update Dockerfile and Readme

Dockerfile

@@ -1,6 +1,8 @@
-FROM zibo.harbor.iluvatar.com.cn:30000/saas/bi100-3.2.1-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.2
+FROM mlu370-pytorch:v25.01-torch2.5.0-torchmlu1.24.1-ubuntu22.04-py310
 
 WORKDIR /workspace
-ENV PT_SDPA_ENABLE_HEAD_DIM_PADDING=1
+ENV PATH=/torch/venv3/pytorch_infer/bin:/workspace/ffmpeg-mlu-v4.2.0/install/bin:/usr/local/neuware/bin:/usr/local/openmpi/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
+
 RUN pip install diffusers==0.34.0
-COPY main.py test.sh dataset.json /workspace/
+RUN sed -i 's|source /torch/venv3/pytorch/bin/activate|source /torch/venv3/pytorch_infer/bin/activate|' /root/.bashrc
+COPY main.py dataset.json /workspace/

Dockerfile.mrv100

@@ -1,7 +0,0 @@
-FROM corex:4.3.0
-
-WORKDIR /workspace
-COPY whls-mrv100 /packages
-RUN pip install diffusers==0.34.0 sentencepiece transformers==4.55.2
-# RUN pip install /packages/*.whl
-COPY main.py test.sh dataset.json /workspace/

README.md

@@ -0,0 +1,24 @@
+## Quickstart
+
+### 构建镜像
+```bash
+docker build -t diffusers:v0.1 .
+```
+其中基础镜像 mlu370-pytorch:v25.01-torch2.5.0-torchmlu1.24.1-ubuntu22.04-py310 联系寒武纪厂商技术支持可获取
+
+### 模型下载
+模型地址：https://modelscope.cn/models/AI-ModelScope/stable-diffusion-v1-5
+并放到目录：`/mnt/contest_ceph/zhanghao/models/stable-diffusion-v1-5`（如更改目录，请修改后面的执行脚本中的模型路径）
+
+### 测试程序
+1. 准备输入数据集，可以参考示例`dataset.json`
+2. 在docker镜像里运行测试程序，会根据`dataset.json`内容，在`output`目录下生成图片文件。
+```bash
+./run_in_docker.sh
+```
+
+## 测试结果
+|      | A100 平均生成时间（秒） | MLU370-x4 平均生成时间（秒） |  MLU370-x8 平均生成时间（秒） |
+|------|-------------------------|----------------------------|----------------------------|
+| 时间 | 1.5                  | 4.2                     | 5.3                     |
+

run_in_docker.sh

@@ -1,3 +1,4 @@
 #! /usr/bin/env bash
-image=harbor-contest.4pd.io/zhanghao/diffusers:bi100-0.2
+image=diffusers:v0.1
-docker run -it -v /root/zhanghao:/workspace -v /mnt:/mnt --device=dev/iluvatar1:/dev/iluvatar0 $image bash
+device=cambricon_dev1
+docker run -v `pwd`:/workspace -v /mnt/contest_ceph/zhanghao/models/stable-diffusion-v1-5:/workspace/stable-diffusion-v1-5 --device=/dev/$device:/dev/cambricon_dev0 --device=/dev/cambricon_ctl:/dev/cambricon_ctl $image python3 main.py --model "./stable-diffusion-v1-5" --json  "dataset.json" --results "results.json" --outdir "output" --device cuda --dtype fp16

run_in_docker_mrv100.sh

@@ -1,3 +0,0 @@
-#! /usr/bin/env bash
-image=harbor-contest.4pd.io/zhanghao/diffusers:mrv100-0.2
-docker run -it -v /root/zhanghao:/workspace -v /mnt:/mnt --device=dev/iluvatar0:/dev/iluvatar0 $image bash

test.py

@@ -1,13 +0,0 @@
-from diffusers import DiffusionPipeline
-import torch
-import time
-
-model_path = "/mnt/contest_ceph/zhanghao/models/stable-diffusion-v1-5"
-# model_path = "/mnt/contest_ceph/zhanghao/models/stable-diffusion-3.5-medium"
-pipeline = DiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
-pipeline.to("cuda")
-start = time.time()
-image = pipeline("An image of a squirrel in Picasso style").images[0]
-end = time.time()
-print(f"elapsed: {end - start}")
-image.save("squirrel_picasso.png")

test.sh

@@ -1,7 +0,0 @@
-python3 main.py \
-  --model "/mnt/contest_ceph/zhanghao/models/stable-diffusion-v1-5" \
-  --json  "dataset.json" \
-  --results "results.json" \
-  --outdir "output" \
-  --device cuda \
-  --dtype fp16