",
+ "wasm-bindgen-futures",
+ "wasm-streams",
+ "web-sys",
+]
+
+[[package]]
+name = "rgb"
+version = "0.8.52"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0c6a884d2998352bb4daf0183589aec883f16a6da1f4dde84d8e2e9a5409a1ce"
+
+[[package]]
+name = "ring"
+version = "0.17.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a4689e6c2294d81e88dc6261c768b63bc4fcdb852be6d1352498b114f61383b7"
+dependencies = [
+ "cc",
+ "cfg-if",
+ "getrandom 0.2.16",
+ "libc",
+ "untrusted",
+ "windows-sys 0.52.0",
+]
+
+[[package]]
+name = "rustc-demangle"
+version = "0.1.26"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "56f7d92ca342cea22a06f2121d944b4fd82af56988c270852495420f961d4ace"
+
+[[package]]
+name = "rustc-hash"
+version = "2.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "357703d41365b4b27c590e3ed91eabb1b663f07c4c084095e60cbed4362dff0d"
+
+[[package]]
+name = "rustix"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cd15f8a2c5551a84d56efdc1cd049089e409ac19a3072d5037a17fd70719ff3e"
+dependencies = [
+ "bitflags 2.9.4",
+ "errno",
+ "libc",
+ "linux-raw-sys",
+ "windows-sys 0.61.0",
+]
+
+[[package]]
+name = "rustls"
+version = "0.23.32"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cd3c25631629d034ce7cd9940adc9d45762d46de2b0f57193c4443b92c6d4d40"
+dependencies = [
+ "once_cell",
+ "rustls-pki-types",
+ "rustls-webpki",
+ "subtle",
+ "zeroize",
+]
+
+[[package]]
+name = "rustls-pki-types"
+version = "1.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "229a4a4c221013e7e1f1a043678c5cc39fe5171437c88fb47151a21e6f5b5c79"
+dependencies = [
+ "zeroize",
+]
+
+[[package]]
+name = "rustls-webpki"
+version = "0.103.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8572f3c2cb9934231157b45499fc41e1f58c589fdfb81a844ba873265e80f8eb"
+dependencies = [
+ "ring",
+ "rustls-pki-types",
+ "untrusted",
+]
+
+[[package]]
+name = "rustversion"
+version = "1.0.22"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b39cdef0fa800fc44525c84ccb54a029961a8215f9619753635a9c0d2538d46d"
+
+[[package]]
+name = "ryu"
+version = "1.0.20"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "28d3b2b1366ec20994f1fd18c3c594f05c5dd4bc44d8bb0c1c632c8d6829481f"
+
+[[package]]
+name = "safetensors"
+version = "0.4.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "44560c11236a6130a46ce36c836a62936dc81ebf8c36a37947423571be0e55b6"
+dependencies = [
+ "serde",
+ "serde_json",
+]
+
+[[package]]
+name = "same-file"
+version = "1.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "93fc1dc3aaa9bfed95e02e6eadabb4baf7e3078b0bd1b4d7b6b0b68378900502"
+dependencies = [
+ "winapi-util",
+]
+
+[[package]]
+name = "schannel"
+version = "0.1.28"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "891d81b926048e76efe18581bf793546b4c0eaf8448d72be8de2bbee5fd166e1"
+dependencies = [
+ "windows-sys 0.61.0",
+]
+
+[[package]]
+name = "security-framework"
+version = "2.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "897b2245f0b511c87893af39b033e5ca9cce68824c4d7e7630b5a1d339658d02"
+dependencies = [
+ "bitflags 2.9.4",
+ "core-foundation",
+ "core-foundation-sys",
+ "libc",
+ "security-framework-sys",
+]
+
+[[package]]
+name = "security-framework-sys"
+version = "2.15.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cc1f0cbffaac4852523ce30d8bd3c5cdc873501d96ff467ca09b6767bb8cd5c0"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
+[[package]]
+name = "seq-macro"
+version = "0.3.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1bc711410fbe7399f390ca1c3b60ad0f53f80e95c5eb935e52268a0e2cd49acc"
+
+[[package]]
+name = "serde"
+version = "1.0.226"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0dca6411025b24b60bfa7ec1fe1f8e710ac09782dca409ee8237ba74b51295fd"
+dependencies = [
+ "serde_core",
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_core"
+version = "1.0.226"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba2ba63999edb9dac981fb34b3e5c0d111a69b0924e253ed29d83f7c99e966a4"
+dependencies = [
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_derive"
+version = "1.0.226"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8db53ae22f34573731bafa1db20f04027b2d25e02d8205921b569171699cdb33"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "serde_html_form"
+version = "0.2.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b2f2d7ff8a2140333718bb329f5c40fc5f0865b84c426183ce14c97d2ab8154f"
+dependencies = [
+ "form_urlencoded",
+ "indexmap",
+ "itoa",
+ "ryu",
+ "serde_core",
+]
+
+[[package]]
+name = "serde_json"
+version = "1.0.145"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "402a6f66d8c709116cf22f558eab210f5a50187f702eb4d7e5ef38d9a7f1c79c"
+dependencies = [
+ "itoa",
+ "memchr",
+ "ryu",
+ "serde",
+ "serde_core",
+]
+
+[[package]]
+name = "serde_plain"
+version = "1.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9ce1fc6db65a611022b23a0dec6975d63fb80a302cb3388835ff02c097258d50"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "serde_spanned"
+version = "0.6.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bf41e0cfaf7226dca15e8197172c295a782857fcb97fad1808a166870dee75a3"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "serde_urlencoded"
+version = "0.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d3491c14715ca2294c4d6a88f15e84739788c1d030eed8c110436aafdaa2f3fd"
+dependencies = [
+ "form_urlencoded",
+ "itoa",
+ "ryu",
+ "serde",
+]
+
+[[package]]
+name = "shlex"
+version = "1.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0fda2ff0d084019ba4d7c6f371c95d8fd75ce3524c3cb8fb653a3023f6323e64"
+
+[[package]]
+name = "simd-adler32"
+version = "0.3.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d66dc143e6b11c1eddc06d5c423cfc97062865baf299914ab64caa38182078fe"
+
+[[package]]
+name = "simd_helpers"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "95890f873bec569a0362c235787f3aca6e1e887302ba4840839bcc6459c42da6"
+dependencies = [
+ "quote",
+]
+
+[[package]]
+name = "slab"
+version = "0.4.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7a2ae44ef20feb57a68b23d846850f861394c2e02dc425a50098ae8c90267589"
+
+[[package]]
+name = "smallvec"
+version = "1.15.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "67b1b7a3b5fe4f1376887184045fcf45c69e92af734b7aaddc05fb777b6fbd03"
+
+[[package]]
+name = "socket2"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "233504af464074f9d066d7b5416c5f9b894a5862a6506e306f7b816cdd6f1807"
+dependencies = [
+ "libc",
+ "windows-sys 0.59.0",
+]
+
+[[package]]
+name = "spm_precompiled"
+version = "0.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5851699c4033c63636f7ea4cf7b7c1f1bf06d0cc03cfb42e711de5a5c46cf326"
+dependencies = [
+ "base64 0.13.1",
+ "nom",
+ "serde",
+ "unicode-segmentation",
+]
+
+[[package]]
+name = "stable_deref_trait"
+version = "1.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a8f112729512f8e442d81f95a8a7ddf2b7c6b8a1a6f509a95864142b30cab2d3"
+
+[[package]]
+name = "static_assertions"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a2eb9349b6444b326872e140eb1cf5e7c522154d69e7a0ffb0fb81c06b37543f"
+
+[[package]]
+name = "strsim"
+version = "0.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7da8b5736845d9f2fcb837ea5d9e2628564b3b043a70948a3f0b778838c5fb4f"
+
+[[package]]
+name = "subtle"
+version = "2.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "13c2bddecc57b384dee18652358fb23172facb8a2c51ccc10d74c157bdea3292"
+
+[[package]]
+name = "syn"
+version = "2.0.106"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ede7c438028d4436d71104916910f5bb611972c5cfd7f89b8300a8186e6fada6"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "sync_wrapper"
+version = "1.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0bf256ce5efdfa370213c1dabab5935a12e49f2c58d15e9eac2870d3b4f27263"
+dependencies = [
+ "futures-core",
+]
+
+[[package]]
+name = "synstructure"
+version = "0.13.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "728a70f3dbaf5bab7f0c4b1ac8d7ae5ea60a4b5549c8a5914361c99147a709d2"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "sysctl"
+version = "0.5.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ec7dddc5f0fee506baf8b9fdb989e242f17e4b11c61dfbb0635b705217199eea"
+dependencies = [
+ "bitflags 2.9.4",
+ "byteorder",
+ "enum-as-inner",
+ "libc",
+ "thiserror 1.0.69",
+ "walkdir",
+]
+
+[[package]]
+name = "sysctl"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "01198a2debb237c62b6826ec7081082d951f46dbb64b0e8c7649a452230d1dfc"
+dependencies = [
+ "bitflags 2.9.4",
+ "byteorder",
+ "enum-as-inner",
+ "libc",
+ "thiserror 1.0.69",
+ "walkdir",
+]
+
+[[package]]
+name = "system-configuration"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3c879d448e9d986b661742763247d3693ed13609438cf3d006f51f5368a5ba6b"
+dependencies = [
+ "bitflags 2.9.4",
+ "core-foundation",
+ "system-configuration-sys",
+]
+
+[[package]]
+name = "system-configuration-sys"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e1d1b10ced5ca923a1fcb8d03e96b8d3268065d724548c0211415ff6ac6bac4"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
+[[package]]
+name = "system-deps"
+version = "6.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a3e535eb8dded36d55ec13eddacd30dec501792ff23a0b1682c38601b8cf2349"
+dependencies = [
+ "cfg-expr",
+ "heck",
+ "pkg-config",
+ "toml",
+ "version-compare",
+]
+
+[[package]]
+name = "target-lexicon"
+version = "0.12.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "61c41af27dd6d1e27b1b16b489db798443478cef1f06a660c96db617ba5de3b1"
+
+[[package]]
+name = "tempfile"
+version = "3.22.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "84fa4d11fadde498443cca10fd3ac23c951f0dc59e080e9f4b93d4df4e4eea53"
+dependencies = [
+ "fastrand",
+ "getrandom 0.3.3",
+ "once_cell",
+ "rustix",
+ "windows-sys 0.61.0",
+]
+
+[[package]]
+name = "thiserror"
+version = "1.0.69"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6aaf5339b578ea85b50e080feb250a3e8ae8cfcdff9a461c9ec2904bc923f52"
+dependencies = [
+ "thiserror-impl 1.0.69",
+]
+
+[[package]]
+name = "thiserror"
+version = "2.0.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3467d614147380f2e4e374161426ff399c91084acd2363eaf549172b3d5e60c0"
+dependencies = [
+ "thiserror-impl 2.0.16",
+]
+
+[[package]]
+name = "thiserror-impl"
+version = "1.0.69"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4fee6c4efc90059e10f81e6d42c60a18f76588c3d74cb83a0b242a2b6c7504c1"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "thiserror-impl"
+version = "2.0.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c5e1be1c48b9172ee610da68fd9cd2770e7a4056cb3fc98710ee6906f0c7960"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "tiff"
+version = "0.10.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "af9605de7fee8d9551863fd692cce7637f548dbd9db9180fcc07ccc6d26c336f"
+dependencies = [
+ "fax",
+ "flate2",
+ "half",
+ "quick-error",
+ "weezl",
+ "zune-jpeg",
+]
+
+[[package]]
+name = "tinystr"
+version = "0.8.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5d4f6d1145dcb577acf783d4e601bc1d76a13337bb54e6233add580b07344c8b"
+dependencies = [
+ "displaydoc",
+ "zerovec",
+]
+
+[[package]]
+name = "tokenizers"
+version = "0.22.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6475a27088c98ea96d00b39a9ddfb63780d1ad4cceb6f48374349a96ab2b7842"
+dependencies = [
+ "ahash",
+ "aho-corasick",
+ "compact_str",
+ "dary_heap",
+ "derive_builder",
+ "esaxx-rs",
+ "getrandom 0.3.3",
+ "indicatif",
+ "itertools 0.14.0",
+ "log",
+ "macro_rules_attribute",
+ "monostate",
+ "onig",
+ "paste",
+ "rand 0.9.2",
+ "rayon",
+ "rayon-cond",
+ "regex",
+ "regex-syntax",
+ "serde",
+ "serde_json",
+ "spm_precompiled",
+ "thiserror 2.0.16",
+ "unicode-normalization-alignments",
+ "unicode-segmentation",
+ "unicode_categories",
+]
+
+[[package]]
+name = "tokio"
+version = "1.47.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "89e49afdadebb872d3145a5638b59eb0691ea23e46ca484037cfab3b76b95038"
+dependencies = [
+ "backtrace",
+ "bytes",
+ "io-uring",
+ "libc",
+ "mio",
+ "pin-project-lite",
+ "slab",
+ "socket2",
+ "windows-sys 0.59.0",
+]
+
+[[package]]
+name = "tokio-native-tls"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bbae76ab933c85776efabc971569dd6119c580d8f5d448769dec1764bf796ef2"
+dependencies = [
+ "native-tls",
+ "tokio",
+]
+
+[[package]]
+name = "tokio-rustls"
+version = "0.26.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05f63835928ca123f1bef57abbcd23bb2ba0ac9ae1235f1e65bda0d06e7786bd"
+dependencies = [
+ "rustls",
+ "tokio",
+]
+
+[[package]]
+name = "tokio-util"
+version = "0.7.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "14307c986784f72ef81c89db7d9e28d6ac26d16213b109ea501696195e6e3ce5"
+dependencies = [
+ "bytes",
+ "futures-core",
+ "futures-sink",
+ "pin-project-lite",
+ "tokio",
+]
+
+[[package]]
+name = "toml"
+version = "0.8.23"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc1beb996b9d83529a9e75c17a1686767d148d70663143c7854d8b4a09ced362"
+dependencies = [
+ "serde",
+ "serde_spanned",
+ "toml_datetime 0.6.11",
+ "toml_edit 0.22.27",
+]
+
+[[package]]
+name = "toml_datetime"
+version = "0.6.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "22cddaf88f4fbc13c51aebbf5f8eceb5c7c5a9da2ac40a13519eb5b0a0e8f11c"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "toml_datetime"
+version = "0.7.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32f1085dec27c2b6632b04c80b3bb1b4300d6495d1e129693bdda7d91e72eec1"
+dependencies = [
+ "serde_core",
+]
+
+[[package]]
+name = "toml_edit"
+version = "0.22.27"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "41fe8c660ae4257887cf66394862d21dbca4a6ddd26f04a3560410406a2f819a"
+dependencies = [
+ "indexmap",
+ "serde",
+ "serde_spanned",
+ "toml_datetime 0.6.11",
+ "winnow",
+]
+
+[[package]]
+name = "toml_edit"
+version = "0.23.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f3effe7c0e86fdff4f69cdd2ccc1b96f933e24811c5441d44904e8683e27184b"
+dependencies = [
+ "indexmap",
+ "toml_datetime 0.7.2",
+ "toml_parser",
+ "winnow",
+]
+
+[[package]]
+name = "toml_parser"
+version = "1.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4cf893c33be71572e0e9aa6dd15e6677937abd686b066eac3f8cd3531688a627"
+dependencies = [
+ "winnow",
+]
+
+[[package]]
+name = "tower"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d039ad9159c98b70ecfd540b2573b97f7f52c3e8d9f8ad57a24b916a536975f9"
+dependencies = [
+ "futures-core",
+ "futures-util",
+ "pin-project-lite",
+ "sync_wrapper",
+ "tokio",
+ "tower-layer",
+ "tower-service",
+]
+
+[[package]]
+name = "tower-http"
+version = "0.6.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "adc82fd73de2a9722ac5da747f12383d2bfdb93591ee6c58486e0097890f05f2"
+dependencies = [
+ "bitflags 2.9.4",
+ "bytes",
+ "futures-util",
+ "http",
+ "http-body",
+ "iri-string",
+ "pin-project-lite",
+ "tower",
+ "tower-layer",
+ "tower-service",
+]
+
+[[package]]
+name = "tower-layer"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "121c2a6cda46980bb0fcd1647ffaf6cd3fc79a013de288782836f6df9c48780e"
+
+[[package]]
+name = "tower-service"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8df9b6e13f2d32c91b9bd719c00d1958837bc7dec474d94952798cc8e69eeec3"
+
+[[package]]
+name = "tracing"
+version = "0.1.41"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "784e0ac535deb450455cbfa28a6f0df145ea1bb7ae51b821cf5e7927fdcfbdd0"
+dependencies = [
+ "pin-project-lite",
+ "tracing-attributes",
+ "tracing-core",
+]
+
+[[package]]
+name = "tracing-attributes"
+version = "0.1.30"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "81383ab64e72a7a8b8e13130c49e3dab29def6d0c7d76a03087b3cf71c5c6903"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "tracing-core"
+version = "0.1.34"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9d12581f227e93f094d3af2ae690a574abb8a2b9b7a96e7cfe9647b2b617678"
+dependencies = [
+ "once_cell",
+]
+
+[[package]]
+name = "try-lock"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"
+
+[[package]]
+name = "ug"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "76b761acf8af3494640d826a8609e2265e19778fb43306c7f15379c78c9b05b0"
+dependencies = [
+ "gemm 0.18.2",
+ "half",
+ "libloading",
+ "memmap2",
+ "num",
+ "num-traits",
+ "num_cpus",
+ "rayon",
+ "safetensors",
+ "serde",
+ "thiserror 1.0.69",
+ "tracing",
+ "yoke 0.7.5",
+]
+
+[[package]]
+name = "ug-cuda"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9f0a1fa748f26166778c33b8498255ebb7c6bffb472bcc0a72839e07ebb1d9b5"
+dependencies = [
+ "cudarc",
+ "half",
+ "serde",
+ "thiserror 1.0.69",
+ "ug",
+]
+
+[[package]]
+name = "unicase"
+version = "2.8.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "75b844d17643ee918803943289730bec8aac480150456169e647ed0b576ba539"
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.19"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f63a545481291138910575129486daeaf8ac54aee4387fe7906919f7830c7d9d"
+
+[[package]]
+name = "unicode-normalization-alignments"
+version = "0.1.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "43f613e4fa046e69818dd287fdc4bc78175ff20331479dab6e1b0f98d57062de"
+dependencies = [
+ "smallvec",
+]
+
+[[package]]
+name = "unicode-segmentation"
+version = "1.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f6ccf251212114b54433ec949fd6a7841275f9ada20dddd2f29e9ceea4501493"
+
+[[package]]
+name = "unicode-width"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4a1a07cc7db3810833284e8d372ccdc6da29741639ecc70c9ec107df0fa6154c"
+
+[[package]]
+name = "unicode_categories"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "39ec24b3121d976906ece63c9daad25b85969647682eee313cb5779fdd69e14e"
+
+[[package]]
+name = "untrusted"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8ecb6da28b8a351d773b68d5825ac39017e680750f980f3a1a85cd8dd28a47c1"
+
+[[package]]
+name = "url"
+version = "2.5.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "08bc136a29a3d1758e07a9cca267be308aeebf5cfd5a10f3f67ab2097683ef5b"
+dependencies = [
+ "form_urlencoded",
+ "idna",
+ "percent-encoding",
+ "serde",
+]
+
+[[package]]
+name = "utf8_iter"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6c140620e7ffbb22c2dee59cafe6084a59b5ffc27a8859a5f0d494b5d52b6be"
+
+[[package]]
+name = "v_frame"
+version = "0.3.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "666b7727c8875d6ab5db9533418d7c764233ac9c0cff1d469aec8fa127597be2"
+dependencies = [
+ "aligned-vec",
+ "num-traits",
+ "wasm-bindgen",
+]
+
+[[package]]
+name = "vcpkg"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "accd4ea62f7bb7a82fe23066fb0957d48ef677f6eeb8215f372f52e48bb32426"
+
+[[package]]
+name = "version-compare"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "852e951cb7832cb45cb1169900d19760cfa39b82bc0ea9c0e5a14ae88411c98b"
+
+[[package]]
+name = "version_check"
+version = "0.9.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
+
+[[package]]
+name = "walkdir"
+version = "2.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "29790946404f91d9c5d06f9874efddea1dc06c5efe94541a7d6863108e3a5e4b"
+dependencies = [
+ "same-file",
+ "winapi-util",
+]
+
+[[package]]
+name = "want"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bfa7760aed19e106de2c7c0b581b509f2f25d3dacaf737cb82ac61bc6d760b0e"
+dependencies = [
+ "try-lock",
+]
+
+[[package]]
+name = "wasi"
+version = "0.11.1+wasi-snapshot-preview1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ccf3ec651a847eb01de73ccad15eb7d99f80485de043efb2f370cd654f4ea44b"
+
+[[package]]
+name = "wasi"
+version = "0.14.7+wasi-0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "883478de20367e224c0090af9cf5f9fa85bed63a95c1abf3afc5c083ebc06e8c"
+dependencies = [
+ "wasip2",
+]
+
+[[package]]
+name = "wasip2"
+version = "1.0.1+wasi-0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0562428422c63773dad2c345a1882263bbf4d65cf3f42e90921f787ef5ad58e7"
+dependencies = [
+ "wit-bindgen",
+]
+
+[[package]]
+name = "wasm-bindgen"
+version = "0.2.103"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ab10a69fbd0a177f5f649ad4d8d3305499c42bab9aef2f7ff592d0ec8f833819"
+dependencies = [
+ "cfg-if",
+ "once_cell",
+ "rustversion",
+ "wasm-bindgen-macro",
+ "wasm-bindgen-shared",
+]
+
+[[package]]
+name = "wasm-bindgen-backend"
+version = "0.2.103"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0bb702423545a6007bbc368fde243ba47ca275e549c8a28617f56f6ba53b1d1c"
+dependencies = [
+ "bumpalo",
+ "log",
+ "proc-macro2",
+ "quote",
+ "syn",
+ "wasm-bindgen-shared",
+]
+
+[[package]]
+name = "wasm-bindgen-futures"
+version = "0.4.53"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a0b221ff421256839509adbb55998214a70d829d3a28c69b4a6672e9d2a42f67"
+dependencies = [
+ "cfg-if",
+ "js-sys",
+ "once_cell",
+ "wasm-bindgen",
+ "web-sys",
+]
+
+[[package]]
+name = "wasm-bindgen-macro"
+version = "0.2.103"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fc65f4f411d91494355917b605e1480033152658d71f722a90647f56a70c88a0"
+dependencies = [
+ "quote",
+ "wasm-bindgen-macro-support",
+]
+
+[[package]]
+name = "wasm-bindgen-macro-support"
+version = "0.2.103"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ffc003a991398a8ee604a401e194b6b3a39677b3173d6e74495eb51b82e99a32"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+ "wasm-bindgen-backend",
+ "wasm-bindgen-shared",
+]
+
+[[package]]
+name = "wasm-bindgen-shared"
+version = "0.2.103"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "293c37f4efa430ca14db3721dfbe48d8c33308096bd44d80ebaa775ab71ba1cf"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "wasm-streams"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "15053d8d85c7eccdbefef60f06769760a563c7f0a9d6902a13d35c7800b0ad65"
+dependencies = [
+ "futures-util",
+ "js-sys",
+ "wasm-bindgen",
+ "wasm-bindgen-futures",
+ "web-sys",
+]
+
+[[package]]
+name = "web-sys"
+version = "0.3.80"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fbe734895e869dc429d78c4b433f8d17d95f8d05317440b4fad5ab2d33e596dc"
+dependencies = [
+ "js-sys",
+ "wasm-bindgen",
+]
+
+[[package]]
+name = "web-time"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5a6580f308b1fad9207618087a65c04e7a10bc77e02c8e84e9b00dd4b12fa0bb"
+dependencies = [
+ "js-sys",
+ "wasm-bindgen",
+]
+
+[[package]]
+name = "weezl"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a751b3277700db47d3e574514de2eced5e54dc8a5436a3bf7a0b248b2cee16f3"
+
+[[package]]
+name = "winapi-util"
+version = "0.1.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c2a7b1c03c876122aa43f3020e6c3c3ee5c05081c9a00739faf7503aeba10d22"
+dependencies = [
+ "windows-sys 0.61.0",
+]
+
+[[package]]
+name = "windows-link"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5e6ad25900d524eaabdbbb96d20b4311e1e7ae1699af4fb28c17ae66c80d798a"
+
+[[package]]
+name = "windows-link"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "45e46c0661abb7180e7b9c281db115305d49ca1709ab8242adf09666d2173c65"
+
+[[package]]
+name = "windows-registry"
+version = "0.5.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5b8a9ed28765efc97bbc954883f4e6796c33a06546ebafacbabee9696967499e"
+dependencies = [
+ "windows-link 0.1.3",
+ "windows-result",
+ "windows-strings",
+]
+
+[[package]]
+name = "windows-result"
+version = "0.3.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "56f42bd332cc6c8eac5af113fc0c1fd6a8fd2aa08a0119358686e5160d0586c6"
+dependencies = [
+ "windows-link 0.1.3",
+]
+
+[[package]]
+name = "windows-strings"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "56e6c93f3a0c3b36176cb1327a4958a0353d5d166c2a35cb268ace15e91d3b57"
+dependencies = [
+ "windows-link 0.1.3",
+]
+
+[[package]]
+name = "windows-sys"
+version = "0.52.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "282be5f36a8ce781fad8c8ae18fa3f9beff57ec1b52cb3de0789201425d9a33d"
+dependencies = [
+ "windows-targets 0.52.6",
+]
+
+[[package]]
+name = "windows-sys"
+version = "0.59.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e38bc4d79ed67fd075bcc251a1c39b32a1776bbe92e5bef1f0bf1f8c531853b"
+dependencies = [
+ "windows-targets 0.52.6",
+]
+
+[[package]]
+name = "windows-sys"
+version = "0.61.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e201184e40b2ede64bc2ea34968b28e33622acdbbf37104f0e4a33f7abe657aa"
+dependencies = [
+ "windows-link 0.2.0",
+]
+
+[[package]]
+name = "windows-targets"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973"
+dependencies = [
+ "windows_aarch64_gnullvm 0.52.6",
+ "windows_aarch64_msvc 0.52.6",
+ "windows_i686_gnu 0.52.6",
+ "windows_i686_gnullvm 0.52.6",
+ "windows_i686_msvc 0.52.6",
+ "windows_x86_64_gnu 0.52.6",
+ "windows_x86_64_gnullvm 0.52.6",
+ "windows_x86_64_msvc 0.52.6",
+]
+
+[[package]]
+name = "windows-targets"
+version = "0.53.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d5fe6031c4041849d7c496a8ded650796e7b6ecc19df1a431c1a363342e5dc91"
+dependencies = [
+ "windows-link 0.1.3",
+ "windows_aarch64_gnullvm 0.53.0",
+ "windows_aarch64_msvc 0.53.0",
+ "windows_i686_gnu 0.53.0",
+ "windows_i686_gnullvm 0.53.0",
+ "windows_i686_msvc 0.53.0",
+ "windows_x86_64_gnu 0.53.0",
+ "windows_x86_64_gnullvm 0.53.0",
+ "windows_x86_64_msvc 0.53.0",
+]
+
+[[package]]
+name = "windows_aarch64_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3"
+
+[[package]]
+name = "windows_aarch64_gnullvm"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "86b8d5f90ddd19cb4a147a5fa63ca848db3df085e25fee3cc10b39b6eebae764"
+
+[[package]]
+name = "windows_aarch64_msvc"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469"
+
+[[package]]
+name = "windows_aarch64_msvc"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c7651a1f62a11b8cbd5e0d42526e55f2c99886c77e007179efff86c2b137e66c"
+
+[[package]]
+name = "windows_i686_gnu"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e9b5ad5ab802e97eb8e295ac6720e509ee4c243f69d781394014ebfe8bbfa0b"
+
+[[package]]
+name = "windows_i686_gnu"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c1dc67659d35f387f5f6c479dc4e28f1d4bb90ddd1a5d3da2e5d97b42d6272c3"
+
+[[package]]
+name = "windows_i686_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66"
+
+[[package]]
+name = "windows_i686_gnullvm"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9ce6ccbdedbf6d6354471319e781c0dfef054c81fbc7cf83f338a4296c0cae11"
+
+[[package]]
+name = "windows_i686_msvc"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66"
+
+[[package]]
+name = "windows_i686_msvc"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "581fee95406bb13382d2f65cd4a908ca7b1e4c2f1917f143ba16efe98a589b5d"
+
+[[package]]
+name = "windows_x86_64_gnu"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78"
+
+[[package]]
+name = "windows_x86_64_gnu"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2e55b5ac9ea33f2fc1716d1742db15574fd6fc8dadc51caab1c16a3d3b4190ba"
+
+[[package]]
+name = "windows_x86_64_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d"
+
+[[package]]
+name = "windows_x86_64_gnullvm"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0a6e035dd0599267ce1ee132e51c27dd29437f63325753051e71dd9e42406c57"
+
+[[package]]
+name = "windows_x86_64_msvc"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"
+
+[[package]]
+name = "windows_x86_64_msvc"
+version = "0.53.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "271414315aff87387382ec3d271b52d7ae78726f5d44ac98b4f4030c91880486"
+
+[[package]]
+name = "winnow"
+version = "0.7.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "21a0236b59786fed61e2a80582dd500fe61f18b5dca67a4a067d0bc9039339cf"
+dependencies = [
+ "memchr",
+]
+
+[[package]]
+name = "wit-bindgen"
+version = "0.46.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f17a85883d4e6d00e8a97c586de764dabcc06133f7f1d55dce5cdc070ad7fe59"
+
+[[package]]
+name = "writeable"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ea2f10b9bb0928dfb1b42b65e1f9e36f7f54dbdf08457afefb38afcdec4fa2bb"
+
+[[package]]
+name = "yoke"
+version = "0.7.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "120e6aef9aa629e3d4f52dc8cc43a015c7724194c97dfaf45180d2daf2b77f40"
+dependencies = [
+ "serde",
+ "stable_deref_trait",
+ "yoke-derive 0.7.5",
+ "zerofrom",
+]
+
+[[package]]
+name = "yoke"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5f41bb01b8226ef4bfd589436a297c53d118f65921786300e427be8d487695cc"
+dependencies = [
+ "serde",
+ "stable_deref_trait",
+ "yoke-derive 0.8.0",
+ "zerofrom",
+]
+
+[[package]]
+name = "yoke-derive"
+version = "0.7.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2380878cad4ac9aac1e2435f3eb4020e8374b5f13c296cb75b4620ff8e229154"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+ "synstructure",
+]
+
+[[package]]
+name = "yoke-derive"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "38da3c9736e16c5d3c8c597a9aaa5d1fa565d0532ae05e27c24aa62fb32c0ab6"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+ "synstructure",
+]
+
+[[package]]
+name = "zerocopy"
+version = "0.8.27"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0894878a5fa3edfd6da3f88c4805f4c8558e2b996227a3d864f47fe11e38282c"
+dependencies = [
+ "zerocopy-derive",
+]
+
+[[package]]
+name = "zerocopy-derive"
+version = "0.8.27"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "88d2b8d9c68ad2b9e4340d7832716a4d21a22a1154777ad56ea55c51a9cf3831"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "zerofrom"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "50cc42e0333e05660c3587f3bf9d0478688e15d870fab3346451ce7f8c9fbea5"
+dependencies = [
+ "zerofrom-derive",
+]
+
+[[package]]
+name = "zerofrom-derive"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d71e5d6e06ab090c67b5e44993ec16b72dcbaabc526db883a360057678b48502"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+ "synstructure",
+]
+
+[[package]]
+name = "zeroize"
+version = "1.8.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ced3678a2879b30306d323f4542626697a464a97c0a07c9aebf7ebca65cd4dde"
+
+[[package]]
+name = "zerotrie"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "36f0bbd478583f79edad978b407914f61b2972f5af6fa089686016be8f9af595"
+dependencies = [
+ "displaydoc",
+ "yoke 0.8.0",
+ "zerofrom",
+]
+
+[[package]]
+name = "zerovec"
+version = "0.11.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e7aa2bd55086f1ab526693ecbe444205da57e25f4489879da80635a46d90e73b"
+dependencies = [
+ "yoke 0.8.0",
+ "zerofrom",
+ "zerovec-derive",
+]
+
+[[package]]
+name = "zerovec-derive"
+version = "0.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5b96237efa0c878c64bd89c436f661be4e46b2f3eff1ebb976f7ef2321d2f58f"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "zip"
+version = "1.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9cc23c04387f4da0374be4533ad1208cbb091d5c11d070dfef13676ad6497164"
+dependencies = [
+ "arbitrary",
+ "crc32fast",
+ "crossbeam-utils",
+ "displaydoc",
+ "indexmap",
+ "num_enum",
+ "thiserror 1.0.69",
+]
+
+[[package]]
+name = "zune-core"
+version = "0.4.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3f423a2c17029964870cfaabb1f13dfab7d092a62a29a89264f4d36990ca414a"
+
+[[package]]
+name = "zune-inflate"
+version = "0.2.54"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "73ab332fe2f6680068f3582b16a24f90ad7096d5d39b974d1c0aff0125116f02"
+dependencies = [
+ "simd-adler32",
+]
+
+[[package]]
+name = "zune-jpeg"
+version = "0.4.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "29ce2c8a9384ad323cf564b67da86e21d3cfdff87908bc1223ed5c99bc792713"
+dependencies = [
+ "zune-core",
+]
diff --git a/Cargo.toml b/Cargo.toml
index 4f7c6c7..bc67317 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -7,3 +7,21 @@ license = "Apache-2.0"
 description = "aha"
 
 [dependencies]
+candle-core = { git = "https://github.com/huggingface/candle.git", version = "0.9.1"}
+candle-nn = { git = "https://github.com/huggingface/candle.git", version = "0.9.1"}
+candle-transformers = { git = "https://github.com/huggingface/candle.git", version = "0.9.1"}
+candle-flash-attn = { git = "https://github.com/huggingface/candle.git", optional = true }
+serde = "1.0.226"
+serde_json = "1.0.145"
+anyhow = "1.0.100"
+ffmpeg-next = "8.0.0"
+image = "0.25.8"
+reqwest = { version = "0.12.23", features = ["blocking"] }
+base64 = "0.22.1"
+num = "0.4.3"
+minijinja = "2.12.0"
+tokenizers = "0.22.1"
+openai_dive = "1.3.0"
+[features]
+flash-attn=["candle-flash-attn"]
+cuda=["candle-nn/cuda", "candle-core/cuda", "candle-transformers/cuda"]
diff --git a/assets/img/ocr_test.png b/assets/img/ocr_test.png
new file mode 100644
index 0000000..b69f68e
Binary files /dev/null and b/assets/img/ocr_test.png differ
diff --git a/assets/img/ocr_test1.png b/assets/img/ocr_test1.png
new file mode 100644
index 0000000..39ab8dd
Binary files /dev/null and b/assets/img/ocr_test1.png differ
diff --git a/src/chat_template/chat_template.rs b/src/chat_template/chat_template.rs
new file mode 100644
index 0000000..b290195
--- /dev/null
+++ b/src/chat_template/chat_template.rs
@@ -0,0 +1,104 @@
+use crate::utils::utils::string_to_static_str;
+use anyhow::{Result, anyhow};
+use minijinja::{Environment, Value as MiniJinjaValue, context};
+use openai_dive::v1::resources::chat::ChatCompletionParameters;
+
+pub fn get_template(path: String) -> Result<String> {
+    let tokenizer_config_file = path.clone() + "/tokenizer_config.json";
+    assert!(
+        std::path::Path::new(&tokenizer_config_file).exists(),
+        "tokenizer_config.json not exists in model path"
+    );
+    let tokenizer_config: serde_json::Value =
+        serde_json::from_slice(&std::fs::read(tokenizer_config_file)?)
+            .map_err(|e| anyhow!(format!("load tokenizer_config file error:{}", e)))?;
+    let chat_template = tokenizer_config["chat_template"]
+        .as_str()
+        .ok_or(anyhow!(format!("chat_template to str error")))?;
+    // 修复模板中的问题行
+    let fixed_template = chat_template
+        .replace(
+            "message.content.startswith('<tool_response>')",
+            "message.content is startingwith('<tool_response>')", // 使用minijinja中的 is startingwith 替换
+        )
+        .replace(
+            "message.content.endswith('</tool_response>')",
+            "message.content is endingwith('</tool_response>')", // 使用minijinja中的 is endingwith 替换
+        )
+        .replace(
+            "content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n')",
+            "((content | split('</think>'))[0] | rstrip('\\n') | split('<think>'))[-1] | lstrip('\\n')", // 使用自定义的split, rstrip, lstrip过滤器替换
+        )
+        .replace(
+            "content.split('</think>')[-1].lstrip('\\n')",
+            "(content | split('</think>'))[-1] | lstrip('\\n')", // 使用自定义的过滤器替换
+        )
+        .replace(
+            "reasoning_content.strip('\\n')",
+            "reasoning_content | strip('\\n')", // 使用自定义的过滤器替换
+        )
+        .replace(
+            "content.lstrip('\\n')",
+            "content | lstrip('\\n')", // 使用自定义的过滤器替换
+        );
+    Ok(fixed_template)
+}
+
+pub struct ChatTemplate<'a> {
+    env: Environment<'a>,
+}
+
+impl<'a> ChatTemplate<'a> {
+    pub fn init(path: &str) -> Result<Self> {
+        let path = path.to_string();
+        assert!(
+            std::path::Path::new(&path).exists(),
+            "model path file not exists"
+        );
+        let template = get_template(path)?;
+        let template = string_to_static_str(template);
+        // 加载jinjaenv处理chat_template
+        let mut env = Environment::new();
+        // 添加自定义过滤器
+        env.add_filter("tojson", |v: MiniJinjaValue| {
+            serde_json::to_string(&v).unwrap()
+        });
+
+        env.add_filter("split", |s: String, delimiter: String| {
+            s.split(&delimiter)
+                .map(|s| s.to_string())
+                .collect::<Vec<String>>()
+        });
+
+        // 添加 lstrip 过滤器
+        env.add_filter("lstrip", |s: String, chars: Option<String>| match chars {
+            Some(chars_str) => s.trim_start_matches(chars_str.as_str()).to_string(),
+            None => s.trim_start().to_string(),
+        });
+
+        // 添加 rstrip 过滤器
+        env.add_filter("rstrip", |s: String, chars: Option<String>| match chars {
+            Some(chars_str) => s.trim_end_matches(chars_str.as_str()).to_string(),
+            None => s.trim_end().to_string(),
+        });
+        // let template = get_template(path.to_string())?;
+        let _ = env.add_template("chat", template);
+
+        Ok(Self { env })
+    }
+
+    pub fn apply_chat_template(&self, messages: &ChatCompletionParameters) -> Result<String> {
+        let context = context! {
+            messages => &messages.messages,
+            add_generation_prompt => true,
+        };
+        let template = self
+            .env
+            .get_template("chat")
+            .map_err(|e| anyhow!(format!("render template error {}", e)))?;
+        let message_str = template
+            .render(context)
+            .map_err(|e| anyhow!(format!("render template error {}", e)))?;
+        Ok(message_str)
+    }
+}
diff --git a/src/chat_template/mod.rs b/src/chat_template/mod.rs
new file mode 100644
index 0000000..6854abd
--- /dev/null
+++ b/src/chat_template/mod.rs
@@ -0,0 +1 @@
+pub mod chat_template;
\ No newline at end of file
diff --git a/src/lib.rs b/src/lib.rs
index e69de29..897c0a9 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -0,0 +1,28 @@
+use crate::models::{GenerateModel, qwen2_5vl::generate::Qwen2_5VLGenerateModel};
+use anyhow::{Ok, Result};
+use candle_core::{DType, Device};
+pub mod chat_template;
+pub mod models;
+pub mod position_embed;
+pub mod tokenizer;
+pub mod utils;
+
+pub enum ModelType {
+    Qwen2_5VL,
+}
+
+impl ModelType {
+    pub fn init(
+        model_type: ModelType,
+        model_path: &str,
+        device: Option<&Device>,
+        dtype: Option<DType>,
+    ) -> Result<Box<dyn GenerateModel>> {
+        match model_type {
+            ModelType::Qwen2_5VL => {
+                let model = Qwen2_5VLGenerateModel::init(model_path, device, dtype)?;
+                Ok(Box::new(model))
+            }
+        }
+    }
+}
diff --git a/src/models/mod.rs b/src/models/mod.rs
new file mode 100644
index 0000000..10140c7
--- /dev/null
+++ b/src/models/mod.rs
@@ -0,0 +1,11 @@
+pub mod qwen2_5vl;
+use anyhow::Result;
+use candle_core::{DType, Device};
+use openai_dive::v1::resources::chat::ChatCompletionParameters;
+
+pub trait GenerateModel {
+    fn init(path: &str, device: Option<&Device>, dtype: Option<DType>) -> Result<Self>
+    where
+        Self: Sized;
+    fn generate(&mut self, mes: ChatCompletionParameters) -> Result<String>;
+}
diff --git a/src/models/qwen2_5vl/config.rs b/src/models/qwen2_5vl/config.rs
new file mode 100644
index 0000000..a3f3a9d
--- /dev/null
+++ b/src/models/qwen2_5vl/config.rs
@@ -0,0 +1,97 @@
+use candle_nn::Activation;
+
+#[derive(Debug, Clone, PartialEq, serde::Deserialize)]
+pub struct VisionConfig {
+    pub depth: usize,
+    pub hidden_act: Activation,
+    pub hidden_size: usize,
+    pub intermediate_size: usize,
+    pub num_heads: usize,
+    pub in_chans: usize,
+    pub out_hidden_size: usize,
+    pub patch_size: usize,
+    pub spatial_merge_size: usize,
+    pub spatial_patch_size: usize,
+    pub window_size: usize,
+    pub fullatt_block_indexes: Vec<usize>,
+    pub tokens_per_second: usize,
+    pub temporal_patch_size: usize,
+}
+#[derive(Debug, Clone, PartialEq, serde::Deserialize)]
+pub struct RopeScaling {
+    pub r#type: String,
+    pub mrope_section: Vec<usize>,
+}
+
+#[derive(Debug, Clone, PartialEq, serde::Deserialize)]
+pub struct Config {
+    pub attention_dropout: f32,
+    pub bos_token_id: usize,
+    pub eos_token_id: usize,
+    pub vision_start_token_id: usize,
+    pub vision_end_token_id: usize,
+    pub vision_token_id: usize,
+    pub image_token_id: usize,
+    pub video_token_id: usize,
+    pub hidden_act: Activation,
+    pub hidden_size: usize,
+    pub initializer_range: f32,
+    pub intermediate_size: usize,
+    pub max_position_embeddings: usize,
+    pub max_window_layers: usize,
+    pub num_attention_heads: usize,
+    pub num_hidden_layers: usize,
+    pub num_key_value_heads: usize,
+    pub rms_norm_eps: f64,
+    pub rope_theta: f32,
+    pub sliding_window: usize,
+    pub tie_word_embeddings: bool,
+    pub torch_dtype: String,
+    pub use_sliding_window: bool,
+    pub vision_config: VisionConfig,
+    pub rope_scaling: RopeScaling,
+    pub vocab_size: usize,
+}
+
+pub struct VisionSetting {
+    pub image_factor: u32,
+    pub min_pixels: u32,
+    pub max_pixels: u32,
+    pub max_ratio: u32,
+    pub temporal_patch_size: usize,
+    pub patch_size: usize,
+    pub merge_size: usize,
+    pub video_min_pixels: u32,
+    pub video_max_pixels: u32,
+    pub video_total_pixels: u32,
+    pub frame_factor: u32,
+    pub fps: f32,
+    pub fps_min_frames: u32,
+    pub fps_max_frames: u32,
+    pub image_mean: Vec<f32>,
+    pub image_std: Vec<f32>,
+}
+
+impl VisionSetting {
+    pub fn default() -> Self {
+        Self {
+            image_factor: 28,
+            min_pixels: 4 * 28 * 28,
+            max_pixels: 16384 * 28 * 28,
+            // max_pixels: 1000 * 28 * 28,
+            max_ratio: 200,
+            temporal_patch_size: 2,
+            patch_size: 14,
+            merge_size: 2,
+            video_min_pixels: 128 * 28 * 28,
+            video_max_pixels: 768 * 28 * 28,
+            video_total_pixels: 24576 * 28 * 28,
+            frame_factor: 2,
+            fps: 2.0,
+            fps_min_frames: 4,
+            fps_max_frames: 768,
+            image_mean: vec![0.48145466_f32, 0.4578275, 0.40821073],
+            image_std: vec![0.26862954, 0.26130258, 0.27577711],
+        }
+    }
+}
diff --git a/src/models/qwen2_5vl/generate.rs b/src/models/qwen2_5vl/generate.rs
new file mode 100644
index 0000000..6006834
--- /dev/null
+++ b/src/models/qwen2_5vl/generate.rs
@@ -0,0 +1,137 @@
+use crate::models::qwen2_5vl::config::Config;
+use crate::utils::utils::{find_safetensors_files, get_device, get_dtype};
+use crate::{
+    chat_template::chat_template::ChatTemplate,
+    models::{
+        GenerateModel,
+        qwen2_5vl::{model::Qwen2_5VLModel, processor::Qwen2_5VLProcessor},
+    },
+    tokenizer::tokenizer::TokenizerModel,
+};
+use anyhow::{Result, anyhow};
+use candle_core::{D, DType, Device, IndexOp, Tensor};
+use candle_nn::VarBuilder;
+use candle_transformers::generation::LogitsProcessor;
+use openai_dive::v1::resources::chat::ChatCompletionParameters;
+
+pub struct Qwen2_5VLGenerateModel<'a> {
+    chat_template: ChatTemplate<'a>,
+    tokenizer: TokenizerModel,
+    pre_processor: Qwen2_5VLProcessor,
+    qwen2_5_vl: Qwen2_5VLModel,
+    device: Device,
+    dtype: DType,
+}
+
+impl<'a> GenerateModel for Qwen2_5VLGenerateModel<'a> {
+    fn init(path: &str, device: Option<&Device>, dtype: Option<DType>) -> Result<Self> {
+        let chat_template = ChatTemplate::init(path)?;
+        let tokenizer = TokenizerModel::init(path)?;
+        let config_path = path.to_string() + "/config.json";
+        let cfg: Config = serde_json::from_slice(&std::fs::read(config_path)?)?;
+        let device = &get_device(device);
+        let cfg_dtype = cfg.torch_dtype.as_str();
+        let dtype = get_dtype(dtype, cfg_dtype);
+        let pre_processor = Qwen2_5VLProcessor::new(device, dtype)?;       
+        
+        let model_list = find_safetensors_files(&path)?;
+        let vb = unsafe { VarBuilder::from_mmaped_safetensors(&model_list, dtype, device)? };
+        let qwen2_5_vl = Qwen2_5VLModel::new(cfg, vb)?;
+        Ok(Qwen2_5VLGenerateModel {
+            chat_template,
+            tokenizer,
+            pre_processor,
+            qwen2_5_vl,
+            device: device.clone(),
+            dtype: dtype,
+        })
+    }
+    fn generate(&mut self, mes: ChatCompletionParameters) -> Result<String> {
+        let temperature = match mes.temperature {
+            Some(temp) => Some(temp as f64),
+            None => None,
+        };
+        let top_p = match mes.top_p {
+            Some(tp) => Some(tp as f64),
+            None => None,
+        };
+        let mut logit_processor = LogitsProcessor::new(34562, temperature, top_p);
+        let mes_render = self.chat_template.apply_chat_template(&mes)?;
+        let input = self.pre_processor.process_info(&mes, &mes_render)?;
+        let mut input_ids = self
+            .tokenizer
+            .text_encode(input.replace_text.clone(), &self.device)?;
+        let mut seq_len = input_ids.dim(1)?;
+        let mut seqlen_offset = 0;
+        let end_of_text_id = self.qwen2_5_vl.cfg.bos_token_id as u32;
+        let im_end_id = self.qwen2_5_vl.cfg.eos_token_id as u32;
+        let mut pixel_values = if input.pixel_values.is_some() {
+            Some(&input.pixel_values.unwrap().clone())
+        } else {
+            None
+        };
+        let image_grid_thw = if input.image_grid_thw.is_some() {
+            Some(&input.image_grid_thw.unwrap().clone())
+        } else {
+            None
+        };
+        let mut pixel_values_video = if input.pixel_values_video.is_some() {
+            Some(&input.pixel_values_video.unwrap().clone())
+        } else {
+            None
+        };
+        let video_grid_thw = if input.video_grid_thw.is_some() {
+            Some(&input.video_grid_thw.unwrap().clone())
+        } else {
+            None
+        };
+        let second_per_grid_ts = if input.second_per_grid_ts.is_some() {
+            Some(input.second_per_grid_ts.unwrap().clone())
+        } else {
+            None
+        };
+
+        let mut mask = Tensor::ones_like(&input_ids)?;
+        let mut cache_position = Tensor::ones_like(&input_ids.i(0)?)?
+            .to_dtype(candle_core::DType::F64)?
+            .cumsum(D::Minus1)?
+            .to_dtype(candle_core::DType::U32)?
+            .broadcast_sub(&Tensor::new(vec![1_u32], input_ids.device())?)?;
+
+        let mut generate = Vec::new();
+        let sample_len = match mes.max_tokens {
+            Some(max) => max,
+            None => 512,
+        };
+        for _ in 0..sample_len {
+            let logits = self.qwen2_5_vl.forward(
+                &input_ids,
+                pixel_values,
+                image_grid_thw,
+                pixel_values_video,
+                video_grid_thw,
+                &mask,
+                Some(&cache_position),
+                seqlen_offset,
+                second_per_grid_ts.clone(),
+            )?;
+            let logits = logits.squeeze(0)?.squeeze(0)?.to_dtype(DType::F32)?;
+            let next_token = logit_processor.sample(&logits)?;
+            generate.push(next_token);
+            if next_token == end_of_text_id || next_token == im_end_id {
+                break;
+            }
+            seqlen_offset += seq_len;
+            seq_len = 1;
+            input_ids = Tensor::from_vec(vec![next_token], (1, 1), &self.device)?;
+            let appendd_mask = Tensor::ones((1, 1), mask.dtype(), &self.device)?;
+            mask = Tensor::cat(&[mask, appendd_mask], 1)?;
+            cache_position = Tensor::from_vec(vec![seqlen_offset as u32], 1, &self.device)?;
+            pixel_values = None;
+            pixel_values_video = None;
+        }
+        let res = self.tokenizer.token_decode(generate)?;
+        self.qwen2_5_vl.clear_kv_cache();
+        Ok(res)
+    }
+}
diff --git a/src/models/qwen2_5vl/mod.rs b/src/models/qwen2_5vl/mod.rs
new file mode 100644
index 0000000..8d2989c
--- /dev/null
+++ b/src/models/qwen2_5vl/mod.rs
@@ -0,0 +1,4 @@
+pub mod config;
+pub mod model;
+pub mod processor;
+pub mod generate;
\ No newline at end of file
diff --git a/src/models/qwen2_5vl/model.rs b/src/models/qwen2_5vl/model.rs
new file mode 100644
index 0000000..0a32f00
--- /dev/null
+++ b/src/models/qwen2_5vl/model.rs
@@ -0,0 +1,1229 @@
+use crate::{
+    models::qwen2_5vl::config::{Config, RopeScaling},
+    position_embed::rope::{
+        Qwen2_5VLTextRotaryEmbedding, Qwen2_5VisionRotaryEmbedding, apply_rotary_pos_emb,
+        apply_rotary_pos_emb_vision,
+    },
+    utils::tensor_utils::{
+        get_equal_mask, get_vision_next_indices, masked_scatter_dim0, nonzero_index,
+        safe_arg_sort_last_dim, zero_index,
+    },
+};
+use anyhow::{Result, anyhow};
+use candle_core::{D, DType, Device, IndexOp, Tensor};
+use candle_nn::{
+    Activation, Init, Linear, Module, RmsNorm, VarBuilder, linear, linear_no_bias, rms_norm,
+};
+
+pub struct Qwen2_5VisionPatchEmbed {
+    conv3d_weight: Tensor,
+}
+
+impl Qwen2_5VisionPatchEmbed {
+    pub fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let patch_size = cfg.vision_config.patch_size;
+        let temporal_patch_size = cfg.vision_config.temporal_patch_size;
+        let in_channels = cfg.vision_config.in_chans;
+        let embed_dim = cfg.vision_config.hidden_size;
+        // conv3d weight key: visual.patch_embed.proj.weight, value: Tensor[dims 1280, 3, 2, 14, 14; bf16, cuda:0]
+        // (1280, 3, 2, 14, 14) -> (1280, 1176) -> (1176, 1280)
+        let conv3d_weight = vb
+            .get_with_hints(
+                (
+                    embed_dim,
+                    in_channels,
+                    temporal_patch_size,
+                    patch_size,
+                    patch_size,
+                ),
+                "proj.weight",
+                Init::Const(1.),
+            )?
+            .flatten(1, 4)?
+            .t()?;
+        Ok(Self { conv3d_weight })
+    }
+
+    pub fn forward(&self, hidden_states: &Tensor) -> Result<Tensor> {
+        // hidden_states shape:  (grid_t*grid_h*grid_w, c*temporal_patch_size*patch_size*patch_size)
+        // ((), 1176) matmul (1176, 1280) -> ((), 1280)
+        let hidden_states = hidden_states.matmul(&self.conv3d_weight)?;
+        Ok(hidden_states)
+    }
+}
+
+pub struct Qwen2_5VLPatchMerger {
+    hidden_size: usize,
+    ln_q: RmsNorm,
+    mlp_0: Linear,
+    mlp_2: Linear,
+}
+
+impl Qwen2_5VLPatchMerger {
+    pub fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let hidden_size =
+            cfg.vision_config.hidden_size * (cfg.vision_config.spatial_merge_size.pow(2));
+        let ln_q = rms_norm(
+            cfg.vision_config.hidden_size,
+            cfg.rms_norm_eps,
+            vb.pp("ln_q"),
+        )?;
+        let mlp_0 = linear(hidden_size, hidden_size, vb.pp("mlp.0"))?;
+        let mlp_2 = linear(
+            hidden_size,
+            cfg.vision_config.out_hidden_size,
+            vb.pp("mlp.2"),
+        )?;
+        Ok(Self {
+            hidden_size,
+            ln_q,
+            mlp_0,
+            mlp_2,
+        })
+    }
+}
+impl Module for Qwen2_5VLPatchMerger {
+    fn forward(&self, xs: &Tensor) -> candle_core::Result<Tensor> {
+        let xs = xs.apply(&self.ln_q)?.reshape(((), self.hidden_size))?;
+        let xs = xs.apply(&self.mlp_0)?.gelu()?.apply(&self.mlp_2)?;
+        Ok(xs)
+    }
+}
+
+#[derive(Debug, Clone)]
+struct Qwen2_5VLVisionMLP {
+    gate_proj: Linear,
+    up_proj: Linear,
+    down_proj: Linear,
+    act_fn: Activation,
+}
+
+impl Qwen2_5VLVisionMLP {
+    fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let hidden_sz = cfg.vision_config.hidden_size;
+        let intermediate_sz = cfg.vision_config.intermediate_size;
+        let gate_proj = linear(hidden_sz, intermediate_sz, vb.pp("gate_proj"))?;
+        let up_proj = linear(hidden_sz, intermediate_sz, vb.pp("up_proj"))?;
+        let down_proj = linear(intermediate_sz, hidden_sz, vb.pp("down_proj"))?;
+        Ok(Self {
+            gate_proj,
+            up_proj,
+            down_proj,
+            act_fn: cfg.hidden_act,
+        })
+    }
+}
+
+impl Module for Qwen2_5VLVisionMLP {
+    fn forward(&self, xs: &Tensor) -> candle_core::Result<Tensor> {
+        let lhs = xs.apply(&self.gate_proj)?.apply(&self.act_fn)?;
+        let rhs = xs.apply(&self.up_proj)?;
+        (lhs * rhs)?.apply(&self.down_proj)
+    }
+}
+
+#[derive(Debug, Clone)]
+struct Qwen2_5VLVisionAttention {
+    qkv: Linear,
+    proj: Linear,
+    num_heads: usize,
+    scale: Tensor,
+}
+
+impl Qwen2_5VLVisionAttention {
+    fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let hidden_size = cfg.vision_config.hidden_size;
+        let num_heads = cfg.vision_config.num_heads;
+        let head_dim = hidden_size / num_heads;
+        let qkv = linear(hidden_size, hidden_size * 3, vb.pp("qkv"))?;
+        let proj = linear(hidden_size, hidden_size, vb.pp("proj"))?;
+        let scale = Tensor::new(vec![1f32 / (head_dim as f32).sqrt()], vb.device())?
+            .to_dtype(vb.dtype())?;
+        Ok(Self {
+            qkv,
+            proj,
+            num_heads,
+            scale,
+        })
+    }
+
+    fn forward(
+        &self,
+        xs: &Tensor,
+        cos: &Tensor,
+        sin: &Tensor,
+        attention_mask: &Tensor,
+    ) -> Result<Tensor> {
+        // xs: (seq_len, hidden_size)
+        let seq_length = xs.dim(0)?;
+        // (seq_len, hidden_size) -> (seq_len, hidden_size*3)
+        // -> (seq_len, 3, num_heads, head_dim) -> (3, seq_len, num_heads, head_dim)
+        let qkv_states = xs
+            .apply(&self.qkv)?
+            .reshape((seq_length, 3, self.num_heads, ()))?
+            .permute((1, 0, 2, 3))?;
+        // (seq_len, num_heads, head_dim)
+        let query_states = qkv_states.i(0)?.contiguous()?;
+        let key_states = qkv_states.i(1)?.contiguous()?;
+        let value_states = qkv_states.i(2)?.contiguous()?;
+        let (query_states, key_states) =
+            apply_rotary_pos_emb_vision(&query_states, &key_states, cos, sin)?;
+        // (seq_len, num_heads, head_dim) -> (num_heads, seq_len, head_dim)
+        let query_states = query_states.transpose(0, 1)?.contiguous()?;
+        let key_states = key_states.transpose(0, 1)?.contiguous()?;
+        let value_states = value_states.transpose(0, 1)?.contiguous()?;
+
+        let attn_output = {
+            let attn_weights = query_states
+                .matmul(&key_states.transpose(D::Minus2, D::Minus1)?)?
+                .broadcast_mul(&self.scale)?;
+            let attn_weights = attn_weights.broadcast_add(&attention_mask)?;
+            let attn_weights = candle_nn::ops::softmax_last_dim(&attn_weights)?;
+            attn_weights.matmul(&value_states)?
+        };
+        // (num_heads, seq_len, head_dim) -> (seq_len, num_heads, head_dim) -> (seq_len, hidden_size)
+        let attn_output = attn_output
+            .transpose(0, 1)?
+            .reshape((seq_length, ()))?
+            .contiguous()?;
+        let attn_ouput = attn_output.apply(&self.proj)?;
+        Ok(attn_ouput)
+    }
+}
+
+#[derive(Debug, Clone)]
+struct Qwen2_5VLVisionBlock {
+    attn: Qwen2_5VLVisionAttention,
+    mlp: Qwen2_5VLVisionMLP,
+    norm1: RmsNorm,
+    norm2: RmsNorm,
+}
+
+impl Qwen2_5VLVisionBlock {
+    fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let attn = Qwen2_5VLVisionAttention::new(cfg, vb.pp("attn"))?;
+        let mlp = Qwen2_5VLVisionMLP::new(cfg, vb.pp("mlp"))?;
+        let norm1 = rms_norm(
+            cfg.vision_config.hidden_size,
+            cfg.rms_norm_eps,
+            vb.pp("norm1"),
+        )?;
+        let norm2 = rms_norm(
+            cfg.vision_config.hidden_size,
+            cfg.rms_norm_eps,
+            vb.pp("norm2"),
+        )?;
+
+        Ok(Self {
+            attn,
+            mlp,
+            norm1,
+            norm2,
+        })
+    }
+
+    fn forward(
+        &self,
+        xs: &Tensor,
+        cos: &Tensor,
+        sin: &Tensor,
+        attention_mask: &Tensor,
+    ) -> Result<Tensor> {
+        let residual = xs;
+        let xs = self.norm1.forward(xs)?;
+        let xs = self.attn.forward(&xs, cos, sin, attention_mask)?;
+        let xs = (xs + residual)?;
+        let residual = &xs;
+        let xs = xs.apply(&self.norm2)?.apply(&self.mlp)?;
+        let xs = (residual + xs)?;
+        Ok(xs)
+    }
+}
+
+pub struct Qwen2_5VLVisionModel {
+    spatial_merge_size: usize,
+    patch_size: usize,
+    fullatt_block_indexes: Vec<usize>,
+    window_size: usize,
+    spatial_merge_unit: usize,
+    patch_embed: Qwen2_5VisionPatchEmbed,
+    rotary_pos_emb: Qwen2_5VisionRotaryEmbedding,
+    blocks: Vec<Qwen2_5VLVisionBlock>,
+    merger: Qwen2_5VLPatchMerger,
+    dtype: DType,
+}
+
+impl Qwen2_5VLVisionModel {
+    pub fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let spatial_merge_size = cfg.vision_config.spatial_merge_size;
+        let patch_size = cfg.vision_config.patch_size;
+        let fullatt_block_indexes = cfg.vision_config.fullatt_block_indexes.clone();
+        let window_size = cfg.vision_config.window_size;
+        let spatial_merge_unit = spatial_merge_size * spatial_merge_size;
+        let head_dim = cfg.vision_config.hidden_size / cfg.vision_config.num_heads;
+        let patch_embed = Qwen2_5VisionPatchEmbed::new(cfg, vb.pp("patch_embed"))?;
+        let rotary_pos_emb = Qwen2_5VisionRotaryEmbedding::new(head_dim / 2, None);
+        let mut blocks = Vec::new();
+        let vb_blocks = vb.pp("blocks");
+        for i in 0..cfg.vision_config.depth {
+            let block = Qwen2_5VLVisionBlock::new(cfg, vb_blocks.pp(i))?;
+            blocks.push(block);
+        }
+        let merger = Qwen2_5VLPatchMerger::new(cfg, vb.pp("merger"))?;
+        let dtype = vb.dtype();
+        Ok(Self {
+            spatial_merge_size,
+            patch_size,
+            fullatt_block_indexes,
+            window_size,
+            spatial_merge_unit,
+            patch_embed,
+            rotary_pos_emb,
+            blocks,
+            merger,
+            dtype,
+        })
+    }
+
+    pub fn rot_pos_emb(&self, grid_thw: &Tensor) -> Result<Tensor> {
+        let mut pos_ids = Vec::new();
+        for i in 0..grid_thw.dim(0)? {
+            let [t, h, w] = grid_thw.i(i)?.to_vec1::<u32>()?[..] else {
+                return Err(anyhow!(format!("grid_thw Expected exactly 3 elements")));
+            };
+            // hpos_ids shape (h, w)
+            let hpos_ids = Tensor::arange(0, h, grid_thw.device())?
+                .unsqueeze(1)?
+                .expand((h as usize, w as usize))?;
+            let hpos_ids = hpos_ids.reshape((
+                h as usize / self.spatial_merge_size,
+                self.spatial_merge_size,
+                w as usize / self.spatial_merge_size,
+                self.spatial_merge_size,
+            ))?;
+            let hpos_ids = hpos_ids.permute((0, 2, 1, 3))?.flatten_all()?;
+            let wpos_ids = Tensor::arange(0, w, grid_thw.device())?
+                .unsqueeze(0)?
+                .expand((h as usize, w as usize))?;
+            let wpos_ids = wpos_ids.reshape((
+                h as usize / self.spatial_merge_size,
+                self.spatial_merge_size,
+                w as usize / self.spatial_merge_size,
+                self.spatial_merge_size,
+            ))?;
+            let wpos_ids = wpos_ids.permute((0, 2, 1, 3))?.flatten_all()?;
+            // thw_pos_ids shape (h*w, 2)
+            let thw_pos_ids =
+                Tensor::stack(&[&hpos_ids, &wpos_ids], D::Minus1)?.repeat((t as usize, 1))?;
+            pos_ids.push(thw_pos_ids);
+        }
+        let pos_ids = Tensor::cat(&pos_ids, 0)?.contiguous()?;
+        let max_grid_size = grid_thw.i((.., 1..))?.max_all()?.to_scalar::<u32>()?;
+        let rotary_pos_emb_full = self
+            .rotary_pos_emb
+            .forward(max_grid_size as usize, grid_thw.device())?;
+
+        // contiguous()一定要加！！！很重要！！！！，不然index_select出来的是错的
+        // 找错找了半天，都是泪啊，做维度索引操作后contiguous顺手写上总没错
+        // 第一列是h维度的索引
+        let pos_ids_h = pos_ids.i((.., 0))?.contiguous()?;
+        // 第二列是w维度的索引
+        let pos_ids_w = pos_ids.i((.., 1))?.contiguous()?;
+        let rotary_pos_emb_h = rotary_pos_emb_full.index_select(&pos_ids_h, 0)?;
+        let rotary_pos_emb_w = rotary_pos_emb_full.index_select(&pos_ids_w, 0)?;
+        // 每个patch融合h索引和w索引两个的位置编码信息
+        let rotary_pos_emb = Tensor::cat(&[rotary_pos_emb_h, rotary_pos_emb_w], 1)?.contiguous()?;
+        Ok(rotary_pos_emb)
+    }
+
+    pub fn get_window_index(&self, grid_thw: &Tensor) -> Result<(Tensor, Tensor)> {
+        let mut window_index = Vec::new();
+        let mut cu_window_seqlens = vec![0];
+        let mut window_index_id = 0_i64;
+
+        let vit_merger_window_size =
+            (self.window_size / self.spatial_merge_size / self.patch_size) as u32;
+        for i in 0..grid_thw.dim(0)? {
+            let [grid_t, grid_h, grid_w] = grid_thw.i(i)?.to_vec1::<u32>()?[..] else {
+                return Err(anyhow!(format!("grid_thw Expected exactly 3 elements")));
+            };
+            let llm_grid_h = grid_h / self.spatial_merge_size as u32;
+            let llm_grid_w = grid_w / self.spatial_merge_size as u32;
+            // 因为后续需要使用-100来做填充，所以需要int类型
+            // candle好像不支持i32， DType里面都没有定义i32, 所以这里使用i64
+            let mut index = Tensor::arange(
+                window_index_id,
+                window_index_id + (grid_t * llm_grid_h * llm_grid_w) as i64,
+                grid_thw.device(),
+            )?
+            .reshape((grid_t as usize, llm_grid_h as usize, llm_grid_w as usize))?
+            .contiguous()?;
+            // python transformers 中实现如下
+            // let pad_h = (vit_merger_window_size - llm_grid_h % vit_merger_window_size);
+            // 后面加上 % vit_merger_window_size，保证llm_grid_h能整除vit_merger_window_size时不需要pad
+            // 按理说能整除应该是不需要pad的，transformers中这样实现不知道是不是有什么其他原因
+            let pad_h = (vit_merger_window_size - llm_grid_h % vit_merger_window_size)
+                % vit_merger_window_size;
+            let pad_w = (vit_merger_window_size - llm_grid_w % vit_merger_window_size)
+                % vit_merger_window_size;
+            let num_window_h = (llm_grid_h + pad_h) / vit_merger_window_size;
+            let num_window_w = (llm_grid_w + pad_w) / vit_merger_window_size;
+            if pad_h > 0 {
+                let pad_h_t = Tensor::new(vec![-100_i64], grid_thw.device())?
+                    .broadcast_as((grid_t as usize, pad_h as usize, llm_grid_w as usize))?
+                    .contiguous()?;
+                index = Tensor::cat(&[&index, &pad_h_t], 1)?;
+            }
+            if pad_w > 0 {
+                let pad_w_t = Tensor::new(vec![-100_i64], grid_thw.device())?
+                    .broadcast_as((
+                        grid_t as usize,
+                        (llm_grid_h + pad_h) as usize,
+                        pad_w as usize,
+                    ))?
+                    .contiguous()?;
+                index = Tensor::cat(&[&index, &pad_w_t], 2)?;
+            }
+            let index_padded = index
+                .reshape((
+                    grid_t as usize,
+                    num_window_h as usize,
+                    vit_merger_window_size as usize,
+                    num_window_w as usize,
+                    vit_merger_window_size as usize,
+                ))?
+                .permute((0, 1, 3, 2, 4))?;
+            let index_padded = index_padded
+                .reshape((
+                    grid_t as usize,
+                    (num_window_h * num_window_w) as usize,
+                    vit_merger_window_size as usize,
+                    vit_merger_window_size as usize,
+                ))?
+                .contiguous()?;
+            let is_pad = Tensor::new(vec![-100_i64], grid_thw.device())?;
+            let seqlens = index_padded
+                .broadcast_ne(&is_pad)?
+                .sum((2, 3))?
+                .flatten_all()?;
+            let index_padded = index_padded.flatten_all()?;
+            let not_pad = index_padded.broadcast_ne(&is_pad)?.to_vec1::<u8>()?;
+            let indices: Vec<u32> = not_pad
+                .iter()
+                .enumerate()
+                .filter_map(|(idx, &val)| if val != 0 { Some(idx as u32) } else { None })
+                .collect();
+            let indices_tensor = Tensor::from_slice(&indices, indices.len(), grid_thw.device())?;
+            let index_new = index_padded.gather(&indices_tensor, 0)?;
+            window_index.push(index_new);
+
+            let seq_len_last = cu_window_seqlens[cu_window_seqlens.len() - 1];
+            // cumsum方法i64类型执行会报错，先转成F64计算后再转回i64
+            let cu_seqlens_tmp = seqlens
+                .to_dtype(candle_core::DType::F64)?
+                .cumsum(0)?
+                .to_dtype(candle_core::DType::I64)?
+                .broadcast_mul(&Tensor::new(
+                    vec![self.spatial_merge_unit as i64],
+                    grid_thw.device(),
+                )?)?
+                .broadcast_add(&Tensor::new(vec![seq_len_last], grid_thw.device())?)?;
+            cu_window_seqlens.extend_from_slice(&cu_seqlens_tmp.to_vec1::<i64>()?);
+            window_index_id += (grid_t * llm_grid_h * llm_grid_w) as i64;
+        }
+        let window_index_tensor = Tensor::cat(&window_index, 0)?;
+        let cu_window_seqlens_tensor = Tensor::from_slice(
+            &cu_window_seqlens,
+            cu_window_seqlens.len(),
+            grid_thw.device(),
+        )?
+        .to_dtype(candle_core::DType::U32)?;
+        Ok((window_index_tensor, cu_window_seqlens_tensor))
+    }
+
+    pub fn get_attention_mask(
+        &self,
+        cu_seqlens: &Tensor,
+        seq_len: usize,
+        device: &Device,
+        dtype: DType,
+    ) -> Result<Tensor> {
+        let mut attention_mask =
+            Tensor::new(f32::NEG_INFINITY, device)?.broadcast_as((1, seq_len, seq_len))?;
+        for i in 1..cu_seqlens.dim(0)? {
+            let start = cu_seqlens.i(i - 1)?.to_scalar::<u32>()? as usize;
+            let end = cu_seqlens.i(i)?.to_scalar::<u32>()? as usize;
+            let block_size = end - start;
+            let zeros =
+                Tensor::zeros((1, block_size, block_size), candle_core::DType::F32, device)?;
+            attention_mask =
+                attention_mask.slice_assign(&[(0..1), (start..end), (start..end)], &zeros)?;
+        }
+        let attention_mask = attention_mask.to_dtype(dtype)?.contiguous()?;
+        Ok(attention_mask)
+    }
+
+    pub fn forward(&self, hidden_states: &Tensor, grid_thw: &Tensor) -> Result<Tensor> {
+        // hidden_states: (seq_len, hidden_size)
+        // grid_thw: (num_images_or_videos, 3), temporal, height, width
+        let hidden_states = hidden_states.to_dtype(self.dtype)?;
+        let hidden_states = self.patch_embed.forward(&hidden_states)?;
+        let rotary_pos_emb = self.rot_pos_emb(grid_thw)?;
+        let (window_index, cu_window_seqlens) = self.get_window_index(grid_thw)?;
+        let seq_len = hidden_states.dim(0)?;
+        let hidden_states = hidden_states
+            .reshape((
+                seq_len / self.spatial_merge_unit,
+                self.spatial_merge_unit,
+                (),
+            ))?
+            .contiguous()?;
+        let hidden_states = hidden_states.index_select(&window_index, 0)?;
+        let mut hidden_states = hidden_states.reshape((seq_len, ()))?;
+        let rotary_pos_emb = rotary_pos_emb.reshape((
+            seq_len / self.spatial_merge_unit,
+            self.spatial_merge_unit,
+            (),
+        ))?;
+        let rotary_pos_emb = rotary_pos_emb.index_select(&window_index, 0)?;
+        let rotary_pos_emb = rotary_pos_emb.reshape((seq_len, ()))?;
+        let emb = Tensor::cat(&[&rotary_pos_emb, &rotary_pos_emb], D::Minus1)?;
+        let cos = emb.cos()?.to_dtype(hidden_states.dtype())?;
+        let sin = emb.sin()?.to_dtype(hidden_states.dtype())?;
+        let cu_seqlens = grid_thw.i((.., 1))?.mul(&grid_thw.i((.., 2))?)?;
+        let grid_t = grid_thw.i((.., 0))?.to_vec1::<u32>()?;
+        let cu_seqlens_full = match cu_seqlens.rank() {
+            1 => cu_seqlens.repeat(grid_t[0] as usize)?,
+            2 => {
+                let mut cu_seqlens_repeat = Vec::new();
+                for (index, t) in grid_t.iter().enumerate() {
+                    cu_seqlens_repeat.push(cu_seqlens.i(index)?.repeat(t.clone() as usize)?);
+                }
+                Tensor::cat(&cu_seqlens_repeat, 0)?.flatten_all()?
+            }
+            _ => {
+                return Err(anyhow!(format!("create cu_seqlens error")));
+            }
+        };
+        let cu_seqlens = cu_seqlens_full
+            .to_dtype(DType::F64)?
+            .cumsum(0)?
+            .to_dtype(DType::U32)?;
+        let pad_zero = Tensor::from_vec(vec![0_u32], 1, hidden_states.device())?;
+        let cu_seqlens = Tensor::cat(&[&pad_zero, &cu_seqlens], D::Minus1)?;
+        let attention_mask_window = self.get_attention_mask(
+            &cu_window_seqlens,
+            seq_len,
+            hidden_states.device(),
+            hidden_states.dtype(),
+        )?;
+        let attention_mask_full = self.get_attention_mask(
+            &cu_seqlens,
+            seq_len,
+            hidden_states.device(),
+            hidden_states.dtype(),
+        )?;
+        let mut attention_mask = attention_mask_window.clone();
+        for (layer_num, block) in self.blocks.iter().enumerate() {
+            if self.fullatt_block_indexes.contains(&layer_num) {
+                attention_mask = attention_mask_full.clone();
+            } else {
+                attention_mask = attention_mask_window.clone();
+            }
+            hidden_states = block.forward(&hidden_states, &cos, &sin, &attention_mask)?;
+        }
+        let hidden_states = self.merger.forward(&hidden_states)?;
+        let reverse_indices = safe_arg_sort_last_dim(&window_index, true)?;
+        let hidden_states = hidden_states.index_select(&reverse_indices, 0)?;
+        Ok(hidden_states)
+    }
+}
+
+pub fn repeat_kv(xs: Tensor, n_rep: usize) -> Result<Tensor> {
+    if n_rep == 1 {
+        Ok(xs)
+    } else {
+        let (b_sz, n_kv_head, seq_len, head_dim) = xs.dims4()?;
+        // Using cat is faster than a broadcast as it avoids going through a potentially
+        // strided copy.
+        // https://github.com/huggingface/candle/pull/2043
+        let kv = Tensor::cat(&vec![&xs; n_rep], 2)?.reshape((
+            b_sz,
+            n_kv_head * n_rep,
+            seq_len,
+            head_dim,
+        ))?;
+        Ok(kv)
+    }
+}
+
+#[derive(Debug, Clone)]
+struct Qwen2_5VLTextMLP {
+    gate_proj: Linear,
+    up_proj: Linear,
+    down_proj: Linear,
+    act_fn: Activation,
+}
+
+impl Qwen2_5VLTextMLP {
+    fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let hidden_sz = cfg.hidden_size;
+        let intermediate_sz = cfg.intermediate_size;
+        let gate_proj = linear_no_bias(hidden_sz, intermediate_sz, vb.pp("gate_proj"))?;
+        let up_proj = linear_no_bias(hidden_sz, intermediate_sz, vb.pp("up_proj"))?;
+        let down_proj = linear_no_bias(intermediate_sz, hidden_sz, vb.pp("down_proj"))?;
+
+        Ok(Self {
+            gate_proj,
+            up_proj,
+            down_proj,
+            act_fn: cfg.hidden_act,
+        })
+    }
+}
+
+impl Module for Qwen2_5VLTextMLP {
+    fn forward(&self, xs: &Tensor) -> candle_core::Result<Tensor> {
+        let lhs = xs.apply(&self.gate_proj)?.apply(&self.act_fn)?;
+        let rhs = xs.apply(&self.up_proj)?;
+        (lhs * rhs)?.apply(&self.down_proj)
+    }
+}
+
+#[derive(Debug, Clone)]
+struct Qwen2_5VLTextAttention {
+    q_proj: Linear,
+    k_proj: Linear,
+    v_proj: Linear,
+    o_proj: Linear,
+    num_heads: usize,
+    num_kv_heads: usize,
+    num_kv_groups: usize,
+    head_dim: usize,
+    hidden_size: usize,
+    kv_cache: Option<(Tensor, Tensor)>,
+}
+
+impl Qwen2_5VLTextAttention {
+    fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let hidden_size = cfg.hidden_size;
+        let num_heads = cfg.num_attention_heads;
+        let num_kv_heads = cfg.num_key_value_heads;
+        let num_kv_groups = num_heads / num_kv_heads;
+        let head_dim = hidden_size / num_heads;
+        let q_proj = linear(hidden_size, num_heads * head_dim, vb.pp("q_proj"))?;
+        let k_proj = linear(hidden_size, num_kv_heads * head_dim, vb.pp("k_proj"))?;
+        let v_proj = linear(hidden_size, num_kv_heads * head_dim, vb.pp("v_proj"))?;
+        let o_proj = linear_no_bias(hidden_size, hidden_size, vb.pp("o_proj"))?;
+        Ok(Self {
+            q_proj,
+            k_proj,
+            v_proj,
+            o_proj,
+            num_heads,
+            num_kv_heads,
+            num_kv_groups,
+            head_dim,
+            hidden_size,
+            kv_cache: None,
+        })
+    }
+
+    fn forward(
+        &mut self,
+        xs: &Tensor,
+        cos: &Tensor,
+        sin: &Tensor,
+        attention_mask: Option<&Tensor>,
+    ) -> Result<Tensor> {
+        let (b_sz, q_len, _) = xs.dims3()?;
+        let query_states = self.q_proj.forward(xs)?;
+        let key_states = self.k_proj.forward(xs)?;
+        let value_states = self.v_proj.forward(xs)?;
+        let query_states = query_states
+            .reshape((b_sz, q_len, self.num_heads, self.head_dim))?
+            .transpose(1, 2)?;
+        let key_states = key_states
+            .reshape((b_sz, q_len, self.num_kv_heads, self.head_dim))?
+            .transpose(1, 2)?;
+        let value_states = value_states
+            .reshape((b_sz, q_len, self.num_kv_heads, self.head_dim))?
+            .transpose(1, 2)?;
+        let (query_states, key_states) =
+            apply_rotary_pos_emb(&query_states, &key_states, cos, sin)?;
+        let (key_states, value_states) = match &self.kv_cache {
+            None => (key_states, value_states),
+            Some((prev_k, prev_v)) => {
+                let key_states = Tensor::cat(&[prev_k, &key_states], 2)?;
+                let value_states = Tensor::cat(&[prev_v, &value_states], 2)?;
+                (key_states, value_states)
+            }
+        };
+
+        self.kv_cache = Some((key_states.clone(), value_states.clone()));
+
+        let key_states = repeat_kv(key_states, self.num_kv_groups)?.contiguous()?;
+        let value_states = repeat_kv(value_states, self.num_kv_groups)?.contiguous()?;
+        let query_states = query_states.contiguous()?;
+        let attn_output = {
+            let scale = 1f64 / f64::sqrt(self.head_dim as f64);
+            #[cfg(not(feature = "flash-attn"))]
+            {
+                let attn_weights =
+                    query_states.matmul(&key_states.transpose(D::Minus2, D::Minus1)?)?;
+                let attn_weights = (attn_weights * scale)?;
+                let attn_weights = match attention_mask {
+                    None => attn_weights,
+                    Some(mask) => attn_weights.broadcast_add(mask)?,
+                };
+                let attn_weights = candle_nn::ops::softmax_last_dim(&attn_weights)?;
+                let attn_weights = attn_weights.matmul(&value_states)?;
+                attn_weights
+            }
+            #[cfg(feature = "flash-attn")]
+            {
+                // use flash-attn,
+                // flash-attn shape: (bs, seq_len, num_head, head_dim)
+                let query_states = query_states.transpose(1, 2)?;
+                let key_states = key_states.transpose(1, 2)?;
+                let value_states = value_states.transpose(1, 2)?;
+                let attn_output = candle_flash_attn::flash_attn(
+                    &query_states,
+                    &key_states,
+                    &value_states,
+                    scale as f32,
+                    attention_mask.is_some(),
+                )?
+                .transpose(1, 2)?;
+                attn_output
+            }
+        };
+        let attn_output =
+            attn_output
+                .transpose(1, 2)?
+                .contiguous()?
+                .reshape((b_sz, q_len, self.hidden_size))?;
+        let attn_output = attn_output.apply(&self.o_proj)?;
+        Ok(attn_output)
+    }
+
+    fn clear_kv_cache(&mut self) {
+        self.kv_cache = None
+    }
+}
+
+#[derive(Debug, Clone)]
+struct Qwen2_5VLTextDecoderLayer {
+    self_attn: Qwen2_5VLTextAttention,
+    mlp: Qwen2_5VLTextMLP,
+    input_layernorm: RmsNorm,
+    post_attention_layernorm: RmsNorm,
+}
+
+impl Qwen2_5VLTextDecoderLayer {
+    fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let self_attn = Qwen2_5VLTextAttention::new(cfg, vb.pp("self_attn"))?;
+        let mlp = Qwen2_5VLTextMLP::new(cfg, vb.pp("mlp"))?;
+        let input_layernorm =
+            rms_norm(cfg.hidden_size, cfg.rms_norm_eps, vb.pp("input_layernorm"))?;
+        let post_attention_layernorm = rms_norm(
+            cfg.hidden_size,
+            cfg.rms_norm_eps,
+            vb.pp("post_attention_layernorm"),
+        )?;
+        Ok(Self {
+            self_attn,
+            mlp,
+            input_layernorm,
+            post_attention_layernorm,
+        })
+    }
+
+    fn forward(
+        &mut self,
+        xs: &Tensor,
+        cos: &Tensor,
+        sin: &Tensor,
+        attention_mask: Option<&Tensor>,
+    ) -> Result<Tensor> {
+        let residual = xs;
+        let xs = self.input_layernorm.forward(xs)?;
+        let xs = self.self_attn.forward(&xs, cos, sin, attention_mask)?;
+        let xs = (xs + residual)?;
+        let residual = &xs;
+        let xs = xs.apply(&self.post_attention_layernorm)?.apply(&self.mlp)?;
+        let xs = (residual + xs)?;
+        Ok(xs)
+    }
+
+    fn clear_kv_cache(&mut self) {
+        self.self_attn.clear_kv_cache()
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct Qwen2_5VLTextModel {
+    pub embed_tokens: candle_nn::Embedding,
+    layers: Vec<Qwen2_5VLTextDecoderLayer>,
+    norm: RmsNorm,
+    rotary_emb: Qwen2_5VLTextRotaryEmbedding,
+    dtype: DType,
+    sliding_window: usize,
+    device: Device,
+    rope_scaling: RopeScaling,
+}
+
+impl Qwen2_5VLTextModel {
+    pub fn new(cfg: &Config, vb: VarBuilder) -> Result<Self> {
+        let embed_tokens =
+            candle_nn::embedding(cfg.vocab_size, cfg.hidden_size, vb.pp("embed_tokens"))?;
+        let head_dim = cfg.hidden_size / cfg.num_attention_heads;
+        let rotary_emb = Qwen2_5VLTextRotaryEmbedding::new(head_dim, cfg.rope_theta);
+        let mut layers = Vec::with_capacity(cfg.num_hidden_layers);
+        let vb_l = vb.pp("layers");
+        for layer_idx in 0..cfg.num_hidden_layers {
+            let layer = Qwen2_5VLTextDecoderLayer::new(cfg, vb_l.pp(layer_idx))?;
+            layers.push(layer)
+        }
+        let norm = rms_norm(cfg.hidden_size, cfg.rms_norm_eps, vb.pp("norm"))?;
+        let sliding_window = cfg.sliding_window;
+        let rope_scaling = cfg.rope_scaling.clone();
+        Ok(Self {
+            embed_tokens,
+            layers,
+            norm,
+            rotary_emb,
+            dtype: vb.dtype(),
+            sliding_window,
+            device: vb.device().clone(),
+            rope_scaling,
+        })
+    }
+
+    fn prepare_causal_attention_mask(
+        &self,
+        b_size: usize,
+        tgt_len: usize,
+        seqlen_offset: usize,
+    ) -> Result<Tensor> {
+        // Sliding window mask?
+        let mask: Vec<_> = (0..tgt_len)
+            .flat_map(|i| {
+                (0..tgt_len).map(move |j| {
+                    if i < j || j + self.sliding_window < i {
+                        f32::NEG_INFINITY
+                    } else {
+                        0.
+                    }
+                })
+            })
+            .collect();
+        let mask = Tensor::from_slice(&mask, (tgt_len, tgt_len), &self.device)?;
+        let mask = if seqlen_offset > 0 {
+            let mask0 = Tensor::zeros((tgt_len, seqlen_offset), self.dtype, &self.device)?;
+            Tensor::cat(&[&mask0, &mask], D::Minus1)?
+        } else {
+            mask
+        };
+        let mask = mask
+            .expand((b_size, 1, tgt_len, tgt_len + seqlen_offset))?
+            .to_dtype(self.dtype)?;
+        Ok(mask)
+    }
+    pub fn forward(
+        &mut self,
+        inputs_embeds: &Tensor,
+        seqlen_offset: usize,
+        position_ids: Option<&Tensor>,
+    ) -> Result<Tensor> {
+        let (b_size, seq_len, _) = inputs_embeds.dims3()?;
+
+        let position_ids = match position_ids {
+            Some(ids) => ids.clone(),
+            None => Tensor::arange(
+                seqlen_offset as u32,
+                (seq_len + seqlen_offset) as u32,
+                inputs_embeds.device(),
+            )?
+            .unsqueeze(0)?
+            .unsqueeze(0)?
+            .broadcast_as((3, b_size, seq_len))?,
+        };
+        let (cos, sin) = self.rotary_emb.forward(
+            &position_ids,
+            self.dtype,
+            self.rope_scaling.mrope_section.clone(),
+        )?;
+        let mut xs = inputs_embeds.clone();
+        let attention_mask: Option<&Tensor> = {
+            if seq_len <= 1 {
+                None
+            } else {
+                Some(&self.prepare_causal_attention_mask(b_size, seq_len, seqlen_offset)?)
+            }
+        };
+        for layer in self.layers.iter_mut() {
+            xs = layer.forward(&xs, &cos, &sin, attention_mask)?;
+        }
+        let xs = xs.apply(&self.norm)?;
+        Ok(xs)
+    }
+
+    pub fn clear_kv_cache(&mut self) {
+        for layer in self.layers.iter_mut() {
+            layer.clear_kv_cache()
+        }
+    }
+}
+
+pub struct Qwen2_5VLModel {
+    visual: Qwen2_5VLVisionModel,
+    model: Qwen2_5VLTextModel,
+    pub cfg: Config,
+    lm_head: Linear,
+    rope_deltas: Option<Tensor>,
+}
+
+impl Qwen2_5VLModel {
+    pub fn new(cfg: Config, vb: VarBuilder) -> Result<Self> {
+        let visual = Qwen2_5VLVisionModel::new(&cfg, vb.pp("visual"))?;
+        let model = Qwen2_5VLTextModel::new(&cfg, vb.pp("model"))?;
+        let vocab_size = cfg.vocab_size;
+        let lm_head = if cfg.tie_word_embeddings {
+            Linear::new(model.embed_tokens.embeddings().clone(), None)
+        } else {
+            linear_no_bias(cfg.hidden_size, vocab_size, vb.pp("lm_head"))?
+        };
+
+        Ok(Self {
+            visual,
+            model,
+            cfg,
+            lm_head,
+            rope_deltas: None,
+        })
+    }
+
+    pub fn get_rope_index(
+        &self,
+        input_ids: &Tensor,
+        image_grid_thw: Option<&Tensor>,
+        video_grid_thw: Option<&Tensor>,
+        mask: Option<&Tensor>,
+        second_per_grid_ts: Option<Vec<f32>>,
+    ) -> Result<(Tensor, Tensor)> {
+        let spatial_merge_size = self.cfg.vision_config.spatial_merge_size;
+        let mut mrope_position_deltas: Vec<i64> = Vec::new();
+        if image_grid_thw.is_some() || video_grid_thw.is_some() {
+            let total_input_ids = input_ids.clone();
+            let mut mask_;
+            if mask.is_none() {
+                mask_ = Tensor::ones_like(&total_input_ids)?;
+            } else {
+                mask_ = mask.unwrap().clone();
+            }
+            let mut position_ids = Tensor::ones(
+                (3, input_ids.dim(0)?, input_ids.dim(1)?),
+                input_ids.dtype(),
+                input_ids.device(),
+            )?;
+            let mut image_index = 0;
+            let mut video_index = 0;
+            for i in 0..total_input_ids.dim(0)? {
+                let mut input_ids_i = total_input_ids.i(i)?;
+                let mask_i = mask_.i(i)?;
+                // 推理时, attention_mask如果是全1向量,取非0索引的操作没必要
+                if mask_i.sum_all()?.to_scalar::<u32>()? != mask_i.dim(0)? as u32 {
+                    let nonzero_idx = nonzero_index(&mask_i)?;
+                    input_ids_i = input_ids_i.gather(&nonzero_idx, 0)?;
+                }
+                let mut text_start = 0;
+                let mut text_end = 0;
+                let mut thw = vec![];
+                let mut second_per_grid_t = 0_f32;
+                let mut llm_pos_ids_list: Vec<Tensor> = Vec::new();
+                // vision start的下一个索引
+                let vision_indices =
+                    get_vision_next_indices(&input_ids_i, self.cfg.vision_start_token_id as u32);
+                match vision_indices {
+                    Ok(indeices) => {
+                        let vision_tokens = input_ids_i.gather(&indeices, 0)?.to_vec1::<u32>()?;
+                        let vision_indices_vec = indeices.to_vec1::<u32>()?;
+                        for (j, &token) in vision_tokens.iter().enumerate() {
+                            if token == self.cfg.image_token_id as u32 {
+                                thw = image_grid_thw.unwrap().i(image_index)?.to_vec1::<u32>()?;
+                                image_index += 1;
+                                text_end = vision_indices_vec[j];
+                                second_per_grid_t = 0.0;
+                            }
+                            if token == self.cfg.video_token_id as u32 {
+                                thw = video_grid_thw.unwrap().i(video_index)?.to_vec1::<u32>()?;
+                                text_end = vision_indices_vec[j];
+                                second_per_grid_t = match second_per_grid_ts {
+                                    None => 1.0,
+                                    Some(ref vec) => vec[video_index],
+                                };
+                                video_index += 1;
+                            }
+                            let llm_grid_t = thw[0];
+                            let llm_grid_h = thw[1] / spatial_merge_size as u32;
+                            let llm_grid_w = thw[2] / spatial_merge_size as u32;
+                            let text_len = text_end - text_start;
+                            let start_idx = if llm_pos_ids_list.len() > 0 {
+                                llm_pos_ids_list[llm_pos_ids_list.len() - 1]
+                                    .max_all()?
+                                    .to_scalar::<u32>()?
+                                    + 1
+                            } else {
+                                0
+                            };
+                            let pos_ids = Tensor::arange(
+                                start_idx,
+                                start_idx + text_len,
+                                input_ids_i.device(),
+                            )?
+                            .unsqueeze(0)?
+                            .broadcast_as((3usize, text_len as usize))?;
+                            llm_pos_ids_list.push(pos_ids);
+                            let range_tensor = Tensor::arange(0, llm_grid_t, input_ids_i.device())?
+                                .unsqueeze(D::Minus1)?;
+                            let expanded_range = range_tensor.broadcast_as((
+                                llm_grid_t as usize,
+                                (llm_grid_h * llm_grid_w) as usize,
+                            ))?;
+                            let time_tensor = expanded_range
+                                .broadcast_mul(&Tensor::new(
+                                    vec![
+                                        (second_per_grid_t
+                                            * self.cfg.vision_config.tokens_per_second as f32)
+                                            as u32,
+                                    ],
+                                    input_ids_i.device(),
+                                )?)?
+                                .broadcast_add(&Tensor::new(
+                                    vec![start_idx + text_len],
+                                    input_ids_i.device(),
+                                )?)?;
+                            let t_index = time_tensor.flatten_all()?;
+                            let h_index = Tensor::arange(
+                                start_idx + text_len,
+                                start_idx + text_len + llm_grid_h,
+                                input_ids_i.device(),
+                            )?
+                            .unsqueeze(0)?
+                            .unsqueeze(D::Minus1)?
+                            .broadcast_as((
+                                llm_grid_t as usize,
+                                llm_grid_h as usize,
+                                llm_grid_w as usize,
+                            ))?
+                            .flatten_all()?;
+                            let w_index = Tensor::arange(
+                                start_idx + text_len,
+                                start_idx + text_len + llm_grid_w,
+                                input_ids_i.device(),
+                            )?
+                            .unsqueeze(0)?
+                            .unsqueeze(0)?
+                            .broadcast_as((
+                                llm_grid_t as usize,
+                                llm_grid_h as usize,
+                                llm_grid_w as usize,
+                            ))?
+                            .flatten_all()?;
+
+                            let thw_index = Tensor::stack(&[t_index, h_index, w_index], 0)?;
+                            llm_pos_ids_list.push(thw_index);
+                            text_start = text_end + llm_grid_t * llm_grid_h * llm_grid_w;
+                        }
+                    }
+                    Err(e) => {
+                        println!("get vision_indices err: {}", e);
+                    }
+                };
+
+                if text_start < input_ids_i.dim(0)? as u32 {
+                    let start_idx = if llm_pos_ids_list.len() > 0 {
+                        llm_pos_ids_list[llm_pos_ids_list.len() - 1]
+                            .max_all()?
+                            .to_scalar::<u32>()?
+                            + 1
+                    } else {
+                        0
+                    };
+                    let text_len = input_ids_i.dim(0)? as u32 - text_start;
+                    let pos_ids =
+                        Tensor::arange(start_idx, start_idx + text_len, input_ids_i.device())?
+                            .unsqueeze(0)?
+                            .broadcast_as((3usize, text_len as usize))?;
+                    llm_pos_ids_list.push(pos_ids);
+                }
+                let llm_position = Tensor::cat(&llm_pos_ids_list, 1)?.reshape((3, 1, ()))?;
+                position_ids = position_ids
+                    .slice_assign(&[(0..3), (i..i + 1), (0..input_ids.dim(1)?)], &llm_position)?;
+                let position_deltas = llm_position.max_all()?.to_scalar::<u32>()? as i64 + 1
+                    - input_ids_i.dim(0)? as i64;
+                mrope_position_deltas.push(position_deltas);
+            }
+
+            let mut mrope_position_deltas = Tensor::new(mrope_position_deltas, input_ids.device())?;
+            if mrope_position_deltas.rank() == 1 {
+                mrope_position_deltas = mrope_position_deltas.unsqueeze(0)?;
+            }
+            return Ok((position_ids.contiguous()?, mrope_position_deltas));
+        } else {
+            if mask.is_some() {
+                let mut position_ids = mask
+                    .unwrap()
+                    .to_dtype(candle_core::DType::F64)?
+                    .cumsum(D::Minus1)?
+                    .to_dtype(candle_core::DType::U32)?
+                    .broadcast_sub(&Tensor::new(vec![1_u32], input_ids.device())?)?;
+                for i in 0..position_ids.dim(0)? {
+                    let mut position_ids_i = position_ids.i(i)?;
+                    let mask_i = mask.unwrap().i(i)?;
+                    // 如果有pad, 将填充位置置为1
+                    // 当bs>1, 可能存在不同序列长度，需要添加pad使seq_len长度一致
+                    if mask_i.sum_all()?.to_scalar::<u32>()? != mask_i.dim(0)? as u32 {
+                        let zero_indices = zero_index(&mask_i)?;
+                        let replace_1 = Tensor::ones(
+                            zero_indices.dim(0)?,
+                            candle_core::DType::U32,
+                            input_ids.device(),
+                        )?;
+                        position_ids_i = position_ids_i
+                            .scatter(&zero_indices, &replace_1, 0)?
+                            .unsqueeze(0)?;
+                        position_ids = position_ids.slice_assign(
+                            &[(i..i + 1), (0..position_ids.dim(1)?)],
+                            &position_ids_i,
+                        )?;
+                    }
+                }
+                position_ids = position_ids
+                    .unsqueeze(0)?
+                    .broadcast_as((3, input_ids.dim(0)?, input_ids.dim(1)?))?
+                    .contiguous()?;
+                let mut mrope_position_deltas = position_ids
+                    .max(0)?
+                    .max(D::Minus1)?
+                    .broadcast_sub(&Tensor::new(
+                        vec![mask.unwrap().dim(D::Minus1)? as u32 - 1],
+                        input_ids.device(),
+                    )?)?
+                    .contiguous()?;
+                if mrope_position_deltas.rank() == 1 {
+                    mrope_position_deltas = mrope_position_deltas.unsqueeze(0)?;
+                }
+                return Ok((position_ids, mrope_position_deltas));
+            } else {
+                let position_ids =
+                    Tensor::arange(0_u32, input_ids.dim(D::Minus1)? as u32, input_ids.device())?
+                        .unsqueeze(0)?
+                        .unsqueeze(0)?
+                        .broadcast_as((3, input_ids.dim(0)?, input_ids.dim(D::Minus1)?))?
+                        .contiguous()?;
+                let mrope_position_deltas = Tensor::zeros(
+                    (input_ids.dim(0)?, 1),
+                    input_ids.dtype(),
+                    input_ids.device(),
+                )?;
+                Ok((position_ids, mrope_position_deltas))
+            }
+        }
+    }
+
+    pub fn forward(
+        &mut self,
+        input_ids: &Tensor,
+        pixel_values: Option<&Tensor>,
+        image_grid_thw: Option<&Tensor>,
+        pixel_values_video: Option<&Tensor>,
+        video_grid_thw: Option<&Tensor>,
+        mask: &Tensor,
+        cache_position: Option<&Tensor>,
+        seqlen_offset: usize,
+        second_per_grid_ts: Option<Vec<f32>>,
+    ) -> Result<Tensor> {
+        // input_ids shape: (bs, seq_len)
+        let mut inputs_embeds = self.model.embed_tokens.forward(&input_ids)?;
+        // inputs_embeds shape: (bs, seq_len, hidden_dim)
+        if pixel_values.is_some() && image_grid_thw.is_some() {
+            // image_embed shape: (seq_len, hidden_dim)
+            let image_embed = self
+                .visual
+                .forward(pixel_values.unwrap(), image_grid_thw.unwrap())?;
+            let vision_mask = get_equal_mask(&input_ids, self.cfg.image_token_id as u32)?;
+
+            let n_image_tokens = vision_mask.sum_all()?.to_scalar::<u32>()?;
+            if n_image_tokens as usize != image_embed.dim(0)? {
+                return Err(anyhow!(format!(
+                    "n_image_token num: {} not equal to image_embed len: {}",
+                    n_image_tokens,
+                    image_embed.dim(0)?
+                )));
+            }
+            inputs_embeds = masked_scatter_dim0(&inputs_embeds, &image_embed, &vision_mask)?;
+        }
+        if pixel_values_video.is_some() && video_grid_thw.is_some() {
+            let video_embed = self
+                .visual
+                .forward(pixel_values_video.unwrap(), video_grid_thw.unwrap())?;
+
+            let vision_mask = get_equal_mask(&input_ids, self.cfg.video_token_id as u32)?;
+            let n_video_tokens = vision_mask.sum_all()?.to_scalar::<u32>()?;
+            if n_video_tokens as usize != video_embed.dim(0)? {
+                return Err(anyhow!(format!(
+                    "n_image_token num: {} not equal to image_embed len: {}",
+                    n_video_tokens,
+                    video_embed.dim(0)?
+                )));
+            }
+            inputs_embeds = masked_scatter_dim0(&inputs_embeds, &video_embed, &vision_mask)?;
+        }
+        let mut position_ids;
+        let mut rope_deltas;
+        if (cache_position.is_some() && cache_position.unwrap().i(0)?.to_scalar::<u32>()? == 0)
+            || self.rope_deltas.is_none()
+        {
+            (position_ids, rope_deltas) = self.get_rope_index(
+                input_ids,
+                image_grid_thw,
+                video_grid_thw,
+                Some(mask),
+                second_per_grid_ts,
+            )?;
+            self.rope_deltas = Some(rope_deltas);
+        } else {
+            let (bs, seq_len, _) = inputs_embeds.dims3()?;
+            let delta = if cache_position.is_some() {
+                cache_position
+                    .unwrap()
+                    .i(0)?
+                    .to_dtype(self.rope_deltas.as_ref().unwrap().dtype())?
+                    .broadcast_add(&self.rope_deltas.as_ref().unwrap())?
+                    .contiguous()?
+                    .to_dtype(candle_core::DType::U32)?
+            } else {
+                Tensor::zeros(1, inputs_embeds.dtype(), inputs_embeds.device())?
+            };
+            position_ids = Tensor::arange(0u32, seq_len as u32, input_ids.device())?
+                .unsqueeze(0)?
+                .broadcast_as((bs, seq_len))?
+                .broadcast_add(&delta)?
+                .unsqueeze(0)?
+                .broadcast_as((3, bs, seq_len))?
+                .contiguous()?;
+        }
+        let outputs = self
+            .model
+            .forward(&inputs_embeds, seqlen_offset, Some(&position_ids))?;
+        let seq_len = outputs.dim(1)?;
+        let hidden_state = outputs.narrow(1, seq_len - 1, 1)?;
+        let logits = self.lm_head.forward(&hidden_state)?;
+        Ok(logits)
+    }
+    pub fn clear_kv_cache(&mut self) {
+        self.model.clear_kv_cache();
+    }
+}
diff --git a/src/models/qwen2_5vl/processor.rs b/src/models/qwen2_5vl/processor.rs
new file mode 100644
index 0000000..eb173d1
--- /dev/null
+++ b/src/models/qwen2_5vl/processor.rs
@@ -0,0 +1,462 @@
+use std::collections::HashMap;
+
+use anyhow::{Result, anyhow};
+use candle_core::{DType, Device, IndexOp, Shape, Tensor};
+use ffmpeg_next as ffmpeg;
+use image::DynamicImage;
+use num::integer::lcm;
+use openai_dive::v1::resources::chat::{
+    ChatCompletionParameters, ChatMessage, ChatMessageContent, ChatMessageContentPart,
+};
+
+use crate::{
+    models::qwen2_5vl::config::VisionSetting,
+    utils::{
+        img_utils::get_image,
+        utils::{ceil_by_factor, floor_by_factor, round_by_factor},
+    },
+};
+
+#[derive(Clone)]
+pub struct VisionInput {
+    pub data: Tensor,
+    pub grid_thw: Tensor,
+}
+
+#[derive(Clone)]
+pub struct GeneralInput {
+    pub replace_text: String,
+    pub pixel_values: Option<Tensor>,
+    pub image_grid_thw: Option<Tensor>,
+    pub pixel_values_video: Option<Tensor>,
+    pub video_grid_thw: Option<Tensor>,
+    pub second_per_grid_ts: Option<Vec<f32>>,
+}
+
+pub struct Qwen2_5VLProcessor {
+    vision_setting: VisionSetting,
+    device: Device,
+    dtype: DType,
+    image_token: String,
+    video_token: String,
+}
+
+impl Qwen2_5VLProcessor {
+    pub fn new(device: &Device, dtype: DType) -> Result<Self> {
+        let vision_setting = VisionSetting::default();
+        let image_token = "<|image_pad|>".to_string();
+        let video_token = "<|video_pad|>".to_string();
+        Ok(Self {
+            vision_setting,
+            device: device.clone(),
+            dtype,
+            image_token,
+            video_token,
+        })
+    }
+
+    pub fn extract_vision_info(
+        &self,
+        mes: &ChatCompletionParameters,
+    ) -> Result<HashMap<String, Vec<String>>> {
+        let mut vision_map = HashMap::new();
+        vision_map.insert("image".to_string(), Vec::new());
+        vision_map.insert("video".to_string(), Vec::new());
+        for chat_mes in mes.messages.clone() {
+            match chat_mes {
+                ChatMessage::User { content, name } => match content {
+                    ChatMessageContent::ContentPart(part_vec) => {
+                        for part in part_vec {
+                            match part {
+                                ChatMessageContentPart::Image(img_part) => {
+                                    let img_url = img_part.image_url;
+                                    vision_map.get_mut("image").unwrap().push(img_url.url);
+                                }
+                                _ => {}
+                            }
+                        }
+                    }
+                    _ => {}
+                },
+                _ => {}
+            }
+        }
+        Ok(vision_map)
+    }
+
+    pub fn process_img(
+        &self,
+        img: &DynamicImage,
+        img_mean: &Tensor,
+        img_std: &Tensor,
+    ) -> Result<Tensor> {
+        let img_h = img.height();
+        let img_w = img.width();
+        //  h,w resize成 28的倍数
+        let (resize_h, resize_w) = smart_resize(img_h, img_w, &self.vision_setting, true, None)?;
+        let img = img.resize_exact(resize_w, resize_h, image::imageops::FilterType::CatmullRom);
+        let img_vec = img.to_rgb8().into_raw();
+        // (h, w, c) => (c, h, w)
+        let img_tensor = Tensor::from_slice(
+            &img_vec,
+            (resize_h as usize, resize_w as usize, 3),
+            &self.device,
+        )?
+        .permute((2, 0, 1))?
+        .to_dtype(self.dtype)?;
+        // 0-255 rescale to 0-1
+        let img_tensor = img_tensor.affine(1.0 / 255.0, 0.)?;
+        // normalize
+        let img_tensor = img_tensor
+            .broadcast_sub(&img_mean)?
+            .broadcast_div(&img_std)?;
+        // (c, h, w) => (1, c, h, w)
+        let img_tensor = img_tensor.unsqueeze(0)?;
+        Ok(img_tensor)
+    }
+
+    pub fn process_vision_tensor(&self, img_tensor: &Tensor) -> Result<(Tensor, Tensor)> {
+        let channel = img_tensor.dim(1)?;
+        let grid_t = img_tensor.dim(0)? / self.vision_setting.temporal_patch_size;
+        let grid_h = img_tensor.dim(2)? / self.vision_setting.patch_size;
+        let grid_w = img_tensor.dim(3)? / self.vision_setting.patch_size;
+        let shape = Shape::from(vec![
+            grid_t,
+            self.vision_setting.temporal_patch_size,
+            channel,
+            grid_h / self.vision_setting.merge_size,
+            self.vision_setting.merge_size,
+            self.vision_setting.patch_size,
+            grid_w / self.vision_setting.merge_size,
+            self.vision_setting.merge_size,
+            self.vision_setting.patch_size,
+        ]);
+        let img_tensor = img_tensor.reshape(shape)?;
+        // shape to // grid_t,
+        // grid_h / merge_size,
+        // grid_w / merge_size,
+        // merge_size,
+        // merge_size,
+        // channel,
+        // temporal_patch_size,
+        // patch_size,
+        // patch_size,
+        let img_tensor = img_tensor.permute(vec![0, 3, 6, 4, 7, 2, 1, 5, 8])?;
+        let img_tensor = img_tensor
+            .reshape((
+                grid_t * grid_h * grid_w,
+                channel
+                    * self.vision_setting.temporal_patch_size
+                    * self.vision_setting.patch_size
+                    * self.vision_setting.patch_size,
+            ))?
+            .contiguous()?;
+        let grid_thw = Tensor::from_vec(
+            vec![grid_t as u32, grid_h as u32, grid_w as u32],
+            (1, 3),
+            &self.device,
+        )?;
+        Ok((img_tensor, grid_thw))
+    }
+
+    pub fn process_images(
+        &self,
+        imgs: Vec<DynamicImage>,
+        img_mean: &Tensor,
+        img_std: &Tensor,
+    ) -> Result<VisionInput> {
+        let mut pixel_values_vec = Vec::new();
+        let mut vision_grid_thws_vec = Vec::new();
+
+        for img in imgs {
+            let img_tensor = self.process_img(&img, &img_mean, &img_std)?;
+            let img_tensor = Tensor::cat(&[&img_tensor, &img_tensor], 0)?.contiguous()?;
+            let (img_tensor, grid_thw) = self.process_vision_tensor(&img_tensor)?;
+            pixel_values_vec.push(img_tensor);
+            vision_grid_thws_vec.push(grid_thw);
+        }
+        let pixel_values = Tensor::cat(&pixel_values_vec, 0)?;
+        let vision_grid_thws = Tensor::cat(&vision_grid_thws_vec, 0)?;
+        Ok(VisionInput {
+            data: pixel_values,
+            grid_thw: vision_grid_thws,
+        })
+    }
+
+    pub fn process_videos(
+        &self,
+        data: Vec<Tensor>,
+        img_mean: &Tensor,
+        img_std: &Tensor,
+    ) -> Result<VisionInput> {
+        let mut pixel_values_vec = Vec::new();
+        let mut vision_grid_thws_vec = Vec::new();
+        for single_video in data {
+            // 0-255 rescale to 0-1
+            let video_tensor = single_video.to_dtype(self.dtype)?.affine(1.0 / 255.0, 0.)?;
+            // normalize
+            let video_tensor = video_tensor
+                .broadcast_sub(&img_mean)?
+                .broadcast_div(&img_std)?
+                .contiguous()?;
+            let (video_tensor, video_grid_thw) = self.process_vision_tensor(&video_tensor)?;
+            pixel_values_vec.push(video_tensor);
+            vision_grid_thws_vec.push(video_grid_thw);
+        }
+        let pixel_values = Tensor::cat(&pixel_values_vec, 0)?.contiguous()?;
+        let vision_grid_thws = Tensor::cat(&vision_grid_thws_vec, 0)?.contiguous()?;
+        Ok(VisionInput {
+            data: pixel_values,
+            grid_thw: vision_grid_thws,
+        })
+    }
+
+    pub fn process_info(
+        &self,
+        messages: &ChatCompletionParameters,
+        text: &str,
+    ) -> Result<GeneralInput> {
+        let mut pixel_values = None;
+        let mut image_grid_thw = None;
+        let mut pixel_values_video = None;
+        let mut video_grid_thw = None;
+        let mut second_per_grid_ts = None;
+        let vision_map = self.extract_vision_info(messages)?;
+        let img_mean =
+            Tensor::from_slice(&self.vision_setting.image_mean, (3, 1, 1), &self.device)?
+                .to_dtype(self.dtype)?;
+        let img_std = Tensor::from_slice(&self.vision_setting.image_std, (3, 1, 1), &self.device)?
+            .to_dtype(self.dtype)?;
+        for (key, vec) in vision_map {
+            // println!("key: {}, \nvalue: {:?}", key, vec);
+            if key.eq("image") {
+                let mut file_vec = Vec::new();
+                for file in &vec {
+                    let image = get_image(file);
+                    match image {
+                        Ok(img) => file_vec.push(img),
+                        Err(e) => println!("get_image err: {:?}", e),
+                    };
+                }
+                if file_vec.len() > 0 {
+                    let vision_input = self.process_images(file_vec, &img_mean, &img_std);
+                    match vision_input {
+                        Ok(img_input) => {
+                            pixel_values = Some(img_input.data);
+                            image_grid_thw = Some(img_input.grid_thw);
+                        }
+                        Err(e) => println!("img process_images err: {:?}", e),
+                    };
+                }
+            }
+            if key.eq("video") {
+                let mut file_vec = Vec::new();
+                for file in &vec {
+                    let video_data = get_video_data(file, &self.vision_setting, &self.device);
+                    match video_data {
+                        Ok(tensor) => file_vec.push(tensor),
+                        Err(e) => println!("get_video_data err: {:?}", e),
+                    };
+                }
+                if file_vec.len() > 0 {
+                    let vision_input = self.process_videos(file_vec, &img_mean, &img_std);
+                    match vision_input {
+                        Ok(video_input) => {
+                            let video_num = video_input.grid_thw.dim(0)?;
+                            pixel_values_video = Some(video_input.data);
+                            video_grid_thw = Some(video_input.grid_thw);
+                            let second_per_grid = vec![
+                                self.vision_setting.temporal_patch_size
+                                    as f32
+                                    / self.vision_setting.fps;
+                                video_num
+                            ];
+                            second_per_grid_ts = Some(second_per_grid);
+                        }
+                        Err(e) => println!("video process_videos err: {:?}", e),
+                    };
+                }
+            }
+        }
+        let merge_length = self.vision_setting.merge_size.pow(2);
+        let mut text = text.to_string();
+        if image_grid_thw.is_some() {
+            let mut index = 0;
+            while text.contains(&self.image_token) {
+                let grid_i = image_grid_thw.as_ref().unwrap().i(index)?;
+                let repeat_num =
+                    grid_i.to_vec1::<u32>()?.iter().product::<u32>() as usize / merge_length;
+                let replace = "<|placeholder|>".repeat(repeat_num);
+                text = text.replacen(&self.image_token, &replace, 1);
+                index += 1;
+            }
+            text = text.replace("<|placeholder|>", &self.image_token);
+        }
+        if video_grid_thw.is_some() {
+            let mut index = 0;
+            while text.contains(&self.video_token) {
+                let grid_i = video_grid_thw.as_ref().unwrap().i(index)?;
+                let repeat_num =
+                    grid_i.to_vec1::<u32>()?.iter().product::<u32>() as usize / merge_length;
+                let replace = "<|placeholder|>".repeat(repeat_num);
+                text = text.replacen(&self.video_token, &replace, 1);
+                index += 1;
+            }
+            text = text.replace("<|placeholder|>", &self.video_token);
+        }
+        let input = GeneralInput {
+            replace_text: text,
+            pixel_values,
+            image_grid_thw,
+            pixel_values_video,
+            video_grid_thw,
+            second_per_grid_ts,
+        };
+        Ok(input)
+    }
+}
+
+pub fn smart_resize(
+    img_h: u32,
+    img_w: u32,
+    vision_setting: &VisionSetting,
+    is_img: bool,
+    video_ratio: Option<u32>,
+) -> Result<(u32, u32)> {
+    if std::cmp::max(img_h, img_w) / std::cmp::min(img_h, img_w) > vision_setting.max_ratio {
+        return Err(anyhow!(format!(
+            "absolute aspect ratio mush be smaller than {}, got {}",
+            vision_setting.max_ratio,
+            std::cmp::max(img_h, img_w) / std::cmp::min(img_h, img_w)
+        )));
+    }
+    let mut image_factor = vision_setting.image_factor;
+    if let Some(ratio) = video_ratio {
+        image_factor = lcm(image_factor, ratio);
+    }
+    let mut h_bar = std::cmp::max(image_factor, round_by_factor(img_h, image_factor));
+    let mut w_bar = std::cmp::max(image_factor, round_by_factor(img_w, image_factor));
+    let mut max_pixels = 0u32;
+    let mut min_pixels = 0u32;
+    if is_img {
+        min_pixels = vision_setting.min_pixels;
+        max_pixels = vision_setting.max_pixels;
+    } else {
+        min_pixels = vision_setting.video_min_pixels;
+        max_pixels = vision_setting.video_max_pixels;
+    }
+    if h_bar * w_bar > max_pixels {
+        let beta = ((img_h * img_w) as f32 / max_pixels as f32).sqrt();
+        h_bar = floor_by_factor(img_h as f32 / beta, image_factor);
+        w_bar = floor_by_factor(img_w as f32 / beta, image_factor);
+    } else if h_bar * w_bar < min_pixels {
+        let beta = (min_pixels as f32 / (img_h * img_w) as f32).sqrt();
+        h_bar = ceil_by_factor(img_h as f32 * beta, image_factor);
+        w_bar = ceil_by_factor(img_w as f32 * beta, image_factor);
+    }
+    Ok((h_bar, w_bar))
+}
+
+pub fn get_video_data(
+    file: &String,
+    vision_setting: &VisionSetting,
+    device: &Device,
+) -> Result<Tensor> {
+    ffmpeg::init().map_err(|e| anyhow!(format!("Failed to initialize ffmpeg: {}", e)))?;
+
+    let mut ictx = ffmpeg::format::input(&file)
+        .map_err(|e| anyhow!(format!("Failed to open video file: {}", e)))?;
+    let input = ictx
+        .streams()
+        .best(ffmpeg::media::Type::Video)
+        .ok_or_else(|| anyhow!(format!("No video stream found")))?;
+    let video_stream_index = input.index();
+    let context_decoder = ffmpeg::codec::context::Context::from_parameters(input.parameters())
+        .map_err(|e| anyhow!(format!("Failed to crate decoder context: {}", e)))?;
+    let mut decoder = context_decoder
+        .decoder()
+        .video()
+        .map_err(|e| anyhow!(format!("Failed to decoder video: {}", e)))?;
+
+    let video_h = decoder.height();
+    let video_w = decoder.width();
+    let format = decoder.format();
+
+    let frames = input.frames();
+    let rate = (input.rate().0 as f32 / input.rate().1 as f32).round() as u32;
+    // 1s取两帧
+    let min_frames = ceil_by_factor(
+        vision_setting.fps_min_frames as f32,
+        vision_setting.frame_factor,
+    );
+    let max_frames = floor_by_factor(
+        vision_setting.fps_max_frames as f32,
+        vision_setting.frame_factor,
+    );
+    let nframes = (frames as f32 / rate as f32 * vision_setting.fps) as u32;
+    let nframes = std::cmp::min(std::cmp::max(nframes, min_frames), max_frames);
+    let nframes = round_by_factor(nframes, vision_setting.frame_factor);
+    let sample_interval = (frames as f32 / nframes as f32).round() as u32;
+    let mut frame_id = 0_u32;
+
+    // 图片帧使用scaler reshape的时候需要保证宽高是16的倍数,不然reshape出来的是损坏的图片
+    // 所以计算resize的目标宽高时,需要用16和image_factor的最小公倍数
+    let (resize_h, resize_w) = smart_resize(video_h, video_w, vision_setting, false, Some(16))?;
+    let mut scaler = ffmpeg::software::scaling::context::Context::get(
+        format,
+        video_w,
+        video_h,
+        ffmpeg::format::Pixel::RGB24,
+        resize_w,
+        resize_h,
+        ffmpeg::software::scaling::flag::Flags::BILINEAR
+            | ffmpeg::software::scaling::flag::Flags::ACCURATE_RND,
+    )
+    .map_err(|e| anyhow!(format!("Failed to crate scaler: {}", e)))?;
+
+    let mut frames_vec = Vec::new();
+    let mut receive_and_process_decoded_frames =
+        |decoder: &mut ffmpeg::decoder::Video| -> Result<()> {
+            let mut decoded = ffmpeg::frame::Video::empty();
+            while decoder.receive_frame(&mut decoded).is_ok() {
+                if frame_id % sample_interval == 0 {
+                    let mut rgb_frame = ffmpeg::frame::Video::empty();
+                    scaler
+                        .run(&decoded, &mut rgb_frame)
+                        .map_err(|e| anyhow!(format!("Failed to scaler run decoded: {}", e)))?;
+
+                    // save_file(&rgb_frame, frame_id as usize);
+                    let frame_data = rgb_frame.data(0);
+                    let frame_tensor = Tensor::from_slice(
+                        frame_data,
+                        (resize_h as usize, resize_w as usize, 3),
+                        device,
+                    )?
+                    .permute((2, 0, 1))?;
+                    frames_vec.push(frame_tensor);
+                }
+                frame_id += 1;
+            }
+            Ok(())
+        };
+
+    for (stream, packet) in ictx.packets() {
+        if stream.index() == video_stream_index {
+            decoder
+                .send_packet(&packet)
+                .map_err(|e| anyhow!(format!("Failed to send packet: {}", e)))?;
+            receive_and_process_decoded_frames(&mut decoder)?;
+        }
+    }
+    decoder
+        .send_eof()
+        .map_err(|e| anyhow!(format!("Failed to decoder.send_eof(): {}", e)))?;
+    receive_and_process_decoded_frames(&mut decoder)?;
+
+    if frames_vec.is_empty() {
+        return Err(anyhow!("No frames extracted from video".to_string()));
+    }
+    // (t, c, h, w)
+    let frames_tensor = Tensor::stack(&frames_vec, 0)?.contiguous()?;
+    Ok(frames_tensor)
+}
diff --git a/src/position_embed/mod.rs b/src/position_embed/mod.rs
new file mode 100644
index 0000000..f67f4f0
--- /dev/null
+++ b/src/position_embed/mod.rs
@@ -0,0 +1 @@
+pub mod rope;
diff --git a/src/position_embed/rope.rs b/src/position_embed/rope.rs
new file mode 100644
index 0000000..165bcf0
--- /dev/null
+++ b/src/position_embed/rope.rs
@@ -0,0 +1,182 @@
+use anyhow::Result;
+use candle_core::{D, DType, Device, IndexOp, Tensor};
+use candle_transformers::models::deepseek2::SplitOp;
+
+use crate::models::qwen2_5vl::config::RopeScaling;
+
+pub fn compute_default_rope_parameters(dim: usize, base: f32) -> Vec<f32> {
+    let inv_freq: Vec<f32> = (0..dim)
+        .step_by(2)
+        .map(|i| 1.0_f32 / base.powf(i as f32 / dim as f32))
+        .collect();
+    inv_freq
+}
+
+pub fn rotate_half(x: &Tensor) -> Result<Tensor> {
+    let half_dim = x.dim(D::Minus1)? / 2;
+    let x1 = x.narrow(D::Minus1, 0, half_dim)?;
+    let x2 = x.narrow(D::Minus1, half_dim, half_dim)?;
+    let x2 = x2.affine(-1.0, 0.0)?;
+    let rotate_x = Tensor::cat(&[&x2, &x1], D::Minus1)?.contiguous()?;
+    Ok(rotate_x)
+}
+
+pub fn apply_multimodel_rotary_pos_emb(
+    q: &Tensor,
+    k: &Tensor,
+    cos: &Tensor,
+    sin: &Tensor,
+    mrope_section: Vec<usize>,
+) -> Result<(Tensor, Tensor)> {
+    let mrope_section = mrope_section.repeat(2);
+    let cos_select: Vec<Tensor> = cos
+        .split(&mrope_section, D::Minus1)?
+        .iter()
+        .enumerate()
+        .map(|(i, m)| m.i(i % 3).unwrap())
+        .collect();
+    let cos = Tensor::cat(&cos_select, D::Minus1)?
+        .unsqueeze(1)?
+        .contiguous()?;
+    let sin_select: Vec<Tensor> = sin
+        .split(&mrope_section, D::Minus1)?
+        .iter()
+        .enumerate()
+        .map(|(i, m)| m.i(i % 3).unwrap())
+        .collect();
+    let sin = Tensor::cat(&sin_select, D::Minus1)?
+        .unsqueeze(1)?
+        .contiguous()?;
+    let q_embed = q
+        .broadcast_mul(&cos)?
+        .add(&rotate_half(&q)?.broadcast_mul(&sin)?)?;
+    let k_embed = k
+        .broadcast_mul(&cos)?
+        .add(&rotate_half(&k)?.broadcast_mul(&sin)?)?;
+    Ok((q_embed, k_embed))
+}
+
+pub fn apply_rotary_pos_emb_vision(
+    q: &Tensor,
+    k: &Tensor,
+    cos: &Tensor,
+    sin: &Tensor,
+) -> Result<(Tensor, Tensor)> {
+    // q, k -> (seq_len, num_heads, head_dim)
+    // cos, sin -> (seq_len, head_dim) -> (seq_len, 1, head_dim)
+    let cos = cos.unsqueeze(D::Minus2)?;
+    let sin = sin.unsqueeze(D::Minus2)?;
+    let q_embed = q
+        .broadcast_mul(&cos)?
+        .add(&rotate_half(q)?.broadcast_mul(&sin)?)?;
+    let k_embed = k
+        .broadcast_mul(&cos)?
+        .add(&rotate_half(k)?.broadcast_mul(&sin)?)?;
+    Ok((q_embed, k_embed))
+}
+
+pub fn apply_rotary_pos_emb(
+    q: &Tensor,
+    k: &Tensor,
+    cos: &Tensor,
+    sin: &Tensor,
+) -> Result<(Tensor, Tensor)> {
+    // sin/cos: (bs, 1, seq_len, head_dim)
+    // q/k: (bs, n_head, seq_len, head_dim)
+    let q_embed = q
+        .broadcast_mul(&cos)?
+        .add(&rotate_half(q)?.broadcast_mul(&sin)?)?;
+    let k_embed = k
+        .broadcast_mul(&cos)?
+        .add(&rotate_half(k)?.broadcast_mul(&sin)?)?;
+    Ok((q_embed, k_embed))
+}
+
+#[derive(Debug, Clone)]
+pub struct Qwen2_5VLTextRotaryEmbedding {
+    inv_freq: Vec<f32>,
+}
+
+impl Qwen2_5VLTextRotaryEmbedding {
+    pub fn new(dim: usize, theta_base: f32) -> Self {
+        let inv_freq = compute_default_rope_parameters(dim, theta_base);
+        Self { inv_freq }
+    }
+    pub fn forward(
+        &self,
+        position_ids: &Tensor,
+        dtype: DType,
+        mrope_section: Vec<usize>,
+    ) -> Result<(Tensor, Tensor)> {
+        // position_ids shape: (3, bs, position) -> (3, bs, 1, position)
+        let position_ids_expanded = position_ids
+            .unsqueeze(D::Minus2)?
+            .to_dtype(DType::F32)?
+            .contiguous()?;
+        // inv_freq Vec<f32> -> Tensor(1, 1, head_dim / 2, 1) -> (3, bs, head_dim / 2, 1)
+        let inv_freq_expanded = Tensor::from_vec(
+            self.inv_freq.clone(),
+            (1, 1, self.inv_freq.len(), 1),
+            position_ids.device(),
+        )?
+        .broadcast_as((3, position_ids.dim(1)?, self.inv_freq.len(), 1))?
+        .to_dtype(DType::F32)?
+        .contiguous()?;
+
+        // (3, bs, head_dim / 2, 1) matmul (3, bs, 1, position)
+        //    -> (3, bs, head_dim / 2, seq_len) -> (3, bs, seq_len, head_dim / 2)
+        let freqs = inv_freq_expanded
+            .matmul(&position_ids_expanded)?
+            .transpose(2, 3)?;
+        // let freqs = position_ids_expanded.matmul(&inv_freq_expanded)?;
+        // (3, bs, seq_len, head_dim / 2) -> (3, bs, seq_len, head_dim)
+        let emb = Tensor::cat(&[&freqs, &freqs], D::Minus1)?.contiguous()?;
+        let cos = emb.cos()?;
+        let sin = emb.sin()?;
+        let mrope_section = mrope_section.repeat(2);
+        let cos_select: Vec<Tensor> = cos
+            .split(&mrope_section, D::Minus1)?
+            .iter()
+            .enumerate()
+            .map(|(i, m)| m.i(i % 3).unwrap())
+            .collect();
+        // (bs, seq_len, head_dim) -> (bs, 1, seq_len, head_dim)
+        let cos = Tensor::cat(&cos_select, D::Minus1)?
+            .unsqueeze(1)?
+            .contiguous()?;
+        let sin_select: Vec<Tensor> = sin
+            .split(&mrope_section, D::Minus1)?
+            .iter()
+            .enumerate()
+            .map(|(i, m)| m.i(i % 3).unwrap())
+            .collect();
+        // (bs, seq_len, head_dim) -> (bs, 1, seq_len, head_dim)
+        let sin = Tensor::cat(&sin_select, D::Minus1)?
+            .unsqueeze(1)?
+            .contiguous()?;
+        Ok((cos.to_dtype(dtype)?, sin.to_dtype(dtype)?))
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct Qwen2_5VisionRotaryEmbedding {
+    inv_freq: Vec<f32>,
+}
+
+impl Qwen2_5VisionRotaryEmbedding {
+    pub fn new(dim: usize, theta_base: Option<f32>) -> Self {
+        let theta_base = match theta_base {
+            Some(theta) => theta,
+            None => 10000.0_f32,
+        };
+        let inv_freq = compute_default_rope_parameters(dim, theta_base);
+        Self { inv_freq }
+    }
+
+    pub fn forward(&self, seqlen: usize, device: &Device) -> Result<Tensor> {
+        let seq = Tensor::arange(0.0_f32, seqlen as f32, device)?.reshape((seqlen, 1))?;
+        let inv_freq = Tensor::from_vec(self.inv_freq.clone(), (1, self.inv_freq.len()), device)?;
+        let freqs = seq.matmul(&inv_freq)?;
+        Ok(freqs)
+    }
+}
diff --git a/src/tokenizer/mod.rs b/src/tokenizer/mod.rs
new file mode 100644
index 0000000..5768976
--- /dev/null
+++ b/src/tokenizer/mod.rs
@@ -0,0 +1 @@
+pub mod tokenizer;
\ No newline at end of file
diff --git a/src/tokenizer/tokenizer.rs b/src/tokenizer/tokenizer.rs
new file mode 100644
index 0000000..8040a76
--- /dev/null
+++ b/src/tokenizer/tokenizer.rs
@@ -0,0 +1,44 @@
+use anyhow::{Result, anyhow};
+use candle_core::{Device, Tensor};
+use tokenizers::Tokenizer;
+
+pub struct TokenizerModel {
+    tokenizer: Tokenizer,
+}
+
+impl TokenizerModel {
+    pub fn init(path: &str) -> Result<Self> {
+        let path = path.to_string();
+        assert!(
+            std::path::Path::new(&path).exists(),
+            "model path file not exists"
+        );
+        let tokenizer_file = path.clone() + "/tokenizer.json";
+        assert!(
+            std::path::Path::new(&tokenizer_file).exists(),
+            "tokenizer.json not exists in model path"
+        );
+        let tokenizer = Tokenizer::from_file(tokenizer_file)
+            .map_err(|e| anyhow!(format!("tokenizer from file error{}", e)))?;
+        Ok(Self { tokenizer })
+    }
+
+    pub fn text_encode(&self, text: String, device: &Device) -> Result<Tensor> {
+        let token_id = self
+            .tokenizer
+            .encode(text, true)
+            .map_err(|e| anyhow!(format!("tokenizer encode error: {}", e)))?
+            .get_ids()
+            .to_vec();
+        let token_tensor = Tensor::from_slice(&token_id, (1, token_id.len()), device)?;
+        Ok(token_tensor)
+    }
+
+    pub fn token_decode(&self, tokens: Vec<u32>) -> Result<String> {
+        let decode = self
+            .tokenizer
+            .decode(&tokens, true)
+            .map_err(|e| anyhow!(format!("tokenizer encode error{}", e)))?;
+        Ok(decode)
+    }
+}
diff --git a/src/utils/audio_utils.rs b/src/utils/audio_utils.rs
new file mode 100644
index 0000000..e69de29
diff --git a/src/utils/img_utils.rs b/src/utils/img_utils.rs
new file mode 100644
index 0000000..84b438b
--- /dev/null
+++ b/src/utils/img_utils.rs
@@ -0,0 +1,62 @@
+use std::io::Cursor;
+
+use anyhow::{Result, anyhow};
+use base64::{Engine, engine::general_purpose};
+use image::{DynamicImage, ImageReader};
+
+pub fn load_image_from_url(url: &str) -> Result<DynamicImage> {
+    let response = reqwest::blocking::get(url)
+        .map_err(|e| anyhow!(format!("Failed to fetch image from url: {}", e)))?;
+    let bytes = response
+        .bytes()
+        .map_err(|e| anyhow!(format!("Failed to get image bytes: {}", e)))?;
+
+    let cursor = Cursor::new(bytes);
+    let img = ImageReader::new(cursor)
+        .with_guessed_format()
+        .map_err(|e| anyhow!(format!("Failed to read image format: {}", e)))?
+        .decode()
+        .map_err(|e| anyhow!(format!("Failed to decode image: {}", e)))?;
+    Ok(img)
+}
+
+pub fn load_image_from_base64(base64_data: &str) -> Result<DynamicImage> {
+    let image_data = general_purpose::STANDARD
+        .decode(base64_data)
+        .map_err(|e| anyhow!(format!("Failed to decode image: {}", e)))?;
+    let cursor = Cursor::new(image_data);
+    let img = ImageReader::new(cursor)
+        .with_guessed_format()
+        .map_err(|e| anyhow!(format!("Failed to read image format: {}", e)))?
+        .decode()
+        .map_err(|e| anyhow!(format!("Failed to decode image: {}", e)))?;
+    Ok(img)
+}
+
+pub fn get_image(file: &String) -> Result<DynamicImage> {
+    let mut img = None;
+    if file.starts_with("http://") || file.starts_with("https://") {
+        img = Some(load_image_from_url(&file)?);
+    }
+    if file.starts_with("file://") {
+        let mut path = file.clone();
+        path = path.split_off(7);
+        img = Some(
+            ImageReader::open(path)
+                .map_err(|e| anyhow!(format!("Failed to open file: {}", e)))?
+                .decode()
+                .map_err(|e| anyhow!(format!("Failed to decode image: {}", e)))?,
+        );
+    }
+    if file.starts_with("data:image") {
+        if file.contains("base64,") {
+            let data: Vec<&str> = file.split("base64,").collect();
+            let data = data[1];
+            img = Some(load_image_from_base64(data)?);
+        }
+    }
+    if img.is_some() {
+        return Ok(img.unwrap());
+    }
+    Err(anyhow!("get image from message failed".to_string()))
+}
diff --git a/src/utils/mod.rs b/src/utils/mod.rs
new file mode 100644
index 0000000..d9f4429
--- /dev/null
+++ b/src/utils/mod.rs
@@ -0,0 +1,4 @@
+pub mod tensor_utils;
+pub mod utils;
+pub mod img_utils;
+pub mod video_utils;
\ No newline at end of file
diff --git a/src/utils/tensor_utils.rs b/src/utils/tensor_utils.rs
new file mode 100644
index 0000000..9426025
--- /dev/null
+++ b/src/utils/tensor_utils.rs
@@ -0,0 +1,202 @@
+use anyhow::{Result, anyhow};
+use candle_core::{D, DType, Device, IndexOp, Tensor, shape::Dim};
+
+pub fn split(t: &Tensor, splits: &[usize], dim: D) -> Result<Vec<Tensor>> {
+    let dim = dim.to_index(t.shape(), "split")?;
+    let mut split_res = Vec::new();
+    let mut index = 0;
+    for split in splits {
+        split_res.push(t.narrow(dim, index, *split)?);
+        index += *split;
+    }
+    Ok(split_res)
+}
+
+pub fn safe_arg_sort_last_dim(t: &Tensor, ascending: bool) -> Result<Tensor> {
+    // tensor在GPU上时，维度超过1024， arg_sort_last_dim方法会报错
+    // 所以维度大于1024时，放到CPU上处理
+    let last_dim = t.dims()[t.rank() - 1];
+    if last_dim <= 1024 {
+        let t = t.arg_sort_last_dim(ascending)?;
+        Ok(t)
+    } else {
+        let cpu_tensor = t.to_device(&Device::Cpu)?;
+        let sorted_indices = cpu_tensor.arg_sort_last_dim(ascending)?;
+        let t = sorted_indices.to_device(t.device())?;
+        Ok(t)
+    }
+}
+
+pub fn nonzero_index_vec(mask: &Tensor) -> Result<Vec<u32>> {
+    // 根据mask矩阵选出其中不为0的元素所在索引, 返回vec
+    // 只能处理1维数据
+    let mut mask = mask.clone();
+    if mask.dtype() != DType::U32 {
+        mask = mask.to_dtype(DType::U32)?;
+    }
+    match mask.rank() {
+        0 => {
+            return Err(anyhow!(format!(
+                "input rank must > 0, the input tensor rank: {}",
+                mask.rank()
+            )));
+        }
+        1 => {
+            let mask_vector = mask.to_vec1::<u32>()?;
+            let indices: Vec<u32> = mask_vector
+                .iter()
+                .enumerate()
+                .filter_map(|(idx, &val)| if val != 0 { Some(idx as u32) } else { None })
+                .collect();
+            Ok(indices)
+        }
+        _ => {
+            return Err(anyhow!(format!(
+                "input rank not support, the input tensor rank: {}",
+                mask.rank()
+            )));
+        }
+    }
+}
+
+pub fn nonzero_index(mask: &Tensor) -> Result<Tensor> {
+    // 根据mask矩阵选出其中不为1的元素所在索引, 返回Tensor
+    let indices_tensor = match mask.rank() {
+        0 => {
+            return Err(anyhow!(format!(
+                "input rank must > 0, the input tensor rank: {}",
+                mask.rank()
+            )));
+        }
+        1 => {
+            let index_vec = nonzero_index_vec(mask)?;
+            let indices_tensor = Tensor::from_slice(&index_vec, index_vec.len(), mask.device())?;
+            indices_tensor
+        }
+        _ => {
+            return Err(anyhow!(format!(
+                "input rank must == 1, the input tensor rank: {}",
+                mask.rank()
+            )));
+        }
+    };
+    Ok(indices_tensor)
+}
+
+pub fn zero_index_vec(mask: &Tensor) -> Result<Vec<u32>> {
+    // 根据mask矩阵选出其中为0的元素所在索引, 返回vec
+    // 只能处理1维数据
+    let mut mask = mask.clone();
+    if mask.dtype() != DType::U32 {
+        mask = mask.to_dtype(DType::U32)?;
+    }
+    match mask.rank() {
+        0 => {
+            return Err(anyhow!(format!(
+                "input rank must > 0, the input tensor rank: {}",
+                mask.rank()
+            )));
+        }
+        1 => {
+            let mask_vector = mask.to_vec1::<u32>()?;
+            let indices: Vec<u32> = mask_vector
+                .iter()
+                .enumerate()
+                .filter_map(|(idx, &val)| if val == 0 { Some(idx as u32) } else { None })
+                .collect();
+            Ok(indices)
+        }
+        _ => {
+            return Err(anyhow!(format!(
+                "input rank not support, the input tensor rank: {}",
+                mask.rank()
+            )));
+        }
+    }
+}
+
+pub fn zero_index(mask: &Tensor) -> Result<Tensor> {
+    let index_vec = zero_index_vec(mask)?;
+    let indices_tensor = Tensor::from_slice(&index_vec, index_vec.len(), mask.device())?;
+    Ok(indices_tensor)
+}
+
+pub fn nonzero_slice(mask: &Tensor) -> Result<Vec<(usize, usize)>> {
+    // 根据mask矩阵选出其中非0的元素所在索引
+    // 根据索引获取连续索引间隔
+    // 如不为零索引元素为[0, 3, 4, 5, 8, 9]
+    // 间隔为: [(0, 1), (3, 6), (8, 10)]
+    // 索引前闭后开
+    let mut index_vec = nonzero_index_vec(mask)?;
+    match index_vec.len() {
+        0 => {
+            return Ok(vec![]);
+        }
+        1 => {
+            return Ok(vec![(index_vec[0] as usize, (index_vec[0] + 1) as usize)]);
+        }
+        _ => {
+            let mut vec_slice = vec![];
+            let mut start = index_vec.remove(0);
+            let mut last = start;
+
+            for i in index_vec {
+                if i == (last + 1) {
+                    last = i;
+                    continue;
+                } else {
+                    vec_slice.push((start as usize, (last + 1) as usize));
+                    start = i;
+                    last = i;
+                }
+            }
+            vec_slice.push((start as usize, (last + 1) as usize));
+            Ok(vec_slice)
+        }
+    }
+}
+
+pub fn masked_scatter_dim0(original: &Tensor, replace: &Tensor, mask: &Tensor) -> Result<Tensor> {
+    // 根据mask中非0元素所在索引,使用replace中的数据替换掉original中的数据
+    // original: rank = 3: (bs, seq_len, hidden_dim)
+    // replace: rank = 2: (seq_len, hidden_dim)
+    // mask: rank = 2: (bs, seq_len)
+    // 推理时bs=1,为了方便替换,将bs squeeze,替换后再unsqueeze
+    // 按行替换
+    if original.dim(0)? != 1 || mask.dim(0)? != 1 {
+        return Err(anyhow!(format!(
+            "masked_scatter_dim0 original bs: {} or mask bs :{} not equal to 1 ",
+            original.dim(0)?,
+            mask.dim(0)? != 1
+        )));
+    }
+    let mut original = original.squeeze(0)?;
+    let mask = mask.squeeze(0)?;
+    let slices = nonzero_slice(&mask)?;
+    let mut sub_start = 0usize;
+    let mut sub_end = 0usize;
+    for (start, end) in slices {
+        sub_end = sub_start + (end - start);
+        let sub_replace = replace.i((sub_start..sub_end, ..))?;
+        original = original.slice_assign(&[(start..end), (0..original.dim(1)?)], &sub_replace)?;
+        sub_start = sub_end;
+    }
+    original = original.unsqueeze(0)?;
+    Ok(original)
+}
+
+pub fn get_equal_mask(input_ids: &Tensor, token_ids: u32) -> Result<Tensor> {
+    let image_token_id_tensor = Tensor::new(vec![token_ids], input_ids.device())?;
+    let mask = input_ids
+        .broadcast_eq(&image_token_id_tensor)?
+        .to_dtype(candle_core::DType::U32)?;
+    Ok(mask)
+}
+
+pub fn get_vision_next_indices(input_ids: &Tensor, token_id: u32) -> Result<Tensor> {
+    // input_ids -> shape: (seq_len)
+    let mask = get_equal_mask(&input_ids, token_id)?;
+    let indices = nonzero_index(&mask)?;
+    let indices = indices.broadcast_add(&Tensor::new(vec![1u32], input_ids.device())?)?;
+    Ok(indices)
+}
diff --git a/src/utils/utils.rs b/src/utils/utils.rs
new file mode 100644
index 0000000..fd26e1d
--- /dev/null
+++ b/src/utils/utils.rs
@@ -0,0 +1,87 @@
+use anyhow::Result;
+use candle_core::{DType, Device};
+
+pub fn get_device(device: Option<&Device>) -> Device {
+    match device {
+        Some(d) => d.clone(),
+        None => {
+            #[cfg(feature = "cuda")]
+            {
+                Device::new_cuda(0).unwrap_or(Device::Cpu)
+            }
+            #[cfg(not(feature = "cuda"))]
+            {
+                Device::Cpu
+            }
+        }
+    }
+}
+
+pub fn get_dtype(dtype: Option<DType>, cfg_dtype: &str) -> DType {
+    match dtype {
+        Some(d) => d,
+        None => {
+            #[cfg(feature = "cuda")]
+            {
+                match cfg_dtype {
+                    "float32" | "float" => DType::F32,
+                    "float64" | "double" => DType::F64,
+                    "float16" => DType::F16,
+                    "bfloat16" => DType::BF16,
+                    "uint8" => DType::U8,
+                    "int8" | "int16" | "int32" | "int64" => DType::I64,
+                    _ => DType::F32,
+                }
+            }
+            #[cfg(not(feature = "cuda"))]
+            {
+                match cfg_dtype {
+                    "float32" | "float" => DType::F32,
+                    "float64" | "double" => DType::F64,
+                    "float16" | "bfloat16" => DType::F16, // cpu上bfloat16有问题
+                    "uint8" => DType::U8,
+                    "int8" | "int16" | "int32" | "int64" => DType::I64,
+                    _ => DType::F32,
+                }
+            }
+        }
+    }
+}
+
+pub fn string_to_static_str(s: String) -> &'static str {
+    Box::leak(s.into_boxed_str())
+}
+
+pub fn find_safetensors_files(path: &str) -> Result<Vec<String>> {
+    let mut files = Vec::new();
+
+    for entry in std::fs::read_dir(path)? {
+        let entry = entry?;
+        let file_path = entry.path();
+
+        if file_path.is_file() {
+            if let Some(extension) = file_path.extension() {
+                if extension == "safetensors" {
+                    files.push(file_path.to_string_lossy().to_string());
+                }
+            }
+        }
+    }
+
+    Ok(files)
+}
+
+pub fn round_by_factor(num: u32, factor: u32) -> u32 {
+    let round = (num as f32 / factor as f32).round() as u32;
+    round * factor
+}
+
+pub fn floor_by_factor(num: f32, factor: u32) -> u32 {
+    let floor = (num / factor as f32).floor() as u32;
+    floor * factor
+}
+
+pub fn ceil_by_factor(num: f32, factor: u32) -> u32 {
+    let ceil = (num / factor as f32).ceil() as u32;
+    ceil * factor
+}
diff --git a/src/utils/video_utils.rs b/src/utils/video_utils.rs
new file mode 100644
index 0000000..417928c
--- /dev/null
+++ b/src/utils/video_utils.rs
@@ -0,0 +1,13 @@
+use ffmpeg_next as ffmpeg;
+use std::{fs::File, io::Write};
+
+#[allow(unused)]
+fn save_file(
+    frame: &ffmpeg::frame::Video,
+    index: usize,
+) -> std::result::Result<(), std::io::Error> {
+    let mut file = File::create(format!("frame{}.ppm", index))?;
+    file.write_all(format!("P6\n{} {}\n255\n", frame.width(), frame.height()).as_bytes())?;
+    file.write_all(frame.data(0))?;
+    Ok(())
+}
diff --git a/tests/config_tests.rs b/tests/config_tests.rs
new file mode 100644
index 0000000..8e41bf1
--- /dev/null
+++ b/tests/config_tests.rs
@@ -0,0 +1,12 @@
+use anyhow::Result;
+use aha::models::qwen2_5vl::config::Config;
+
+#[test]
+fn qwen2_5vl_config() -> Result<()> {
+    // cargo test qwen2_5vl_config -- --nocapture
+    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen2.5-VL-3B-Instruct/";
+    let config_path = model_path.to_string() + "/config.json";
+    let config: Config = serde_json::from_slice(&std::fs::read(config_path)?)?;
+    println!("{:?}", config);
+    Ok(())
+}
\ No newline at end of file
diff --git a/tests/test_qwen2_5vl.rs b/tests/test_qwen2_5vl.rs
new file mode 100644
index 0000000..3c5bcd5
--- /dev/null
+++ b/tests/test_qwen2_5vl.rs
@@ -0,0 +1,56 @@
+use std::time::Instant;
+
+use aha::{models::{qwen2_5vl::generate::Qwen2_5VLGenerateModel, GenerateModel}, ModelType};
+use anyhow::{Result};
+use candle_core::{DType, Device};
+use openai_dive::v1::resources::chat::ChatCompletionParameters;
+
+
+#[test]
+fn qwen2_5vl_generate() -> Result<()> {
+    // test with cpu :(太慢了, : RUST_BACKTRACE=1 cargo test qwen2_5vl_generate -- --nocapture
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda qwen2_5vl_generate -- --nocapture
+    // test with cuda+flash-attn: RUST_BACKTRACE=1 cargo test -F cuda,flash-attn qwen2_5vl_generate -- --nocapture
+    let device = Device::cuda_if_available(0)?;
+    let dtype = DType::BF16;
+
+    let model_path = "/home/jhq/huggingface_model/Qwen/Qwen2.5-VL-3B-Instruct/";
+    
+    let message = r#"
+    {
+        "model": "qwen2.5vl",
+        "messages": [
+            {
+                "role": "user",
+                "content": [ 
+                    {
+                        "type": "image",
+                        "image_url": 
+                        {
+                            "url": "file://./assets/img/ocr_test.png"
+                        }
+                    },               
+                    {
+                        "type": "text", 
+                        "text": "请分析图片并提取所有可见文本内容，按从左到右、从上到下的布局，返回纯文本"
+                    }
+                ]
+            }
+        ]
+    }
+    "#;
+    let mes:ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    // let mut model = Qwen2_5VLGenerateModel::init(model_path, &device, dtype)?;
+    let mut model = ModelType::init(ModelType::Qwen2_5VL, model_path, None, None)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in load model is: {:?}", i_duration);
+
+    let i_start = Instant::now();
+    let result = model.generate(mes)?;
+    println!("generate: \n{}", result);    
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in generate is: {:?}", i_duration);
+
+    Ok(())
+}
\ No newline at end of file