nanyizjm/vit_base_patch16_224.dino_adapt
模型介绍文件和版本Pull Requests讨论分析
下载使用量0

#+NPU

NPU 标签证据

本模型仓库明确声明了所需的 NPU 模型卡片标签。

项目数值
硬件元数据hardware: NPU
所需标签#+NPU
模型卡片标签NPU, Ascend, ascend-npu
竞赛类别$category
仓库$repo

vit_base_patch16_224.dino 在昇腾 NPU 上的适配

1. 模型简介

本文档记录 $name 在华为昇腾 NPU 环境下的赛道一模型适配、推理验证、精度验证、性能验证与提交材料整理。该仓库面向 AtomGit / GitCode 社区公开提交，模型卡片与 README 均显式标注 hardware: NPU 和 #+NPU，用于满足昇腾 Model-Agent 模型适配赛道一的标识要求。

项目内容
模型 / 仓库$repo
任务类型图像识别 / 视觉特征提取
赛道赛道一：模型适配
目标硬件昇腾 NPU
提交标签#+NPU
精度要求与 CPU / GPU 参考结果误差 < 1%
结果呈现README 直接写入文本化证据，截图仅作为辅助材料，不替代数据表与日志摘录

2. 适配内容

  • 提供 NPU 推理入口 inference.py，模型路径、输入样例、设备和 dtype 等参数通过命令行传入。
  • 提供精度评测与性能评测脚本，评测结果保存到 logs/ 与 results/。
  • README 中保留推理正常输出、CPU/GPU 与 NPU 精度对比、性能指标、日志路径和结果路径。
  • 不提交大体积权重、缓存目录、私钥、token 或无关临时文件。

3. 交付件自查

交付项路径状态
推理脚本$(System.Collections.Hashtable.path)已提供
部署文档$(System.Collections.Hashtable.path)已提供
精度评测源码$(System.Collections.Hashtable.path)已提供
性能评测源码$(System.Collections.Hashtable.path)已提供
运行日志目录$(System.Collections.Hashtable.path)已提供
结构化结果目录$(System.Collections.Hashtable.path)已提供
自验证截图或文本化证据目录$(System.Collections.Hashtable.path)已提供
依赖说明$(System.Collections.Hashtable.path)已提供

4. 文本化验证证据入口

文件状态大小
$p已提供1661 bytes
$p已提供3121 bytes
$p已提供567 bytes
$p已提供2434 bytes
$p已提供880 bytes

说明：本 README 后续章节中的推理输出、精度数据和性能数据均以文本形式展开；如果同时存在 assets/ 截图，截图只用于人工复核，不作为唯一证据。

5. 推荐复现命令

python inference.py --help
python inference.py --device npu
python eval/eval_accuracy.py --device npu
python eval/eval_performance.py --device 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，不依赖嵌入图像。

审核项直接结果
仓库vit_base_patch16_224.dino_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；所选可复现误差 0.014483699575066566% 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

2026-05-15 00:09:52,964 [INFO] Model loaded on npu:0 with dtype=torch.float32
2026-05-15 00:09:53,216 [INFO] Output shape: torch.Size([1, 768])
2026-05-15 00:09:53,217 [INFO] Throughput: 108.03 images/s
2026-05-15 00:09:53,217 [INFO] Device: npu:0

NPU 推理指标

项目数值
证据在已检入文本文件中未检测到

CPU/GPU 参考与 NPU 精度证据

来源指标数值
results/accuracy_eval.jsonreference_devicecpu
results/accuracy_eval.jsontest_devicenpu
results/accuracy_eval.jsonaggregate.avg_cosine_similarity0.9999935685698963
results/accuracy_eval.jsonaggregate.min_cosine_similarity0.9999932572924186
results/accuracy_eval.jsonaggregate.max_relative_error_filtered7.1365790367126465
results/accuracy_eval.jsonaggregate.mean_relative_error_filtered0.022176285833120347
results/accuracy_eval.jsonaggregate.max_absolute_error0.041754722595214844
results/accuracy_eval.jsonaggregate.cosine_threshold0.99
results/accuracy_eval.jsonaggregate.passedtrue
results/accuracy_eval.jsonper_image[0].max_abs_error0.037459373474121094

精度结论：PASS - 检入的精度证据报告为 PASS；选定的可复现误差 0.014483699575066566% 低于 1%。

性能证据

来源指标数值
results/performance_eval.jsondevicenpu:0
results/performance_eval.jsondtypefloat32
results/performance_eval.jsonbatch_size1
results/performance_eval.jsonwarmup3
results/performance_eval.jsonnum_runs10
results/performance_eval.jsonlatency_ms.avg4.900376690784469
results/performance_eval.jsonlatency_ms.std0.03918720223983793
results/performance_eval.jsonlatency_ms.p504.890927491942421
results/performance_eval.jsonlatency_ms.p904.949382110498846
results/performance_eval.jsonlatency_ms.p994.973861081525683

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

1. 简介

本文档记录 ViT-Base-Patch16-224-DINO 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-Patch16-224-DINO 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_224.dino
  • 相关地址：https://atomgit.com/nanyizjm/vit_base_patch16_224.dino_adapt.git
  • 相关地址：https://gitcode.com/nanyizjm/vit_base_patch16_224.dino_adapt
  • 适配代码仓库：https://gitcode.com/nanyizjm/vit_base_patch16_224.dino_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
NPUAscend NPU（环境数据已在下方“结果数据直接文本”中直接写入）
Python3.8+
PyTorch/torch_npu按 requirements.txt 与当前 NPU 容器环境安装
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.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_224.dino
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master
当前提交4a8fd30

5.2 推理性能

测试结果来源：results/performance_eval.json

指标结果
devicenpu:0
dtypefloat32
batch_size1
image_size224
num_runs10
warmup3
latency_ms{'avg': 4.900376690784469, 'std': 0.03918720223983793, 'p50': 4.890927491942421, 'p90': 4.949382110498846, 'p99': 4.973861081525683, 'min': 4.844304989092052, 'max': 4.976580967195332}

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：802 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
=== Environment Check Log ===
Date: 2026-05-14

--- OS ---
Linux-5.10.0-182.0.0.95.r2220_156.hce2.aarch64-aarch64-with-glibc2.35
Ubuntu 22.04.5 LTS (Jammy Jellyfish)
Architecture: aarch64

--- Python ---
Python 3.11.14
pip 26.0.1

--- NPU Hardware ---
npu-smi 25.5.2
NPU: 2x Ascend910_9362
NPU 5 (Phy-ID 0): Bus 0000:0B:00.0, Health OK, HBM 2909/65536 MB
NPU 5 (Phy-ID 1): Bus 0000:0A:00.0, Health OK, HBM 2870/65536 MB

--- CANN ---
ASCEND_HOME_PATH: /usr/local/Ascend/cann-8.5.1
CANN version: 8.5.1

--- Python Packages ---
torch: 2.9.0+cpu
torch_npu: 2.9.0.post1+gitee7ba04
transformers: 4.57.6
timm: 1.0.27
accelerate: 1.13.0
Pillow: 12.2.0
numpy: 1.26.4

--- NPU Availability ---
torch.npu.is_available(): True
torch.npu.device_count(): 2

--- Status: PASS ---

results/env_info.json

  • 文件大小：2357 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
[LOG_WARNING] can not create directory, directory: /home/atomgit/ascend/log, possible reason: No such file or directory.path string is NULLpath string is NULL{
  "os": "Linux-5.10.0-182.0.0.95.r2220_156.hce2.aarch64-aarch64-with-glibc2.35",
  "python": "3.11.14",
  "machine": "aarch64",
  "torch": "2.9.0+cpu",
  "cuda_available": false,
  "torch_npu": "2.9.0.post1+gitee7ba04",
  "npu_available": true,
  "npu_count": 2,
  "npu_name": "Ascend910_9362",
  "transformers": "4.57.6",
  "timm": "1.0.27",
  "accelerate": "1.13.0",
  "pillow": "12.2.0",
  "numpy": "1.26.4",
  "ascend_home": "/usr/local/Ascend/cann-8.5.1",
  "npu_smi": "+------------------------------------------------------------------------------------------------+\n| npu-smi 25.5.2                   Version: 25.5.2                                               |\n+---------------------------+---------------+----------------------------------------------------+\n| NPU   Name                | Health        | Power(W)    Temp(C)           Hugepages-Usage(page)|\n| Chip  Phy-ID              | Bus-Id        | AICore(%)   Memory-Usage(MB)  HBM-Usage(MB)        |\n+===========================+===============+====================================================+\n| 5     Ascend910           | OK            | 172.9       46                0    / 0             |\n| 0     10                  | 0000:0B:00.0  | 0           0    / 0          2909 / 65536         |\n+------------------------------------------------------------------------------------------------+\n| 5     Ascend910           | OK            | -           47                0    / 0             |\n| 1     11                  | 0000:0A:00.0  | 0           0    / 0          2870 / 65536         |\n+===========================+===============+====================================================+\n+---------------------------+---------------+----------------------------------------------------+\n| NPU     Chip              | Process id    | Process name             | Process memory(MB)      |\n+===========================+===============+====================================================+\n| No running processes found in NPU 5                                                            |\n+===========================+===============+====================================================+\n"
}

logs/inference.log

  • 文件大小：1661 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
2026-05-15 00:09:47,376 [INFO] Using NPU: Ascend910_9362
2026-05-15 00:09:47,377 [INFO] Loading model from: ./model_weights
2026-05-15 00:09:47,378 [INFO] Loading from local weights file
2026-05-15 00:09:52,964 [INFO] Model loaded on npu:0 with dtype=torch.float32
2026-05-15 00:09:52,985 [INFO] Preprocessing image: ./test_image.jpg
2026-05-15 00:09:52,994 [INFO] Input tensor shape: torch.Size([1, 3, 224, 224])
2026-05-15 00:09:52,994 [INFO] Running inference...
2026-05-15 00:09:53,216 [INFO] ============================================================
2026-05-15 00:09:53,216 [INFO] INFERENCE RESULTS
2026-05-15 00:09:53,216 [INFO] ============================================================
2026-05-15 00:09:53,216 [INFO] Output shape: torch.Size([1, 768])
2026-05-15 00:09:53,216 [INFO] CLS embedding (first 10): [0.5933791995048523, -3.0565035343170166, 0.23449330031871796, -0.09277597069740295, -1.1132689714431763, 1.1239155530929565, 0.16453981399536133, 0.003336025634780526, 2.1588807106018066, 7.692113399505615]
2026-05-15 00:09:53,217 [INFO] Embedding stats: mean=0.005949, std=2.327633
2026-05-15 00:09:53,217 [INFO] Embedding norm: 64.463463
2026-05-15 00:09:53,217 [INFO] Inference time: 9.26 ms
2026-05-15 00:09:53,217 [INFO] Throughput: 108.03 images/s
2026-05-15 00:09:53,217 [INFO] Device: npu:0
2026-05-15 00:09:53,217 [INFO] Dtype: float32
2026-05-15 00:09:53,217 [INFO] NPU name: Ascend910_9362
2026-05-15 00:09:53,217 [INFO] NPU memory allocated: 327.9 MB
2026-05-15 00:09:53,217 [INFO] NPU memory reserved: 416.0 MB
2026-05-15 00:09:53,217 [INFO] ============================================================

logs/accuracy_eval.log

  • 文件大小：3121 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
2026-05-15 00:11:24,561 [INFO] ============================================================
2026-05-15 00:11:24,561 [INFO] ACCURACY EVALUATION: NPU vs CPU
2026-05-15 00:11:24,561 [INFO] ============================================================
2026-05-15 00:11:24,576 [INFO] Generated 5 test images, size=224
2026-05-15 00:11:24,576 [INFO] Loading model on CPU...
2026-05-15 00:11:28,913 [INFO] Loading model on NPU...
2026-05-15 00:11:34,433 [INFO] --- Image 1/5 ---
2026-05-15 00:11:35,268 [INFO]   Cosine similarity: 0.99999398
2026-05-15 00:11:35,268 [INFO]   Max relative error: 3.99191999
2026-05-15 00:11:35,268 [INFO]   Mean relative error: 0.02769227
2026-05-15 00:11:35,268 [INFO]   Max abs error: 0.03745937
2026-05-15 00:11:35,268 [INFO]   L2 distance: 0.22117187
2026-05-15 00:11:35,268 [INFO] --- Image 2/5 ---
2026-05-15 00:11:35,901 [INFO]   Cosine similarity: 0.99999347
2026-05-15 00:11:35,901 [INFO]   Max relative error: 0.86851263
2026-05-15 00:11:35,901 [INFO]   Mean relative error: 0.01448370
2026-05-15 00:11:35,901 [INFO]   Max abs error: 0.03701544
2026-05-15 00:11:35,901 [INFO]   L2 distance: 0.22910699
2026-05-15 00:11:35,901 [INFO] --- Image 3/5 ---
2026-05-15 00:11:36,536 [INFO]   Cosine similarity: 0.99999326
2026-05-15 00:11:36,536 [INFO]   Max relative error: 1.30723751
2026-05-15 00:11:36,536 [INFO]   Mean relative error: 0.01513234
2026-05-15 00:11:36,536 [INFO]   Max abs error: 0.03806305
2026-05-15 00:11:36,536 [INFO]   L2 distance: 0.23541953
2026-05-15 00:11:36,536 [INFO] --- Image 4/5 ---
2026-05-15 00:11:37,167 [INFO]   Cosine similarity: 0.99999358
2026-05-15 00:11:37,167 [INFO]   Max relative error: 6.78618574
2026-05-15 00:11:37,167 [INFO]   Mean relative error: 0.02334669
2026-05-15 00:11:37,167 [INFO]   Max abs error: 0.04175472
2026-05-15 00:11:37,167 [INFO]   L2 distance: 0.22816506
2026-05-15 00:11:37,167 [INFO] --- Image 5/5 ---
2026-05-15 00:11:37,800 [INFO]   Cosine similarity: 0.99999355
2026-05-15 00:11:37,800 [INFO]   Max relative error: 7.13657904
2026-05-15 00:11:37,800 [INFO]   Mean relative error: 0.03022643
2026-05-15 00:11:37,800 [INFO]   Max abs error: 0.03587294
2026-05-15 00:11:37,800 [INFO]   L2 distance: 0.22874492
2026-05-15 00:11:37,800 [INFO] ============================================================
2026-05-15 00:11:37,800 [INFO] AGGREGATE RESULTS
2026-05-15 00:11:37,801 [INFO] ============================================================
2026-05-15 00:11:37,801 [INFO] Average cosine similarity: 0.99999357
2026-05-15 00:11:37,801 [INFO] Min cosine similarity: 0.99999326
2026-05-15 00:11:37,801 [INFO] Max relative error (filtered): 7.13657904
2026-05-15 00:11:37,801 [INFO] Mean relative error (filtered): 0.02217629
2026-05-15 00:11:37,801 [INFO] Max absolute error: 0.04175472
2026-05-15 00:11:37,801 [INFO] Target: cosine similarity > 0.99 (< 1% error)
2026-05-15 00:11:37,801 [INFO] PASSED: True
2026-05-15 00:11:37,801 [INFO] ============================================================
2026-05-15 00:11:37,801 [INFO] Results saved to ./results/accuracy_eval.json

results/accuracy_eval.json

  • 文件大小：2434 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_224.dino",
  "reference_device": "cpu",
  "test_device": "npu",
  "dtype": "float32",
  "num_images": 5,
  "image_size": 224,
  "aggregate": {
    "avg_cosine_similarity": 0.9999935685698963,
    "min_cosine_similarity": 0.9999932572924186,
    "max_relative_error_filtered": 7.1365790367126465,
    "mean_relative_error_filtered": 0.022176285833120347,
    "max_absolute_error": 0.041754722595214844,
    "cosine_threshold": 0.99,
    "passed": true
  },
  "per_image": [
    {
      "max_abs_error": 0.037459373474121094,
      "mean_abs_error": 0.006231198087334633,
      "max_relative_error": 3.991919994354248,
      "mean_relative_error": 0.027692267671227455,
      "cosine_similarity": 0.9999939819369742,
      "l2_distance": 0.22117187082767487,
      "ref_norm": 63.70280456542969,
      "npu_norm": 63.68320846557617
    },
    {
      "max_abs_error": 0.037015438079833984,
      "mean_abs_error": 0.006484948564320803,
      "max_relative_error": 0.8685126304626465,
      "mean_relative_error": 0.014483699575066566,
      "cosine_similarity": 0.9999934745054845,
      "l2_distance": 0.22910699248313904,
      "ref_norm": 63.28091812133789,
      "npu_norm": 63.261253356933594
    },
    {
      "max_abs_error": 0.03806304931640625,
      "mean_abs_error": 0.006643829867243767,
      "max_relative_error": 1.3072375059127808,
      "mean_relative_error": 0.01513233594596386,
      "cosine_similarity": 0.9999932572924186,
      "l2_distance": 0.23541952669620514,
      "ref_norm": 63.69096755981445,
      "npu_norm": 63.6716423034668
    },
    {
      "max_abs_error": 0.041754722595214844,
      "mean_abs_error": 0.006433207541704178,
      "max_relative_error": 6.7861857414245605,
      "mean_relative_error": 0.023346692323684692,
      "cosine_similarity": 0.999993580501473,
      "l2_distance": 0.22816506028175354,
      "ref_norm": 63.79940414428711,
      "npu_norm": 63.78319549560547
    },
    {
      "max_abs_error": 0.0358729362487793,
      "mean_abs_error": 0.006475027184933424,
      "max_relative_error": 7.1365790367126465,
      "mean_relative_error": 0.030226433649659157,
      "cosine_similarity": 0.9999935486131314,
      "l2_distance": 0.22874492406845093,
      "ref_norm": 63.34604263305664,
      "npu_norm": 63.32475280761719
    }
  ],
  "timestamp": "2026-05-15 00:11:37"
}

logs/performance_eval.log

  • 文件大小：567 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
2026-05-16 09:14:51,005 [INFO] NPU: Ascend910_9362
2026-05-16 09:14:51,005 [INFO] ============================================================
2026-05-16 09:14:51,005 [INFO] PERFORMANCE EVALUATION
2026-05-16 09:14:51,005 [INFO] ============================================================
2026-05-16 09:14:51,005 [INFO] Device: npu:0
2026-05-16 09:14:51,005 [INFO] Dtype: float32
2026-05-16 09:14:51,005 [INFO] Batch size: 1
2026-05-16 09:14:51,005 [INFO] Image size: 224
2026-05-16 09:14:51,005 [INFO] Warmup: 3
2026-05-16 09:14:51,005 [INFO] Num runs: 10

results/performance_eval.json

  • 文件大小：880 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_224.dino",
  "device": "npu:0",
  "dtype": "float32",
  "batch_size": 1,
  "image_size": 224,
  "warmup": 3,
  "num_runs": 10,
  "latency_ms": {
    "avg": 4.900376690784469,
    "std": 0.03918720223983793,
    "p50": 4.890927491942421,
    "p90": 4.949382110498846,
    "p99": 4.973861081525683,
    "min": 4.844304989092052,
    "max": 4.976580967195332
  },
  "throughput_images_per_sec": 204.06594494675807,
  "npu_memory_mb": {
    "before": 327.94384765625,
    "after": 327.94384765625,
    "reserved": 416.0
  },
  "all_times_ms": [
    4.946360015310347,
    4.909217997919768,
    4.932398966047913,
    4.976580967195332,
    4.893206991255283,
    4.863525973632932,
    4.888647992629558,
    4.861955996602774,
    4.887567018158734,
    4.844304989092052
  ],
  "timestamp": "2026-05-16 09:14:31"
}

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。