nanyizjm/C-RADIOv2-B-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，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/C-RADIOv2-B-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；不使用图像替代数据。

Track 1 模型卡片摘要

项目数值
模型仓库https://gitcode.com/nanyizjm/C-RADIOv2-B-adapt
原始模型或权重来源https://gitcode.com/hf_mirrors/nvidia/C-RADIOv2-B
竞赛赛道Track 1：模型适配
目标硬件昇腾 NPU
必备功能NPU 推理成功运行或明确记录阻塞原因
必备精度NPU 结果与 CPU/GPU 参考结果相比，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 中。

#+NPU

C-RADIOv2-B 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C-RADIOv2-B-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 0.004422400008059308% 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": [
"global_feature_shape": [
"patch_feature_shape": [
device: npu:0
input_shape: [1, 3, 224, 224]
global_feature_shape: [1, 1280]
patch_feature_shape: [1, 196, 1280]

NPU 推理指标

来源指标数值
logs/inference.jsondevicenpu:0
logs/inference.jsoninput_shape[1,3,224,224]
logs/inference.jsonglobal_feature_shape[1,1280]
logs/inference.jsonpatch_feature_shape[1,196,1280]

CPU/GPU 参考与 NPU 精度验证

来源指标数值
results/accuracy_eval.jsonreference_devicecpu
results/accuracy_eval.jsontest_devicenpu:0
results/accuracy_eval.jsonnpu_nameAscend910_9362
results/accuracy_eval.jsonglobal_max_rel_error_pct0.045111858009508274
results/accuracy_eval.jsonglobal_mean_rel_error_pct0.00856077733904385
results/accuracy_eval.jsonglobal_min_cosine_similarity0.9999999398898396
results/accuracy_eval.jsonpatch_max_rel_error_pct0.027480260917988133
results/accuracy_eval.jsonpatch_mean_rel_error_pct0.004422400008059308
results/accuracy_eval.jsonpatch_min_cosine_similarity0.9999999672046266
results/accuracy_eval.jsonall_pass_lt_1pcttrue

精度结论：通过 - 已提交的精度验证报告显示通过；选定的可复现误差 0.004422400008059308% 低于 1%。

性能验证

来源指标数值
results/performance_eval.jsondevicenpu:0
results/performance_eval.jsondtypefloat32
results/performance_eval.jsonmemory_before.free_mb61982.71
results/performance_eval.jsonmemory_before.total_mb62740
results/performance_eval.jsonmemory_before.used_mb757.29
results/performance_eval.jsonmemory_after.free_mb61690.71
results/performance_eval.jsonmemory_after.total_mb62740
results/performance_eval.jsonmemory_after.used_mb1049.29
results/performance_eval.jsonper_batch_results.bs1.avg_time_sec0.0050974607467651365
results/performance_eval.jsonper_batch_results.bs1.std_time_sec0.000132789990316491

C-RADIOv2-B on Ascend NPU

1. 简介

本文档记录 C-RADIOv2-B 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。

C-RADIOv2-B 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/nvidia/C-RADIOv2-B
  • 相关地址：https://atomgit.com/nanyizjm/C-RADIOv2-B-adapt.git
  • 相关地址：https://gitcode.com/nanyizjm/C-RADIOv2-B-adapt
  • 适配代码仓库：https://gitcode.com/nanyizjm/C-RADIOv2-B-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
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C-RADIOv2-B
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72b65bc

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]

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
精度评估结果assets/accuracy_eval_result.png
环境检查assets/env_check.png
Git 提交结果assets/git_submit_result.png
推理结果assets/inference_result.png
性能评估结果assets/performance_eval_result.png

推理正常输出证据

  • 仓库：C-RADIOv2-B-adapt
  • 原始模型/权重来源：https://gitcode.com/hf_mirrors/nvidia/C-RADIOv2-B
  • 目标硬件：Ascend NPU
  • 证据来源：logs/inference.json
  • 渲染证据图片文件：assets/inference_result.png
  • 证据策略：截图内容已转录为以下 README 文本；图片未嵌入。
项目证据
状态PASS - NPU 推理生成了全局特征和补丁特征
设备npu:0
输入形状[1, 3, 224, 224]
全局特征形状[1, 1280]
补丁特征形状[1, 196, 1280]
推理时间0.009391 秒
吞吐量106.48 张/秒

推理证据图片全文转录

# Inference Evidence

Repository: C-RADIOv2-B-adapt
Model: C-RADIOv2-B
Date: 2026-05-16 07:03:22

Command:
python inference.py --model_path <model_path> --device npu

Output (from logs/inference.log):
=== C-RADIOv2-B Inference Log ===
model: C-RADIOv2-B
device: npu:0
dtype: float32
image_path: ./test_image.jpg
input_shape: [1, 3, 224, 224]
global_feature_shape: [1, 1280]
patch_feature_shape: [1, 196, 1280]
global_feature_dim: 1280
patch_feature_dim: 1280
num_patches: 196
global_l2_norm: 1.0
global_feature_sample: [0.014378, 0.029448, -0.001252, 0.030316, -0.046129, -0.026451, -0.009364, -1e-05]
patch_l2_norm_mean: 1.0
inference_time_sec: 0.009391
images_per_sec: 106.48
npu_available: True
npu_count: 2
npu_name: Ascend910_9362
model_load_time_sec: 5.61

timestamp: 2026-05-15 01:36:58


Status:
See log for details.

Log File:
logs/inference.log

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：2672 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: C-RADIOv2-B-adapt
# Model: C-RADIOv2-B
# 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            | 174.8       48                0    / 0             |
| 0     0                   | 0000:0A:00.0  | 0           0    / 0          3106 / 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: 7a235978ae3fb32ac62e897fe75d92a42ddec517

<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：565 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 01:38:32",
  "os": "Linux-5.10.0-182.0.0.95.r2220_156.hce2.aarch64-aarch64-with-glibc2.35",
  "python": "3.11.14",
  "model": "C-RADIOv2-B",
  "model_class": "CRADIOv2B (ViT-Base)",
  "model_params_m": 98.23,
  "input_size": [
    3,
    224,
    224
  ],
  "global_feature_dim": 1280,
  "patch_feature_dim": 1280,
  "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"
}

logs/inference.json

  • 文件大小：759 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{
  "model": "C-RADIOv2-B",
  "device": "npu:0",
  "dtype": "float32",
  "image_path": "./test_image.jpg",
  "input_shape": [
    1,
    3,
    224,
    224
  ],
  "global_feature_shape": [
    1,
    1280
  ],
  "patch_feature_shape": [
    1,
    196,
    1280
  ],
  "global_feature_dim": 1280,
  "patch_feature_dim": 1280,
  "num_patches": 196,
  "global_l2_norm": 1.0,
  "global_feature_sample": [
    0.014378,
    0.029448,
    -0.001252,
    0.030316,
    -0.046129,
    -0.026451,
    -0.009364,
    -1e-05
  ],
  "patch_l2_norm_mean": 1.0,
  "inference_time_sec": 0.009391,
  "images_per_sec": 106.48,
  "npu_available": true,
  "npu_count": 2,
  "npu_name": "Ascend910_9362",
  "model_load_time_sec": 5.61
}

logs/inference.log

  • 文件大小：615 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
=== C-RADIOv2-B Inference Log ===
model: C-RADIOv2-B
device: npu:0
dtype: float32
image_path: ./test_image.jpg
input_shape: [1, 3, 224, 224]
global_feature_shape: [1, 1280]
patch_feature_shape: [1, 196, 1280]
global_feature_dim: 1280
patch_feature_dim: 1280
num_patches: 196
global_l2_norm: 1.0
global_feature_sample: [0.014378, 0.029448, -0.001252, 0.030316, -0.046129, -0.026451, -0.009364, -1e-05]
patch_l2_norm_mean: 1.0
inference_time_sec: 0.009391
images_per_sec: 106.48
npu_available: True
npu_count: 2
npu_name: Ascend910_9362
model_load_time_sec: 5.61

timestamp: 2026-05-15 01:36:58

logs/accuracy_eval.log

  • 文件大小：1385 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
=== C-RADIOv2-B Accuracy Evaluation ===
Time: 2026-05-15 01:37:31
NPU available: True (2x Ascend910_9362)
Created 5 test images

Loading model on CPU...
Loading model on NPU...

--- Image 1/5 ---
  Global - Max rel error: 0.045112%
  Global - Mean rel error: 0.008459%
  Global - Cosine sim: 0.99999995
  Patch  - Max rel error: 0.025088%
  Patch  - Cosine sim: 0.99999997

--- Image 2/5 ---
  Global - Max rel error: 0.038468%
  Global - Mean rel error: 0.008881%
  Global - Cosine sim: 0.99999994
  Patch  - Max rel error: 0.026775%
  Patch  - Cosine sim: 0.99999997

--- Image 3/5 ---
  Global - Max rel error: 0.036205%
  Global - Mean rel error: 0.008957%
  Global - Cosine sim: 0.99999994
  Patch  - Max rel error: 0.027104%
  Patch  - Cosine sim: 0.99999997

--- Image 4/5 ---
  Global - Max rel error: 0.038305%
  Global - Mean rel error: 0.008583%
  Global - Cosine sim: 0.99999994
  Patch  - Max rel error: 0.026408%
  Patch  - Cosine sim: 0.99999997

--- Image 5/5 ---
  Global - Max rel error: 0.033702%
  Global - Mean rel error: 0.007924%
  Global - Cosine sim: 0.99999995
  Patch  - Max rel error: 0.027480%
  Patch  - Cosine sim: 0.99999997

=== Summary ===
  Global max rel error: 0.045112%
  Global min cosine sim: 0.99999994
  Patch max rel error: 0.027480%
  Patch min cosine sim: 0.99999997
  All pass (< 1%): True

results/accuracy_eval.json

  • 文件大小：4363 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{
  "model": "C-RADIOv2-B",
  "reference_device": "cpu",
  "test_device": "npu:0",
  "npu_name": "Ascend910_9362",
  "num_images": 5,
  "global_max_rel_error_pct": 0.045111858009508274,
  "global_mean_rel_error_pct": 0.00856077733904385,
  "global_min_cosine_similarity": 0.9999999398898396,
  "patch_max_rel_error_pct": 0.027480260917988133,
  "patch_mean_rel_error_pct": 0.004422400008059308,
  "patch_min_cosine_similarity": 0.9999999672046266,
  "all_pass_lt_1pct": true,
  "per_image_results": [
    {
      "image_idx": 0,
      "global_metrics": {
        "max_abs_error": 3.933161497116089e-05,
        "mean_abs_error": 7.3754422771799e-06,
        "max_rel_error_pct": 0.045111858009508274,
        "mean_rel_error_pct": 0.008459350194733274,
        "cosine_similarity": 0.999999945797489,
        "mse": 8.437891305003364e-11,
        "snr_db": 66.27078727670488
      },
      "patch_metrics": {
        "max_abs_error": 3.3268705010414124e-05,
        "mean_abs_error": 5.70628939251373e-06,
        "max_rel_error_pct": 0.02508838877734748,
        "mean_rel_error_pct": 0.004303191444050031,
        "cosine_similarity": 0.9999999672046266,
        "mse": 5.122039139091182e-11,
        "snr_db": 67.13179674395565
      }
    },
    {
      "image_idx": 1,
      "global_metrics": {
        "max_abs_error": 3.346707671880722e-05,
        "mean_abs_error": 7.72615878901206e-06,
        "max_rel_error_pct": 0.038467935030542695,
        "mean_rel_error_pct": 0.008880649386516394,
        "cosine_similarity": 0.9999999409699577,
        "mse": 9.192193239493194e-11,
        "snr_db": 66.09665347621667
      },
      "patch_metrics": {
        "max_abs_error": 3.460980951786041e-05,
        "mean_abs_error": 5.662270723190421e-06,
        "max_rel_error_pct": 0.02677502214512353,
        "mean_rel_error_pct": 0.004380475539077181,
        "cosine_similarity": 0.9999999676253919,
        "mse": 5.056295164585698e-11,
        "snr_db": 67.15071910894048
      }
    },
    {
      "image_idx": 2,
      "global_metrics": {
        "max_abs_error": 3.089476376771927e-05,
        "mean_abs_error": 7.643185477945736e-06,
        "max_rel_error_pct": 0.03620473213409483,
        "mean_rel_error_pct": 0.00895684087312434,
        "cosine_similarity": 0.9999999416698779,
        "mse": 9.082831800642454e-11,
        "snr_db": 66.12147227038024
      },
      "patch_metrics": {
        "max_abs_error": 3.4224241971969604e-05,
        "mean_abs_error": 5.698677798360421e-06,
        "max_rel_error_pct": 0.02710391843534953,
        "mean_rel_error_pct": 0.0045130728786513755,
        "cosine_similarity": 0.9999999672487938,
        "mse": 5.115138503768461e-11,
        "snr_db": 67.13377898730332
      }
    },
    {
      "image_idx": 3,
      "global_metrics": {
        "max_abs_error": 3.4518539905548096e-05,
        "mean_abs_error": 7.734583189744625e-06,
        "max_rel_error_pct": 0.03830502773952052,
        "mean_rel_error_pct": 0.0085830230492796,
        "cosine_similarity": 0.9999999398898396,
        "mse": 9.360962433852213e-11,
        "snr_db": 66.05863091788996
      },
      "patch_metrics": {
        "max_abs_error": 3.432575613260269e-05,
        "mean_abs_error": 5.6279736017349176e-06,
        "max_rel_error_pct": 0.026408327585985523,
        "mean_rel_error_pct": 0.004329849863925641,
        "cosine_similarity": 0.999999968039511,
        "mse": 4.9915889443785816e-11,
        "snr_db": 67.16942369816202
      }
    },
    {
      "image_idx": 4,
      "global_metrics": {
        "max_abs_error": 3.010965883731842e-05,
        "mean_abs_error": 7.079475335802776e-06,
        "max_rel_error_pct": 0.03370159843773025,
        "mean_rel_error_pct": 0.007924023191565646,
        "cosine_similarity": 0.9999999492324732,
        "mse": 7.901175571241861e-11,
        "snr_db": 66.39908412523303
      },
      "patch_metrics": {
        "max_abs_error": 3.4061260521411896e-05,
        "mean_abs_error": 5.683528835079659e-06,
        "max_rel_error_pct": 0.027480260917988133,
        "mean_rel_error_pct": 0.004585410314592317,
        "cosine_similarity": 0.9999999674409206,
        "mse": 5.0851188458352576e-11,
        "snr_db": 67.14241293587304
      }
    }
  ]
}

logs/performance_eval.log

  • 文件大小：714 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
=== C-RADIOv2-B Performance Evaluation ===
Time: 2026-05-16 10:11:12
NPU: True (2x Ascend910_9362)
Loading model...
Model loaded in 5.40s
Global feature dim: 1280
Patch feature dim: 1280

--- BS=1 ---
  Avg: 0.0051s  P50: 0.0051s  P90: 0.0052s
  Throughput: 196.18 img/s

--- BS=4 ---
  Avg: 0.0054s  P50: 0.0054s  P90: 0.0055s
  Throughput: 184.21 img/s

--- BS=8 ---
  Avg: 0.0081s  P50: 0.0081s  P90: 0.0081s
  Throughput: 123.94 img/s

--- BS=16 ---
  Avg: 0.0141s  P50: 0.0141s  P90: 0.0141s
  Throughput: 71.06 img/s

=== Summary ===
  bs1: avg=0.0051s thr=196.18
  bs4: avg=0.0054s thr=184.21
  bs8: avg=0.0081s thr=123.94
  bs16: avg=0.0141s thr=71.06
  NPU mem: 1049/62740MB

results/performance_eval.json

  • 文件大小：1971 bytes
  • 以下内容为 README 直接文本转写，不是外部路径引用。
{
  "model": "C-RADIOv2-B",
  "device": "npu:0",
  "dtype": "float32",
  "npu_name": "Ascend910_9362",
  "input_size": [
    3,
    224,
    224
  ],
  "global_feature_dim": 1280,
  "patch_feature_dim": 1280,
  "memory_before": {
    "free_mb": 61982.71,
    "total_mb": 62740.0,
    "used_mb": 757.29
  },
  "memory_after": {
    "free_mb": 61690.71,
    "total_mb": 62740.0,
    "used_mb": 1049.29
  },
  "per_batch_results": {
    "bs1": {
      "avg_time_sec": 0.0050974607467651365,
      "std_time_sec": 0.000132789990316491,
      "min_time_sec": 0.00500178337097168,
      "max_time_sec": 0.005484580993652344,
      "p50_sec": 0.005055785179138184,
      "p90_sec": 0.005160140991210938,
      "throughput_img_per_sec": 196.1761060415429,
      "num_runs": 10,
      "warmup": 3
    },
    "bs4": {
      "avg_time_sec": 0.0054285526275634766,
      "std_time_sec": 1.9010200413185365e-05,
      "min_time_sec": 0.0054073333740234375,
      "max_time_sec": 0.0054666996002197266,
      "p50_sec": 0.005423069000244141,
      "p90_sec": 0.005451679229736328,
      "throughput_img_per_sec": 184.2111643023409,
      "num_runs": 10,
      "warmup": 3
    },
    "bs8": {
      "avg_time_sec": 0.008068346977233886,
      "std_time_sec": 1.4662044787665235e-05,
      "min_time_sec": 0.00804901123046875,
      "max_time_sec": 0.008090734481811523,
      "p50_sec": 0.008069276809692383,
      "p90_sec": 0.008085155487060547,
      "throughput_img_per_sec": 123.94112484523258,
      "num_runs": 10,
      "warmup": 3
    },
    "bs16": {
      "avg_time_sec": 0.014073443412780762,
      "std_time_sec": 2.1829834770604343e-05,
      "min_time_sec": 0.01404714584350586,
      "max_time_sec": 0.014113664627075195,
      "p50_sec": 0.014072179794311523,
      "p90_sec": 0.014107227325439453,
      "throughput_img_per_sec": 71.05581560031375,
      "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。