add PaddleOCR-VL-1.6

README.md

@@ -35,7 +35,7 @@ aha is a high-performance, cross-platform AI inference engine built with Rust an
 |----------|--------|
 | **Text** | Qwen3, MiniCPM4, MiniCPM5, LFM2, LFM2.5 |
 | **Vision** | Qwen2.5-VL, Qwen3-VL, Qwen3.5, <br> LFM2.5-VL, LFM2-VL |
-| **OCR** | DeepSeek-OCR, DeepSeek-OCR-2 , PaddleOCR-VL <br> PaddleOCR-VL1.5, Hunyuan-OCR, GLM-OCR |
+| **OCR** | DeepSeek-OCR, DeepSeek-OCR-2 , PaddleOCR-VL <br> PaddleOCR-VL1.5, PaddleOCR-VL1.5, Hunyuan-OCR <br> GLM-OCR |
 | **ASR** | GLM-ASR-Nano, Fun-ASR-Nano, Qwen3-ASR |
 | **TTS** | VoxCPM, VoxCPM1.5, VoxCPM2, Moss-TTS-Nano |
 | **Image** | RMBG-2.0 (background removal) |
@@ -43,6 +43,9 @@ aha is a high-performance, cross-platform AI inference engine built with Rust an
 | **Reranker** | Qwen3-Reranker |
 
 ## Changelog
+### 2026-06-06
+- add PaddleOCR-VL-1.6
+
 ### 2026-05-29
 - generate code refactored
 
@@ -58,8 +61,6 @@ aha is a high-performance, cross-platform AI inference engine built with Rust an
 ### 2026-05-11
 - add Moss-TTS-Nano，its performance is worse than the original Python version
 
-### 2026-05-09
-- merge pr/eastgold15/46, add aha-ui
 
 **[View full changelog](docs/changelog.md)** →
 

README.zh-CN.md

@@ -42,6 +42,9 @@ aha 是一款基于 Rust 和 Candle 框架构建的高性能跨平台 AI 推理
 | **重排序** | Qwen3-Reranker |
 
 ## 更新日志
+### 2026-06-06
+- 新增 PaddleOCR-VL-1.6
+
 ### 2026-05-29
 - generate代码重构完成
 
@@ -57,10 +60,6 @@ aha 是一款基于 Rust 和 Candle 框架构建的高性能跨平台 AI 推理
 ### 2026-05-11
 - 添加Moss-TTS-Nano，该模型比较小，计算误差影响较大，效果比python原版差
 
-### 2026-05-09
-- 合并 pr/eastgold15/46, 添加 aha-ui
-
-
 **[查看完整更新日志](docs/changelog.zh-CN.md)** →
 
 

docs/changelog.md

@@ -5,6 +5,9 @@ All notable changes to aha will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+### 2026-06-06
+- add PaddleOCR-VL-1.6
+
 ### 2026-05-29
 - generate code refactored
 

docs/changelog.zh-CN.md

@@ -5,6 +5,9 @@
 格式基于 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.0.0/)，
 本项目遵循 [语义化版本](https://semver.org/lang/zh-CN/spec/v2.0.0.html)。
 
+### 2026-06-06
+- 新增 PaddleOCR-VL-1.6
+
 ### 2026-05-29
 - generate代码重构完成
 

docs/model-card.md

@@ -3,7 +3,7 @@
 #### Options prompt: 
 "OCR:" | "Table Recognition:" | "Formula Recognition:" | "Chart Recognition:"
 
-## PaddleOCR-VL1.5
+## PaddleOCR-VL1.5/1.6
 #### Options prompt: 
 "OCR:" | "Table Recognition:" | "Formula Recognition:" | "Chart Recognition:" | "Spotting:" | "Seal Recognition:"
 

docs/model-card.zh-CN.md

@@ -3,7 +3,7 @@
 #### 可选提示词: 
 "OCR:" | "Table Recognition:" | "Formula Recognition:" | "Chart Recognition:"
 
-## PaddleOCR-VL1.5
+## PaddleOCR-VL1.5/1.6
 #### 可选提示词: 
 "OCR:" | "Table Recognition:" | "Formula Recognition:" | "Chart Recognition:" | "Spotting:" | "Seal Recognition:"
 

src/cli/mod.rs

@@ -298,7 +298,9 @@ pub(crate) fn run_run(args: RunArgs) -> anyhow::Result<()> {
         WhichModel::HunyuanOCR => {
             hunyuan_ocr::HunyuanORExec::run(&input, output.as_deref(), &weight_path)?;
         }
-        WhichModel::PaddleOCRVL | WhichModel::PaddleOCRVL1_5 => {
+        WhichModel::PaddleOCRVL 
+        | WhichModel::PaddleOCRVL1_5
+        | WhichModel::PaddleOCRVL1_6 => {
             paddleocr_vl::PaddleOVLExec::run(&input, output.as_deref(), &weight_path)?;
         }
         WhichModel::RMBG2_0 => {

src/models/common/model_mapping.rs

@@ -72,6 +72,8 @@ pub enum WhichModel {
     PaddleOCRVL,
     #[value(name = "PaddlePaddle/PaddleOCR-VL-1.5")]
     PaddleOCRVL1_5,
+    #[value(name = "PaddlePaddle/PaddleOCR-VL-1.6")]
+    PaddleOCRVL1_6,
     #[value(name = "AI-ModelScope/RMBG-2.0")]
     RMBG2_0,
     #[value(name = "OpenBMB/VoxCPM-0.5B")]
@@ -165,7 +167,8 @@ impl WhichModel {
             | WhichModel::HunyuanOCR
             | WhichModel::GlmOCR
             | WhichModel::PaddleOCRVL
-            | WhichModel::PaddleOCRVL1_5 => "ocr",
+            | WhichModel::PaddleOCRVL1_5
+            | WhichModel::PaddleOCRVL1_6 => "ocr",
             // ASR models
             WhichModel::Qwen3ASR0_6B
             | WhichModel::Qwen3ASR1_7B

src/models/mod.rs

@@ -284,7 +284,9 @@ pub fn load_model<'a>(
             let model = HunyuanOCRGenerateModel::init(path, device, dtype)?;
             ModelInstance::HunyuanOCR(model)
         }
-        WhichModel::PaddleOCRVL | WhichModel::PaddleOCRVL1_5 => {
+        WhichModel::PaddleOCRVL 
+        | WhichModel::PaddleOCRVL1_5
+        | WhichModel::PaddleOCRVL1_6 => {
             let model = PaddleOCRVLGenerateModel::init(path, device, dtype)?;
             ModelInstance::PaddleOCRVL(Box::new(model))
         }

tests/test_paddleocr_vl.rs

@@ -5,6 +5,55 @@ use aha::params::chat::ChatCompletionParameters;
 use anyhow::Result;
 use rocket::futures::StreamExt;
 
+#[test]
+fn paddleocr_vl_1_6_generate() -> Result<()> {
+    // RUST_BACKTRACE=1 cargo test -F cuda --test test_paddleocr_vl paddleocr_vl_1_6_generate -r -- --nocapture
+    let message = r#"
+    {
+        "model": "paddleocr_vl1.6",
+        "messages": [
+            {
+                "role": "user",
+                "content": [ 
+                    {
+                        "type": "image",
+                        "image_url": 
+                        {
+                            "url": "file://./assets/img/ocr_test1.png"
+                        }
+                    },              
+                    {
+                        "type": "text", 
+                        "text": "Spotting:"
+                    }
+                ]
+            }
+        ],
+        "stream": false
+    }
+    "#;
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/PaddlePaddle/PaddleOCR-VL-1.6/", save_dir);
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut model = PaddleOCRVLGenerateModel::init(&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 res = model.generate(mes)?;
+    let i_duration = i_start.elapsed();
+    println!("generate: \n {:?}", res);
+    if let Some(usage) = &res.usage {
+        let num_token = usage.total_tokens;
+        let duration_secs = i_duration.as_secs_f64();
+        let tps = num_token as f64 / duration_secs;
+        println!("Tokens per second (TPS): {:.2}", tps);
+    }
+    println!("Time elapsed in generate is: {:?}", i_duration);
+    Ok(())
+}
+
 #[test]
 fn paddleocr_vl_1_5_generate() -> Result<()> {
     // RUST_BACKTRACE=1 cargo test -F cuda --test test_paddleocr_vl paddleocr_vl_1_5_generate -r -- --nocapture