From 01b28dad9144340baf848bc88e813033892ac303 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Mon, 13 Oct 2025 11:09:38 +0800 Subject: [PATCH 1/7] fix whisper metric --- ACL_PyTorch/built-in/audio/whisper/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ACL_PyTorch/built-in/audio/whisper/README.md b/ACL_PyTorch/built-in/audio/whisper/README.md index bba8f63259..024d54530d 100644 --- a/ACL_PyTorch/built-in/audio/whisper/README.md +++ b/ACL_PyTorch/built-in/audio/whisper/README.md @@ -101,4 +101,4 @@ warmup结束之后，开始推理librispeech_asr_dummy数据集，推理过程 | 模型 | 芯片 | 平均E2E时间 | WER | |---------|------------|----------|-------| - | whisper base | 800I A2 64G | 71.73ms | 8.21% | + | whisper base | 800I A2 64G | 65.82ms | 8.21% | -- Gitee From cee6c09a38d5423b9071fe028529855722bc0cfb Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Mon, 13 Oct 2025 11:29:52 +0800 Subject: [PATCH 2/7] 1 --- ACL_PyTorch/built-in/audio/whisper/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ACL_PyTorch/built-in/audio/whisper/README.md b/ACL_PyTorch/built-in/audio/whisper/README.md index 024d54530d..fd6a094b9b 100644 --- a/ACL_PyTorch/built-in/audio/whisper/README.md +++ b/ACL_PyTorch/built-in/audio/whisper/README.md @@ -101,4 +101,4 @@ warmup结束之后，开始推理librispeech_asr_dummy数据集，推理过程 | 模型 | 芯片 | 平均E2E时间 | WER | |---------|------------|----------|-------| - | whisper base | 800I A2 64G | 65.82ms | 8.21% | + | whisper base | 800I A2 64G | 67.68ms | 8.21% | -- Gitee From fea13f9c0eac0ab31de5da499c8bad4f4e535d78 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Tue, 14 Oct 2025 09:34:19 +0800 Subject: [PATCH 3/7] fix patch bug --- ACL_PyTorch/built-in/ocr/MinerU/README.md | 10 +++++++--- ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch | 3 ++- ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch | 3 ++- 3 files changed, 11 insertions(+), 5 deletions(-) diff --git a/ACL_PyTorch/built-in/ocr/MinerU/README.md b/ACL_PyTorch/built-in/ocr/MinerU/README.md index cdf01eb0ea..d4306ded69 100644 --- a/ACL_PyTorch/built-in/ocr/MinerU/README.md +++ b/ACL_PyTorch/built-in/ocr/MinerU/README.md @@ -154,7 +154,8 @@ python3 infer.py --data_path=OmniDocBench_dataset --model_source=local 使用`OmniDocBench`数据集配套评测代码测试精度。 1. 推理结果整理 -将解析结果文件夹中的markdown文件整理放置于同一目录，本例将所有markdown文件存放于OmniDocBench_dataset目录下的results_md文件夹 + + 将解析结果文件夹中的markdown文件整理放置于同一目录，本例将所有markdown文件存放于OmniDocBench_dataset目录下的results_md文件夹 ``` cp OmniDocBench_dataset/output/*/auto/*.md OmniDocBench_dataset/results_md/ ``` @@ -164,13 +165,15 @@ python3 infer.py --data_path=OmniDocBench_dataset --model_source=local ``` git clone https://github.com/opendatalab/OmniDocBench.git cd OmniDocBench + git reset --hard dc96d812d219960773399c02ae8f89e4706120d4 conda create -n omnidocbench python=3.10 conda activate omnidocbench pip install -r requirements.txt ``` 3. 测评配置修改 -修改`OmniDocBench`测评代码中的config文件，具体来说，我们使用端到端测评配置，修改configs/end2end.yaml文件中的ground_truth的data_path为下载的OmniDocBench.json路径，修改prediction的data_path中提供整理的推理结果的文件夹路径，如下： + + 修改`OmniDocBench`测评代码中的config文件，具体来说，我们使用端到端测评配置，修改configs/end2end.yaml文件中的ground_truth的data_path为下载的OmniDocBench.json路径，修改prediction的data_path中提供整理的推理结果的文件夹路径，如下： ``` # -----以下是需要修改的部分 ----- dataset: @@ -182,7 +185,8 @@ python3 infer.py --data_path=OmniDocBench_dataset --model_source=local ``` 4. 精度测量结果 -配置好config文件后，只需要将config文件作为参数传入，运行以下代码即可进行评测： + + 配置好config文件后，只需要将config文件作为参数传入，运行以下代码即可进行评测： ``` python pdf_validation.py --config ./configs/end2end.yaml ``` diff --git a/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch b/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch index 7cf22c0b32..9b526e909e 100644 --- a/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch +++ b/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch @@ -47,4 +47,5 @@ diff -ruN doclayout_yolo-0.0.4/doclayout_yolo/utils/tal.py doclayout_yolo-0.0.4_ - stride_tensor.append(torch.full((h * w, 1), stride, dtype=dtype, device=device)) + # stride_tensor.append(torch.full((h * w, 1), stride, dtype=dtype, device=device)) + stride_tensor.append(torch.ones((h * w, 1), dtype=dtype, device=device)*stride) - return torch.cat(anchor_points), torch.cat(stride_tensor) \ No newline at end of file + return torch.cat(anchor_points), torch.cat(stride_tensor) + \ No newline at end of file diff --git a/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch b/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch index 4fab87d605..70feba449f 100644 --- a/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch +++ b/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch @@ -74,4 +74,5 @@ diff -ruN ultralytics-8.3.193/ultralytics/utils/tal.py ultralytics_/ultralytics/ - stride_tensor.append(torch.full((h * w, 1), stride, dtype=dtype, device=device)) + # stride_tensor.append(torch.full((h * w, 1), stride, dtype=dtype, device=device)) + stride_tensor.append(torch.ones((h * w, 1), dtype=dtype, device=device)*stride) - return torch.cat(anchor_points), torch.cat(stride_tensor) \ No newline at end of file + return torch.cat(anchor_points), torch.cat(stride_tensor) + \ No newline at end of file -- Gitee From 9e6c195bb87f9b7da748d0617173aa3753495b58 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Tue, 14 Oct 2025 10:23:29 +0800 Subject: [PATCH 4/7] 1 --- ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch | 1 - ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch | 1 - 2 files changed, 2 deletions(-) diff --git a/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch b/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch index 9b526e909e..b5fd6669aa 100644 --- a/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch +++ b/ACL_PyTorch/built-in/ocr/MinerU/doclayout_yolo.patch @@ -48,4 +48,3 @@ diff -ruN doclayout_yolo-0.0.4/doclayout_yolo/utils/tal.py doclayout_yolo-0.0.4_ + # stride_tensor.append(torch.full((h * w, 1), stride, dtype=dtype, device=device)) + stride_tensor.append(torch.ones((h * w, 1), dtype=dtype, device=device)*stride) return torch.cat(anchor_points), torch.cat(stride_tensor) - \ No newline at end of file diff --git a/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch b/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch index 70feba449f..5511fa6a9e 100644 --- a/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch +++ b/ACL_PyTorch/built-in/ocr/MinerU/ultralytics.patch @@ -75,4 +75,3 @@ diff -ruN ultralytics-8.3.193/ultralytics/utils/tal.py ultralytics_/ultralytics/ + # stride_tensor.append(torch.full((h * w, 1), stride, dtype=dtype, device=device)) + stride_tensor.append(torch.ones((h * w, 1), dtype=dtype, device=device)*stride) return torch.cat(anchor_points), torch.cat(stride_tensor) - \ No newline at end of file -- Gitee From 796693c60d4941f981ceb67402728c9ea7e8a611 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Tue, 14 Oct 2025 14:46:33 +0800 Subject: [PATCH 5/7] 1 --- ACL_PyTorch/built-in/ocr/MinerU/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ACL_PyTorch/built-in/ocr/MinerU/README.md b/ACL_PyTorch/built-in/ocr/MinerU/README.md index d4306ded69..af345076b8 100644 --- a/ACL_PyTorch/built-in/ocr/MinerU/README.md +++ b/ACL_PyTorch/built-in/ocr/MinerU/README.md @@ -181,7 +181,7 @@ python3 infer.py --data_path=OmniDocBench_dataset --model_source=local ground_truth: data_path: ../OmniDocBench_dataset/OmniDocBench.json prediction: - data_path: ../OmniDocBench_dataset/result_md + data_path: ../OmniDocBench_dataset/results_md ``` 4. 精度测量结果 -- Gitee From adbd35983673cf7084a4c27b8921f545f28e2602 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Tue, 14 Oct 2025 19:26:36 +0800 Subject: [PATCH 6/7] 1 --- ACL_PyTorch/README.md | 61 +++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 59 insertions(+), 2 deletions(-) diff --git a/ACL_PyTorch/README.md b/ACL_PyTorch/README.md index 51cd1c0085..1651f4ee21 100755 --- a/ACL_PyTorch/README.md +++ b/ACL_PyTorch/README.md @@ -3,7 +3,7 @@

如何贡献

在开始贡献之前，请先阅读CONTRIBUTING。 谢谢！

-

目前ACL_PyTorch仓库已有模型398个

+

目前ACL_PyTorch仓库已有模型401个

注意：
在提交新模型时，请加上模型ID用于区分，为防止重复提交模型，请执行脚本get_modelID.py，该脚本会自动检索ACL_PyTorch仓库中所有与您提交模型相关的已有模型，请自行查看脚本给出的链接，如果均不同，则可以输入1或true用于获取模型ID。由于该脚本使用正则匹配，后续新模型刷新到主页需要添加README内容时，格式请参考其余模型，并且同步刷新上文模型数量。脚本执行方式如下：
@@ -4654,6 +4654,7 @@ python3 get_modelID.py --model your_model_name

ROC_AUC mel_loss 300I Pro + 800I A2 100313 @@ -4824,7 +4825,7 @@ python3 get_modelID.py --model your_model_name

多尺度 - 100321 + 100409 whisper @@ -4837,6 +4838,34 @@ python3 get_modelID.py --model your_model_name

67.32(bs1) bs x 80 x 3000 + + 100410 + + CosyVoice2 + + 代码仓提供 + + + + + 0.75 + 0.28 + 多尺度 + + + 100411 + + whisperx + + librispeech dev clean + + 0.050 + + + + 70（转录比） + 多尺度 +

Knowledge

@@ -5258,6 +5287,34 @@ python3 get_modelID.py --model your_model_name

多尺度 +

OCR

+ + + + + + + + + + + + + + + + + + + + + + +
IDNameDataset精度300I Pro最优性能（对应bs）输入shape
overall_ENoverall_CH
100408 + + MinerU-ocr + OmniDocBench0.15880.2527多尺度
+

RL

-- Gitee From d0340be1a65cd893edb92d09e2b20420d975fbf8 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E9=82=B9=E4=BA=A6=E8=88=9F?= Date: Tue, 14 Oct 2025 19:31:17 +0800 Subject: [PATCH 7/7] 1 --- ACL_PyTorch/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/ACL_PyTorch/README.md b/ACL_PyTorch/README.md index 1651f4ee21..2f625f52fb 100755 --- a/ACL_PyTorch/README.md +++ b/ACL_PyTorch/README.md @@ -5295,6 +5295,7 @@ python3 get_modelID.py --model your_model_name

 + @@ -5310,6 +5311,7 @@ python3 get_modelID.py --model your_model_name

+ -- Gitee
Dataset 精度 300I Pro最优性能（对应bs）800I A2最优性能（对应bs） 输入shape
OmniDocBench 0.1588 0.2527 多尺度