add setuptools and add deepspeed

Dockerfile.H100

@@ -22,15 +22,15 @@ RUN wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfilter
 
 RUN /workspace/bin/conda init bash
 
+RUN /workspace/bin/conda create -n vllmenv python=3.10 -y && \
+    /workspace/bin/conda run -n vllmenv pip install -r requirements_H100.txt && \
+    /workspace/bin/conda run -n vllmenv pip install -e vllm/ -v
+
 RUN /workspace/bin/conda create -n axolotlenv python=3.10 -y && \
     /workspace/bin/conda run -n axolotlenv pip install packaging torch ninja -v && \
     /workspace/bin/conda run -n axolotlenv pip install -e axolotl/[flash-attn,deepspeed] -v && \
     /workspace/bin/conda run -n axolotlenv pip install -r requirements_H100.txt && \
     /workspace/bin/conda run -n axolotlenv pip uninstall flash-attn -v -y && \
     FLASH_ATTENTION_FORCE_BUILD=TRUE /workspace/bin/conda run -n axolotlenv pip install flash-attn -v
 
-RUN /workspace/bin/conda create -n vllmenv python=3.10 -y && \
-    /workspace/bin/conda run -n vllmenv pip install -e vllm/ -v && \
-    /workspace/bin/conda run -n vllmenv pip install -r requirements_H100.txt
-
 CMD ["python3", "script.py"]

Makefile

@@ -3,7 +3,7 @@ build-and-push:
 	sudo docker push pierre781/benchmark:latest
 
 build-and-push-h100:
-	sudo docker build --no-cache -f Dockerfile.H100 -t pierre781/benchmark:H100 .
+	sudo docker build --no-cache  --progress=plain -f Dockerfile.H100 -t pierre781/benchmark:H100 .
 	sudo docker push pierre781/benchmark:H100
 
 build-and-push-candidate:

script.py

@@ -158,7 +158,7 @@ def run_single_gpu_finetune(device_count):
 def run_multi_gpu_finetune(device_count):
     total_vram = get_vram()
     config = get_finetune_config(total_vram)
-    command = ["/workspace/bin/conda", "run", "-n" "axolotlenv", "accelerate", "launch", "-m", "axolotl.cli.train", config]
+    command = ["/workspace/bin/conda", "run", "-n" "axolotlenv", "accelerate", "launch", "-m", "axolotl.cli.train", config, "--deepspeed", "deepspeed_configs/zero1.json"]
     out, err = shell_cmd(command)
     if err != None:
         return None, err