硬件:Ascend NPU(Ascend 910B4) · 推理引擎:torch_npu · 设备:npu:0 精度验收:CPU vs NPU 推理精度差异 < 1%(实测 SKIP,cosine=)
**A TINY version of the original FloodDiffusion** [Paper](https://arxiv.org/abs/2512.03520) | [Github](https://github.com/ShandaAI/FloodDiffusion) | [Project Page](https://shand
| 属性 | 值 |
|---|---|
| 模型名 | ShandaAI/FloodDiffusionTiny |
| 原始权重 | ModelScope / HuggingFace / AtomGit 镜像 |
| 任务类型 | feature-extraction |
| 输入 | 见 inference.py |
| 输出 | 特征向量 |
| 框架 | PyTorch + torch_npu |
| 官方实现 | https://github.com/ShandaAI/FloodDiffusion |
官方实现 / 论文仓库:
| 依赖项 | 版本 | 说明 |
|---|---|---|
| Python | 3.11 | conda 环境 shandaai-flooddiffusiontiny-npu |
| torch / torch_npu | 2.10.0 / 2.10.0 | 由 setup_env.sh 装,须严格配套 |
| diffusers | ==0.35.2 | |
| transformers | ==4.51.3 | |
| accelerate | ==1.14.0 | |
| pillow | ==12.3.0 | |
| safetensors | ==0.8.0 | |
| modelscope | ==1.37.0 | |
| huggingface-hub | ==0.36.2 | |
| pyyaml | >=6.0 | |
| einops | ==0.8.0 | |
| lightning | ==2.6.5 | |
| ftfy | ==6.3.1 | |
| sentencepiece | ==0.2.2 | |
| omegaconf | ==2.3.0 |
一键建环境(从零创建 conda 环境并装依赖):
bash setup_env.sh # conda create -n <env> python=<ver> + torch/torch_npu + requirements.txt⚠️ 重要:若环境已装 torch/torch_npu,不要强制覆盖 torch,否则破坏 torch_npu 算子插件绑定。
国内 pip 源(Tsinghua):https://pypi.tuna.tsinghua.edu.cn/simple
权重不随仓库分发,推理前先用国内下载源拉取到本地缓存(推荐 ModelScope,失败回落 HF 镜像):
# 方式一:ModelScope(国内直连,最快)
pip install modelscope -i https://pypi.tuna.tsinghua.edu.cn/simple
python -c "from modelscope import snapshot_download; snapshot_download('ShandaAI/FloodDiffusionTiny', cache_dir='./weights')"
# 方式二:HF 镜像(hf-mirror.com)
export HF_ENDPOINT=https://hf-mirror.com
huggingface-cli download ShandaAI/FloodDiffusionTiny --local-dir ./weights/FloodDiffusionTinyinference.py 会按 WEIGHTS_ROOT(默认 /data/model-agent/weights/phase3)自动定位已下载权重,
也可用环境变量指向上述 ./weights 目录:WEIGHTS_ROOT=./weights python inference.py
硬件:单卡 Ascend 910B4(npu:2,物理卡 6)
评委裸 clone 后本机没有 conda 环境,conda activate 不会创建环境。必须先建环境再推理:
# 第一步:从零创建 conda 环境并装依赖(conda create + torch/torch_npu + requirements.txt)
bash setup_env.sh
# 第二步:激活环境并一键推理(环境自检 + inference + eval + benchmark)
conda activate shandaai-flooddiffusiontiny-npu
bash run.shimport torch, torch_npu
from transformers import AutoModel, AutoTokenizer
m = AutoModel.from_pretrained("ShandaAI/FloodDiffusionTiny", trust_remote_code=True).to("npu:0").eval()
tok = AutoTokenizer.from_pretrained("ShandaAI/FloodDiffusionTiny", trust_remote_code=True)
inputs = tok("How to use this model?", return_tensors="pt").to("npu:0")
outputs = m(**inputs)logs/inference.log)输入:示例序列/文本(见 inference.py)
输出:
=== Inference: ShandaAI/FloodDiffusionTiny on Ascend NPU ===
NPU: npu:0
加载方式: ldf_motion(trust_remote_code,LDF+WanVAE+umt5-xxl 随仓)
Prompt: a person walks forward slowly | length=16
输出 motion shape: (61, 263)(HumanML3D 263 维)
推理耗时: 4.18 s
数值范围: [-1.0035, 1.5841] mean=0.1993logs/accuracy.log)
精度对比:SKIP(cpu-forward-infeasible-large-model)
注:本次适配无精度要求,精度数据作为质量证据。详见
logs/accuracy.log。
| 指标 | 数值 |
|---|---|
| avg latency | 2.93 s |
| p50 | N/A |
| throughput | 0.341 motions/sec |



截图由真实执行日志(
logs/)忠实渲染为终端风格 PNG,渲染脚本tools/gen_screenshots_pw.py(playwright),可复现,无虚构结果。
#NPU #Ascend #Ascend910 #model-agent-tagged
仓库:https://gitcode.com/v_50/ShandaAI-FloodDiffusionTiny-NPU 适配日期:2026-08-12 | 赛道:第二季·模型适配·阶段三