diff --git a/src/models/qwen2_5vl/model.rs b/src/models/qwen2_5vl/model.rs index 27ba304..9776858 100644 --- a/src/models/qwen2_5vl/model.rs +++ b/src/models/qwen2_5vl/model.rs @@ -493,19 +493,24 @@ impl Qwen2_5VLVisionModel { 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::()?; - 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 as usize)?); - } - Tensor::cat(&cu_seqlens_repeat, 0)?.flatten_all()? - } - _ => { - return Err(anyhow!(format!("create cu_seqlens error"))); - } - }; + // 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 as usize)?); + // } + // Tensor::cat(&cu_seqlens_repeat, 0)?.flatten_all()? + // } + // _ => { + // return Err(anyhow!(format!("create cu_seqlens error"))); + // } + // }; + 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 as usize)?); + } + let cu_seqlens_full = Tensor::cat(&cu_seqlens_repeat, 0)?.flatten_all()?; let cu_seqlens = cu_seqlens_full .to_dtype(DType::F64)? .cumsum(0)? diff --git a/tests/messy_test.rs b/tests/messy_test.rs index a64b3c8..72d7062 100644 --- a/tests/messy_test.rs +++ b/tests/messy_test.rs @@ -1,4 +1,3 @@ -use aha::utils::tensor_utils::bitor_tensor; use anyhow::Result; use candle_core::{IndexOp, Tensor}; diff --git a/tests/test_qwen2_5vl.rs b/tests/test_qwen2_5vl.rs index a5b65c0..2c54889 100644 --- a/tests/test_qwen2_5vl.rs +++ b/tests/test_qwen2_5vl.rs @@ -28,7 +28,7 @@ fn qwen2_5vl_generate() -> Result<()> { { "url": "file://./assets/img/ocr_test1.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"