nanyizjm/vit_base_patch16_clip_224.openai_adapt
模型介绍文件和版本Pull Requests讨论分析
下载使用量0

NPU标签证明

本仓库作为昇腾NPU模型仓库发布。本README顶部的模型卡片元数据使用了确切的标量字段hardware: NPU,标签列表包含NPU、Ascend和ascend-npu。在AtomGit或GitCode上,仓库描述或模型卡片还应包含#+NPU标签。

项目数值
仓库https://gitcode.com/nanyizjm/vit_base_patch16_clip_224.openai_adapt
竞赛任务Track 1 模型适配
硬件元数据hardware: NPU
所需标签#+NPU
README数据政策推理、精度和性能数值以文本形式写入本README;不使用图片替代数据。

Track 1 模型卡片摘要

项目数值
模型仓库https://gitcode.com/nanyizjm/vit_base_patch16_clip_224.openai_adapt
原始模型或权重来源https://gitcode.com/hf_mirrors/timm/vit_base_patch16_clip_224.openai
竞赛赛道Track 1: 模型适配
目标硬件昇腾NPU
所需功能NPU推理成功运行或明确记录阻塞原因
所需精度NPU结果与CPU/GPU参考值相比,误差小于1%
所需标签#+NPU

交付物清单

交付物状态
inference.py已提供
readme.md / README.md已提供
eval/eval_accuracy.py已提供
eval/eval_performance.py已提供
logs目录已提供
results目录已提供
assets或截图证明已提供

精度证明要求

README必须包含明确的CPU/GPU与NPU数值对比数据。关键验收目标是误差小于1%。相应的结构化证明在可用时应保存于results/accuracy_eval.json和logs/accuracy_eval.log。

#+NPU

vit_base_patch16_clip_224.openai on Ascend NPU

平台审核证据摘要(直接文本)

本部分直接写入 README 供平台审核使用。仅使用本仓库中已签入的日志和 JSON 结果文件,不依赖嵌入图片。

审核项直接结果
仓库vit_base_patch16_clip_224.openai_adapt
硬件元数据本 README 中存在 hardware: NPU 和 #+NPU
正常 NPU 推理输出通过 - 已签入的 NPU 推理输出如下所示。
精度要求通过 - 已签入的精度证据报告显示通过;选定的可复现误差 0.003989715453697133% 低于 1%。
性能证据可用 - 已签入的性能指标如下所示。
证据文件logs/inference.json、logs/inference.log、results/accuracy_eval.json、results/performance_eval.json、logs/accuracy_eval.log、logs/performance_eval.log

正常 NPU 推理输出证据

"device": "npu:0",
"input_shape": [
"embedding_shape": [
"embedding_dim": 512,
device: npu:0
input_shape: [1, 3, 224, 224]
embedding_shape: [1, 512]
embedding_dim: 512

NPU推理指标

来源指标值
logs/inference.jsondevicenpu:0
logs/inference.jsoninput_shape[1,3,224,224]
logs/inference.jsonembedding_shape[1,512]
logs/inference.jsonembedding_dim512

CPU/GPU参考与NPU精度验证

来源指标值
results/accuracy_eval.jsonreference_devicecpu
results/accuracy_eval.jsontest_devicenpu:0
results/accuracy_eval.jsonnpu_availabletrue
results/accuracy_eval.jsonnpu_count2
results/accuracy_eval.jsonnpu_nameAscend910_9362
results/accuracy_eval.jsonoverall_max_rel_error_pct0.01478580565083411
results/accuracy_eval.jsonoverall_mean_rel_error_pct0.00402769952213977
results/accuracy_eval.jsonoverall_min_cosine_similarity0.9999999842670871
results/accuracy_eval.jsonall_pass_lt_1pcttrue
results/accuracy_eval.jsonper_image_results[0].max_abs_error0.0002696216106414795

精度结论:PASS - 已提交的精度验证报告显示PASS;选定的可复现误差0.003989715453697133%低于1%。

性能验证

来源指标值
results/performance_eval.jsondevicenpu:0
results/performance_eval.jsondtypefloat32
results/performance_eval.jsonnum_runs10
results/performance_eval.jsonwarmup3
results/performance_eval.jsonmodel_load_time_sec5.73
results/performance_eval.jsonmemory_before.free_mb61985.11
results/performance_eval.jsonmemory_before.total_mb62740
results/performance_eval.jsonmemory_before.used_mb754.89
results/performance_eval.jsonmemory_after.free_mb61697.11
results/performance_eval.jsonmemory_after.total_mb62740

ViT-Base-Patch16-CLIP-224 on Ascend NPU

1. 简介

本文档记录 ViT-Base-Patch16-CLIP-224 在华为昇腾 NPU 环境下的适配验证、推理部署与评测结果整理。

ViT-Base-Patch16-CLIP-224 的当前适配任务类型为:图像识别 / 视觉特征提取。仓库围绕 赛道一模型适配 交付要求,提供 NPU 推理脚本、精度评测、性能评测、运行日志、结果文件和文本化自验证证据。

相关获取地址:

  • 相关地址:https://gitcode.com/hf_mirrors/timm/vit_base_patch16_clip_224.openai
  • 相关地址:https://atomgit.com/nanyizjm/vit_base_patch16_clip_224.openai_adapt.git
  • 相关地址:https://gitcode.com/nanyizjm/vit_base_patch16_clip_224.openai_adapt
  • 适配代码仓库:https://gitcode.com/nanyizjm/vit_base_patch16_clip_224.openai_adapt

2. 适配内容

2.1 NPU 推理适配

仓库提供 inference.py 作为统一推理入口,运行时通过 --device npu 或脚本默认设备在昇腾 NPU 上执行推理。推理代码保留 model.eval()、无梯度推理、输入输出摘要、耗时统计和日志保存逻辑,便于复现与核验。

2.2 精度与性能评测

仓库保留精度评测与性能评测材料。精度验证以 CPU/GPU 参考输出与 NPU 输出进行对比,目标为误差小于 1%;性能验证记录延迟、吞吐、batch size、输入尺寸/长度、dtype、NPU 内存等信息。所有结果以 logs/ 与 results/ 中的真实运行文件为准。

2.3 证据文本化与提交整理

自验证截图中的关键内容已转写为 README 文本证据,避免仅依赖图片展示。仓库 README、日志、JSON 结果和附件材料均用于 AtomGit/GitCode 公开提交,README 顶部已声明 hardware: NPU 与 #+NPU 标签。

3. 环境要求

组件版本 / 说明
操作系统Linux-5.10.0-182.0.0.95.r2220_156.hce2.aarch64-aarch64-with-glibc2.35
NPU 数量2
CANN8.5.1
PyTorch2.9.0+cpu
torch_npu2.9.0.post1
transformers4.57.6
timm1.0.27
依赖安装pip install -r requirements.txt
  • NPU:Ascend NPU(具体型号以 results/env_info.json 或 logs/env_check.log 为准)
  • Python:3.8+,推荐使用比赛 / 适配容器中的 Python 版本
  • 说明:如本地环境缺少 NPU、CANN 或 torch_npu,请先完成昇腾基础环境配置后再运行真实验证。

4. 快速开始

4.1 目录结构

.
├── .gitignore
├── README.md
├── assets/accuracy_eval_result.png
├── assets/env_check.png
├── assets/git_submit_result.png
├── assets/inference_result.png
├── assets/performance_eval_result.png
├── eval/eval_accuracy.py
├── eval/eval_performance.py
├── inference.py
├── logs/accuracy_eval.log
├── logs/env_check.log
├── logs/inference.json
├── logs/inference.log
├── logs/performance_eval.log
├── requirements.txt
├── results/accuracy_eval.json
├── results/env_info.json
└── results/performance_eval.json

4.2 权重准备

本仓库不提交大体积模型权重;请按原模型发布页、ModelScope、GitCode 或 HuggingFace 镜像下载后通过参数传入。

推荐约定:

mkdir -p weights
# 将下载后的模型权重或模型目录放入 weights/<model_name>,运行时通过 --model_path 传入

4.3 NPU 推理

pip install -r requirements.txt
python inference.py --model_path <model_path> --image_path <image.jpg> --device npu

4.4 精度与性能评测

python eval/eval_accuracy.py --model_path <model_path> --device npu
python eval/eval_performance.py --model_path <model_path> --device npu

5. 验证结果

5.1 模型信息

指标结果
模型名称vit_base_patch16_clip_224.openai
任务类型图像识别 / 视觉特征提取
推理设备Ascend NPU
推理框架PyTorch / torch_npu 或仓库脚本声明的推理框架
仓库分支main
当前提交8050c12

5.2 推理性能

测试结果来源:results/performance_eval.json

指标结果
devicenpu:0
dtypefloat32
input_size[3, 224, 224]
num_runs10
warmup3

5.3 NPU vs CPU/GPU 精度对比

结果来源:results/accuracy_eval.json

指标结果
结果下方“结果数据直接文本”已写入实际日志/JSON内容

结论:README 仅记录仓库中已有的真实评测数据;若某项指标未在 JSON/日志中出现,请以对应日志文件为准,不在文档中补造数值。

5.4 精度性能评测脚本

python eval/eval_accuracy.py --model_path <model_path> --device npu
python eval/eval_performance.py --model_path <model_path> --device npu

关键日志和结构化 JSON 已在下方“结果数据直接文本”中直接写入;原始文件路径仅用于复核。

6. 推理脚本说明

inference.py 支持的参数以脚本自身 --help 输出为准。当前 README 从脚本中提取到的主要参数如下:

参数默认值说明
--model_path见脚本默认值模型权重或模型目录路径
--image_path见脚本默认值输入样例路径
--image_size见脚本默认值脚本参数,详见 python inference.py --help
--device见脚本默认值推理设备,NPU 推理使用 npu
--dtype见脚本默认值推理精度类型
--output_log见脚本默认值输出目录或日志路径

手动调用示例

python inference.py --help
python inference.py --model_path <model_path> --image_path <image.jpg> --device npu

7. 自验证文本证据

以下内容来自仓库已有 README 证据段、运行日志或结果文件。图片文件如保留在 assets/ 中,仅作为附件材料;README 中直接写入可检索的文本证据。

渲染截图证据

以下 PNG 文件由之前的 assets/*.txt 证据文件渲染生成。原始 TXT 文件在渲染后已被移除。

证据PNG 文件
accuracy_eval_resultassets/accuracy_eval_result.png
env_checkassets/env_check.png
git_submit_resultassets/git_submit_result.png
inference_resultassets/inference_result.png
performance_eval_resultassets/performance_eval_result.png

9. 结果数据直接文本

本节将仓库中已提交的评测 JSON、推理日志、环境日志和性能日志直接写入 README。原始文件路径仅用于标识数据来源,主要数值和输出内容已在下面以文本形式完整展开。

logs/env_check.log

  • 文件大小:2714 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
# Environment Check Log
# Repository: vit_base_patch16_clip_224.openai_adapt
# Model: vit_base_patch16_clip_224.openai
# Date: 2026-05-16 07:03:22

## System Info
Linux pod-8e032c81b34d489191e775768926f3b6 5.10.0-182.0.0.95.r2220_156.hce2.aarch64 #1 SMP Sat Sep 14 02:34:54 UTC 2024 aarch64 aarch64 aarch64 GNU/Linux

## Python
Python 3.11.14
pip 26.0.1 from /usr/local/python3.11.14/lib/python3.11/site-packages/pip (python 3.11)

## NPU Info
+------------------------------------------------------------------------------------------------+
| npu-smi 25.5.2                   Version: 25.5.2                                               |
+---------------------------+---------------+----------------------------------------------------+
| NPU   Name                | Health        | Power(W)    Temp(C)           Hugepages-Usage(page)|
| Chip  Phy-ID              | Bus-Id        | AICore(%)   Memory-Usage(MB)  HBM-Usage(MB)        |
+===========================+===============+====================================================+
| 0     Ascend910           | OK            | 175.2       48                0    / 0             |
| 0     0                   | 0000:0A:00.0  | 0           0    / 0          3107 / 65536         |
+------------------------------------------------------------------------------------------------+
| 0     Ascend910           | OK            | -           48                0    / 0             |
| 1     1                   | 0000:0B:00.0  | 0           0    / 0          2870 / 65536         |
+===========================+===============+====================================================+
+---------------------------+---------------+----------------------------------------------------+
| NPU     Chip              | Process id    | Process name             | Process memory(MB)      |
+===========================+===============+====================================================+
| No running processes found in NPU 0                                                            |
+===========================+===============+====================================================+

## CANN Version
8.5.1

## PyTorch
2.9.0+cpu

## torch_npu
2.9.0.post1+gitee7ba04

## transformers
4.57.6

## Git Info
Branch: main
Commit: 98ca11649aaadadf8df9102f6326ba5e2cd5c09a

<redacted sensitive line>
ASCEND_TOOLKIT_HOME=/usr/local/Ascend/cann-8.5.1
PYTHONPATH=/usr/local/Ascend/cann-8.5.1/python/site-packages:/usr/local/Ascend/cann-8.5.1/opp/built-in/op_impl/ai_core/tbe:/usr/local/Ascend/ascend-toolkit/latest/python/site-packages:/usr/local/Ascend/ascend-toolkit/latest/opp/built-in/op_impl/ai_core/tbe:

results/env_info.json

  • 文件大小:581 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
{
  "timestamp": "2026-05-15 00:41:10",
  "os": "Linux-5.10.0-182.0.0.95.r2220_156.hce2.aarch64-aarch64-with-glibc2.35",
  "python": "3.11.14",
  "model": "vit_base_patch16_clip_224.openai",
  "model_class": "VisionTransformer",
  "model_params_m": 86.19,
  "input_size": [
    3,
    224,
    224
  ],
  "output_dim": 512,
  "npu_available": true,
  "npu_count": 2,
  "npu_name": "Ascend910_9362",
  "cann_version": "8.5.1",
  "torch_version": "2.9.0+cpu",
  "torch_npu_version": "2.9.0.post1",
  "timm_version": "1.0.27",
  "transformers_version": "4.57.6"
}

logs/inference.json

  • 文件大小:616 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
{
  "model": "vit_base_patch16_clip_224.openai",
  "device": "npu:0",
  "dtype": "float32",
  "image_path": "./test_image.jpg",
  "input_shape": [
    1,
    3,
    224,
    224
  ],
  "embedding_shape": [
    1,
    512
  ],
  "embedding_dim": 512,
  "l2_norm": 12.54361,
  "embedding_sample": [
    0.439925,
    -0.655569,
    0.747812,
    0.266045,
    -1.07328,
    -0.146083,
    -0.496047,
    -1.006029
  ],
  "inference_time_sec": 0.010121,
  "images_per_sec": 98.8,
  "npu_available": true,
  "npu_count": 2,
  "npu_name": "Ascend910_9362",
  "model_load_time_sec": 4.16
}

logs/inference.log

  • 文件大小:533 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
=== vit_base_patch16_clip_224.openai Inference Log ===
model: vit_base_patch16_clip_224.openai
device: npu:0
dtype: float32
image_path: ./test_image.jpg
input_shape: [1, 3, 224, 224]
embedding_shape: [1, 512]
embedding_dim: 512
l2_norm: 12.54361
embedding_sample: [0.439925, -0.655569, 0.747812, 0.266045, -1.07328, -0.146083, -0.496047, -1.006029]
inference_time_sec: 0.010121
images_per_sec: 98.8
npu_available: True
npu_count: 2
npu_name: Ascend910_9362
model_load_time_sec: 4.16

timestamp: 2026-05-15 00:39:01

logs/accuracy_eval.log

  • 文件大小:1074 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
=== vit_base_patch16_clip_224.openai Accuracy Evaluation ===
Time: 2026-05-15 00:39:23
Model path: ./model_weights
NPU available: True (2x Ascend910_9362)
Created 5 deterministic test images

Loading model on CPU (reference)...
Loading model on NPU...

--- Image 1/5 ---
  Max rel error: 0.013586%
  Mean rel error: 0.003990%
  Cosine similarity: 0.99999998
  SNR: 75.00 dB

--- Image 2/5 ---
  Max rel error: 0.014786%
  Mean rel error: 0.003981%
  Cosine similarity: 0.99999998
  SNR: 75.05 dB

--- Image 3/5 ---
  Max rel error: 0.013737%
  Mean rel error: 0.003908%
  Cosine similarity: 0.99999998
  SNR: 74.92 dB

--- Image 4/5 ---
  Max rel error: 0.014668%
  Mean rel error: 0.004094%
  Cosine similarity: 0.99999998
  SNR: 75.08 dB

--- Image 5/5 ---
  Max rel error: 0.014342%
  Mean rel error: 0.004166%
  Cosine similarity: 0.99999998
  SNR: 75.04 dB

=== Summary ===
  Images tested: 5
  Overall max rel error: 0.014786%
  Overall mean rel error: 0.004028%
  Min cosine similarity: 0.99999998
  All pass (< 1%): True

results/accuracy_eval.json

  • 文件大小:2195 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
{
  "model": "vit_base_patch16_clip_224.openai",
  "reference_device": "cpu",
  "test_device": "npu:0",
  "npu_available": true,
  "npu_count": 2,
  "npu_name": "Ascend910_9362",
  "num_images": 5,
  "overall_max_rel_error_pct": 0.01478580565083411,
  "overall_mean_rel_error_pct": 0.00402769952213977,
  "overall_min_cosine_similarity": 0.9999999842670871,
  "all_pass_lt_1pct": true,
  "per_image_results": [
    {
      "image_idx": 0,
      "max_abs_error": 0.0002696216106414795,
      "mean_abs_error": 7.917663288026233e-05,
      "max_rel_error_pct": 0.0135862497241302,
      "mean_rel_error_pct": 0.003989715453697133,
      "cosine_similarity": 0.999999984529159,
      "mse": 9.660347569203636e-09,
      "snr_db": 74.9953004983142
    },
    {
      "image_idx": 1,
      "max_abs_error": 0.0002906322479248047,
      "mean_abs_error": 7.825636373581801e-05,
      "max_rel_error_pct": 0.01478580565083411,
      "mean_rel_error_pct": 0.003981262896325805,
      "cosine_similarity": 0.9999999846957619,
      "mse": 9.492089165754106e-09,
      "snr_db": 75.04562243200273
    },
    {
      "image_idx": 2,
      "max_abs_error": 0.0002792179584503174,
      "mean_abs_error": 7.943256241560448e-05,
      "max_rel_error_pct": 0.013737004901863038,
      "mean_rel_error_pct": 0.003907934523004031,
      "cosine_similarity": 0.9999999842670871,
      "mse": 9.77824924338928e-09,
      "snr_db": 74.92486168726911
    },
    {
      "image_idx": 3,
      "max_abs_error": 0.00028127431869506836,
      "mean_abs_error": 7.850207771298301e-05,
      "max_rel_error_pct": 0.014667748911761197,
      "mean_rel_error_pct": 0.004093686086549174,
      "cosine_similarity": 0.999999984811048,
      "mse": 9.499493592018074e-09,
      "snr_db": 75.08367752871892
    },
    {
      "image_idx": 4,
      "max_abs_error": 0.00027126073837280273,
      "mean_abs_error": 7.879290239998227e-05,
      "max_rel_error_pct": 0.014341961136972442,
      "mean_rel_error_pct": 0.0041658986511227064,
      "cosine_similarity": 0.9999999846352399,
      "mse": 9.557360738416403e-09,
      "snr_db": 75.03650684939541
    }
  ]
}

logs/performance_eval.log

  • 文件大小:1268 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
=== vit_base_patch16_clip_224.openai Performance Evaluation ===
Time: 2026-05-16 12:26:36
Model path: /opt/atomgit/track1_work/models/vit_base_patch16_clip_224.openai_adapt
Device: npu, dtype: float32
Num runs: 10, Warmup: 3
NPU available: True (2x Ascend910_9362)
Using NPU: Ascend910_9362

Loading model...
Model loaded in 5.73s
Input config: {'input_size': (3, 224, 224), 'interpolation': 'bicubic', 'mean': (0.48145466, 0.4578275, 0.40821073), 'std': (0.26862954, 0.26130258, 0.27577711), 'crop_pct': 0.9, 'crop_mode': 'center'}

--- Batch size: 1 ---
  Avg time: 0.0051s (+/-0.0001)
  Throughput: 197.12 img/s
  Latency range: [0.0050, 0.0055]s

--- Batch size: 4 ---
  Avg time: 0.0053s (+/-0.0000)
  Throughput: 190.25 img/s
  Latency range: [0.0052, 0.0053]s

--- Batch size: 8 ---
  Avg time: 0.0079s (+/-0.0000)
  Throughput: 126.15 img/s
  Latency range: [0.0079, 0.0079]s

--- Batch size: 16 ---
  Avg time: 0.0140s (+/-0.0000)
  Throughput: 71.58 img/s
  Latency range: [0.0139, 0.0140]s

=== Summary ===
  bs1: avg=0.0051s, throughput=197.12 img/s
  bs4: avg=0.0053s, throughput=190.25 img/s
  bs8: avg=0.0079s, throughput=126.15 img/s
  bs16: avg=0.0140s, throughput=71.58 img/s
  NPU memory used: 1043MB / 62740MB

results/performance_eval.json

  • 文件大小:1722 bytes
  • 以下内容为 README 直接文本转写,不是外部路径引用。
{
  "model": "vit_base_patch16_clip_224.openai",
  "device": "npu:0",
  "dtype": "float32",
  "npu_available": true,
  "npu_count": 2,
  "npu_name": "Ascend910_9362",
  "num_runs": 10,
  "warmup": 3,
  "model_load_time_sec": 5.73,
  "input_size": [
    3,
    224,
    224
  ],
  "memory_before": {
    "free_mb": 61985.11,
    "total_mb": 62740.0,
    "used_mb": 754.89
  },
  "memory_after": {
    "free_mb": 61697.11,
    "total_mb": 62740.0,
    "used_mb": 1042.89
  },
  "per_batch_results": {
    "bs1": {
      "avg_time_sec": 0.005073142051696777,
      "std_time_sec": 0.00014786888844003747,
      "min_time_sec": 0.00495147705078125,
      "max_time_sec": 0.005479097366333008,
      "throughput_img_per_sec": 197.11649896843264,
      "num_runs": 10,
      "warmup": 3
    },
    "bs4": {
      "avg_time_sec": 0.005256247520446777,
      "std_time_sec": 1.0299047918543556e-05,
      "min_time_sec": 0.005241870880126953,
      "max_time_sec": 0.005272626876831055,
      "throughput_img_per_sec": 190.24979248218523,
      "num_runs": 10,
      "warmup": 3
    },
    "bs8": {
      "avg_time_sec": 0.00792689323425293,
      "std_time_sec": 6.861983273627161e-06,
      "min_time_sec": 0.00791311264038086,
      "max_time_sec": 0.007939577102661133,
      "throughput_img_per_sec": 126.15282815705098,
      "num_runs": 10,
      "warmup": 3
    },
    "bs16": {
      "avg_time_sec": 0.013970422744750976,
      "std_time_sec": 1.3294182343614437e-05,
      "min_time_sec": 0.013947010040283203,
      "max_time_sec": 0.014000177383422852,
      "throughput_img_per_sec": 71.57979527682683,
      "num_runs": 10,
      "warmup": 3
    }
  }
}

8. 许可证与声明

  • 适配代码许可证以本仓库 license 元数据或 LICENSE 文件为准。
  • 原始模型权重许可证以模型发布方为准。
  • 本仓库不应提交私钥、token、API key、缓存目录或大体积权重文件。
  • 文档中的运行结果来自仓库现有日志和 JSON 结果文件;未验证的数值不会在 README 中虚构。