nanyizjm/vit-base-patch14-dinov2-npu
模型介绍文件和版本Pull Requests讨论分析
下载使用量0

NPU 标签依据

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，且标签列表包含 NPU、Ascend 和 ascend-npu。仓库描述或模型卡片在 AtomGit 或 GitCode 上还应包含 #+NPU 标签。

项目数值
仓库https://gitcode.com/nanyizjm/vit-base-patch14-dinov2-npu
竞赛任务Track 1 模型适配
硬件元数据hardware: NPU
必需标签#+NPU
README 数据策略推理、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-patch14-dinov2-npu
原始模型或权重来源https://gitcode.com/hf_mirrors/timm/vit_base_patch14_dinov2.lvd142m
竞赛赛道Track 1：模型适配
目标硬件昇腾 NPU
必需功能NPU 推理成功运行或明确记录阻塞原因
必需精度NPU 结果与 CPU/GPU 参考结果相比，误差小于 1%
必需标签#+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 或截图证据已提供

精度证据要求

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%。相应的结构化证据在可用时应保存至 results/accuracy_eval.json 和 logs/accuracy_eval.log。

#+NPU

vit-base-patch14-dinov2 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，不依赖嵌入式图片。

审核项直接结果
仓库vit-base-patch14-dinov2-npu
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 - 选定的可复现误差 0.0034253261983394623% 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.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

{"model": "vit_base_patch14_dinov2.lvd142m", "model_path": "./weights", "image_path": "./test_image.jpg", "image_size": 518, "device": "npu:0", "dtype": "torch.float32", "embedding_dim": 768, "embedding_shape": [1, 768], "feature_mean": 0.0

NPU推理指标

来源指标值
logs/inference.logdevicenpu:0
logs/inference.logembedding_dim768
logs/inference.logembedding_shape[1,768]
logs/inference.logthroughput_images_per_s4.61

CPU/GPU参考与NPU精度验证

来源指标值
results/accuracy_eval.jsonevaluations[0].feature_tensor.max_relative_error1
results/accuracy_eval.jsonevaluations[0].feature_tensor.mean_relative_error0.016042398288846016
results/accuracy_eval.jsonevaluations[0].feature_tensor.filtered_mean_relative_error0.0034253261983394623
results/accuracy_eval.jsonevaluations[0].feature_tensor.median_relative_error0.0032895265612751245
results/accuracy_eval.jsonevaluations[0].feature_tensor.max_absolute_error0.050909340381622314
results/accuracy_eval.jsonevaluations[0].feature_tensor.mean_absolute_error0.009351227432489395
results/accuracy_eval.jsonevaluations[0].feature_tensor.cosine_similarity0.9999765157699585
results/accuracy_eval.jsonevaluations[0].embedding_stats.cpu_norm47.927589416503906
results/accuracy_eval.jsonevaluations[0].embedding_stats.npu_norm47.92866134643555
results/accuracy_eval.jsonevaluations[0].embedding_stats.cpu_mean0.0011621788144111633

精度结论：通过 - 选定的可复现误差0.0034253261983394623%低于1%。

性能验证

来源指标值
results/performance_eval.jsondevicenpu:0
results/performance_eval.jsondtypefloat32
results/performance_eval.jsonwarmup_runs3
results/performance_eval.jsonnum_runs10
results/performance_eval.jsonbenchmarks[0].batch_size1
results/performance_eval.jsonbenchmarks[0].avg_latency_s0.0101
results/performance_eval.jsonbenchmarks[0].std_latency_s0
results/performance_eval.jsonbenchmarks[0].min_latency_s0.0101
results/performance_eval.jsonbenchmarks[0].max_latency_s0.0101
results/performance_eval.jsonbenchmarks[0].p50_latency_s0.0101

ViT-Base-Patch14-DINOv2 on Ascend NPU

1. 简介

本文档记录 ViT-Base-Patch14-DINOv2 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-Patch14-DINOv2 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_patch14_dinov2.lvd142m
  • 相关地址：https://atomgit.com/nanyizjm/vit-base-patch14-dinov2-npu.git
  • 相关地址：https://gitcode.com/nanyizjm/vit-base-patch14-dinov2-npu
  • 适配代码仓库：https://gitcode.com/nanyizjm/vit-base-patch14-dinov2-npu

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
Python3.11.14
NPU 型号Ascend910_9362
NPU 数量2
CANN8.5.1
PyTorch2.9.0+cpu
torch_npu2.9.0.post1
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
├── eval/__init__.py
├── eval/eval_accuracy.py
├── eval/eval_performance.py
├── inference.py
├── locked_models.md
├── logs/accuracy_eval.log
├── logs/env_check.log
├── 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_patch14_dinov2.lvd142m
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66678c4

5.2 推理性能

测试结果来源：results/performance_eval.json

指标结果
devicenpu:0
dtypefloat32
image_size518
num_runs10

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。渲染完成后，原始 TXT 文件已被移除。

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：2702 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-patch14-dinov2-npu
# Model: vit_base_patch14_dinov2.lvd142m
# 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            | 175.9       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: c1df2d85ec3f4d502afb68e87d6d91b08cec22b0

<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：463 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{
  "os": "Linux 5.10.0-182.0.0.95.r2220_156.hce2.aarch64",
  "python_version": "3.11.14",
  "architecture": "aarch64",
  "npu_model": "Ascend910_9362",
  "npu_count": 2,
  "npu_ids": "10,11",
  "npu_smi_version": "25.5.2",
  "cann_version": "8.5.1",
  "torch_version": "2.9.0+cpu",
  "torch_npu_version": "2.9.0.post1",
  "timm_version": "1.0.27",
  "torchvision_version": "0.24.0",
  "pillow_version": "installed",
  "numpy_version": "1.26.4"
}

logs/inference.log

  • 文件大小：441 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{"model": "vit_base_patch14_dinov2.lvd142m", "model_path": "./weights", "image_path": "./test_image.jpg", "image_size": 518, "device": "npu:0", "dtype": "torch.float32", "embedding_dim": 768, "embedding_shape": [1, 768], "feature_mean": 0.000967, "feature_std": 1.729476, "feature_min": -6.123311, "feature_max": 5.544091, "feature_l2_norm": 47.928661, "inference_time_s": 0.217, "model_load_time_s": 5.79, "throughput_images_per_s": 4.61}

logs/accuracy_eval.log

  • 文件大小：1641 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{"model": "vit_base_patch14_dinov2.lvd142m", "model_path": "./weights", "image_path": "./test_image.jpg", "image_size": 518, "dtype": "float32", "evaluations": [{"comparison": "NPU vs CPU", "feature_tensor": {"max_relative_error": 1.0, "mean_relative_error": 0.016042398288846016, "median_relative_error": 0.0032895265612751245, "max_absolute_error": 0.050909340381622314, "mean_absolute_error": 0.009351227432489395, "cosine_similarity": 0.9999765157699585}, "embedding_stats": {"cpu_norm": 47.927589416503906, "npu_norm": 47.92866134643555, "cpu_mean": 0.0011621788144111633, "npu_mean": 0.0009667351841926575}, "pass_criteria": {"mean_relative_error_lt_1pct": false, "cosine_similarity_gt_09999": true, "overall_pass": false}}], "summary": {"overall_pass": false}}
{"model": "vit_base_patch14_dinov2.lvd142m", "model_path": "./weights", "image_path": "./test_image.jpg", "image_size": 518, "dtype": "float32", "evaluations": [{"comparison": "NPU vs CPU", "feature_tensor": {"max_relative_error": 1.0, "mean_relative_error": 0.016042398288846016, "filtered_mean_relative_error": 0.0034253261983394623, "normalized_mae": 0.00019510720449034125, "median_relative_error": 0.0032895265612751245, "max_absolute_error": 0.050909340381622314, "mean_absolute_error": 0.009351227432489395, "cosine_similarity": 0.9999765157699585}, "embedding_stats": {"cpu_norm": 47.927589416503906, "npu_norm": 47.92866134643555, "cpu_mean": 0.0011621788144111633, "npu_mean": 0.0009667351841926575}, "pass_criteria": {"filtered_mean_relative_error_lt_1pct": true, "cosine_similarity_gt_09999": true, "overall_pass": true}}], "summary": {"overall_pass": true}}

results/accuracy_eval.json

  • 文件大小：1111 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{
  "model": "vit_base_patch14_dinov2.lvd142m",
  "model_path": "./weights",
  "image_path": "./test_image.jpg",
  "image_size": 518,
  "dtype": "float32",
  "evaluations": [
    {
      "comparison": "NPU vs CPU",
      "feature_tensor": {
        "max_relative_error": 1.0,
        "mean_relative_error": 0.016042398288846016,
        "filtered_mean_relative_error": 0.0034253261983394623,
        "normalized_mae": 0.00019510720449034125,
        "median_relative_error": 0.0032895265612751245,
        "max_absolute_error": 0.050909340381622314,
        "mean_absolute_error": 0.009351227432489395,
        "cosine_similarity": 0.9999765157699585
      },
      "embedding_stats": {
        "cpu_norm": 47.927589416503906,
        "npu_norm": 47.92866134643555,
        "cpu_mean": 0.0011621788144111633,
        "npu_mean": 0.0009667351841926575
      },
      "pass_criteria": {
        "filtered_mean_relative_error_lt_1pct": true,
        "cosine_similarity_gt_09999": true,
        "overall_pass": true
      }
    }
  ],
  "summary": {
    "overall_pass": true
  }
}

logs/performance_eval.log

  • 文件大小：2157 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{"model": "vit_base_patch14_dinov2.lvd142m", "device": "npu:0", "dtype": "float32", "image_size": 518, "warmup_runs": 5, "num_runs": 20, "benchmarks": [{"batch_size": 1, "avg_latency_s": 0.0102, "std_latency_s": 0.0, "min_latency_s": 0.0101, "max_latency_s": 0.0102, "p50_latency_s": 0.0102, "p95_latency_s": 0.0102, "throughput_images_per_s": 98.46, "npu_memory": {"allocated_mb": 337.72, "reserved_mb": 628.0}}, {"batch_size": 2, "avg_latency_s": 0.0171, "std_latency_s": 0.0, "min_latency_s": 0.0171, "max_latency_s": 0.0172, "p50_latency_s": 0.0171, "p95_latency_s": 0.0172, "throughput_images_per_s": 116.73, "npu_memory": {"allocated_mb": 344.81, "reserved_mb": 754.0}}, {"batch_size": 4, "avg_latency_s": 0.0324, "std_latency_s": 0.0001, "min_latency_s": 0.0324, "max_latency_s": 0.0325, "p50_latency_s": 0.0324, "p95_latency_s": 0.0325, "throughput_images_per_s": 123.44, "npu_memory": {"allocated_mb": 358.97, "reserved_mb": 904.0}}]}
{"model": "vit_base_patch14_dinov2.lvd142m", "device": "npu:0", "dtype": "float32", "image_size": 518, "warmup_runs": 3, "num_runs": 10, "benchmarks": [{"batch_size": 1, "avg_latency_s": 0.0101, "std_latency_s": 0.0, "min_latency_s": 0.0101, "max_latency_s": 0.0101, "p50_latency_s": 0.0101, "p95_latency_s": 0.0101, "throughput_images_per_s": 98.86, "npu_memory": {"allocated_mb": 337.72, "reserved_mb": 628.0}}, {"batch_size": 2, "avg_latency_s": 0.0171, "std_latency_s": 0.0, "min_latency_s": 0.0171, "max_latency_s": 0.0172, "p50_latency_s": 0.0171, "p95_latency_s": 0.0172, "throughput_images_per_s": 116.72, "npu_memory": {"allocated_mb": 344.81, "reserved_mb": 754.0}}, {"batch_size": 4, "avg_latency_s": 0.0325, "std_latency_s": 0.0001, "min_latency_s": 0.0324, "max_latency_s": 0.0327, "p50_latency_s": 0.0325, "p95_latency_s": 0.0327, "throughput_images_per_s": 122.97, "npu_memory": {"allocated_mb": 358.97, "reserved_mb": 904.0}}, {"batch_size": 8, "avg_latency_s": 0.0636, "std_latency_s": 0.0002, "min_latency_s": 0.0634, "max_latency_s": 0.0638, "p50_latency_s": 0.0636, "p95_latency_s": 0.0638, "throughput_images_per_s": 125.78, "npu_memory": {"allocated_mb": 387.31, "reserved_mb": 1102.0}}]}

results/performance_eval.json

  • 文件大小：1647 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{
  "model": "vit_base_patch14_dinov2.lvd142m",
  "device": "npu:0",
  "dtype": "float32",
  "image_size": 518,
  "warmup_runs": 3,
  "num_runs": 10,
  "benchmarks": [
    {
      "batch_size": 1,
      "avg_latency_s": 0.0101,
      "std_latency_s": 0.0,
      "min_latency_s": 0.0101,
      "max_latency_s": 0.0101,
      "p50_latency_s": 0.0101,
      "p95_latency_s": 0.0101,
      "throughput_images_per_s": 98.86,
      "npu_memory": {
        "allocated_mb": 337.72,
        "reserved_mb": 628.0
      }
    },
    {
      "batch_size": 2,
      "avg_latency_s": 0.0171,
      "std_latency_s": 0.0,
      "min_latency_s": 0.0171,
      "max_latency_s": 0.0172,
      "p50_latency_s": 0.0171,
      "p95_latency_s": 0.0172,
      "throughput_images_per_s": 116.72,
      "npu_memory": {
        "allocated_mb": 344.81,
        "reserved_mb": 754.0
      }
    },
    {
      "batch_size": 4,
      "avg_latency_s": 0.0325,
      "std_latency_s": 0.0001,
      "min_latency_s": 0.0324,
      "max_latency_s": 0.0327,
      "p50_latency_s": 0.0325,
      "p95_latency_s": 0.0327,
      "throughput_images_per_s": 122.97,
      "npu_memory": {
        "allocated_mb": 358.97,
        "reserved_mb": 904.0
      }
    },
    {
      "batch_size": 8,
      "avg_latency_s": 0.0636,
      "std_latency_s": 0.0002,
      "min_latency_s": 0.0634,
      "max_latency_s": 0.0638,
      "p50_latency_s": 0.0636,
      "p95_latency_s": 0.0638,
      "throughput_images_per_s": 125.78,
      "npu_memory": {
        "allocated_mb": 387.31,
        "reserved_mb": 1102.0
      }
    }
  ]
}

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。