Go to 12.1 instead of 12.2 so gpu system doesn't have to upgrade driv…

…er from 530.30.02 cuda12.1

docker_build_script_ubuntu.sh

@@ -90,7 +90,7 @@ python -m venv vllm_env --system-site-packages
 /h2ogpt_conda/vllm_env/bin/python -m pip install https://h2o-release.s3.amazonaws.com/h2ogpt/megablocks-0.5.1-cp310-cp310-linux_x86_64.whl
 /h2ogpt_conda/vllm_env/bin/python -m pip install https://h2o-release.s3.amazonaws.com/h2ogpt/triton-2.2.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
 /h2ogpt_conda/vllm_env/bin/python -m pip install https://h2o-release.s3.amazonaws.com/h2ogpt/mosaicml_turbo-0.0.9-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
-/h2ogpt_conda/vllm_env/bin/python -m pip install https://h2o-release.s3.amazonaws.com/h2ogpt/vllm-0.2.7%2Bcu123-cp310-cp310-linux_x86_64.whl
+/h2ogpt_conda/vllm_env/bin/python -m pip install https://h2o-release.s3.amazonaws.com/h2ogpt/vllm-0.2.7-cp310-cp310-linux_x86_64.whl
 mkdir $VLLM_CACHE
 chmod -R a+rwx /h2ogpt_conda
 

docs/README_InferenceServers.md

@@ -260,8 +260,7 @@ Assuming torch was installed with CUDA 12.1, and you have installed cuda locally
 ```bash
 export CUDA_HOME=/usr/local/cuda-12.1
 export PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cu123"
-pip install mosaicml-turbo --upgrade  # see docker_build_script_ubuntu.sh for x86 prebuilt wheel on s3
-pip install git+https://github.com/stanford-futuredata/megablocks.git  # see docker_build_script_ubuntu.sh for x86 prebuilt wheel on s3
+pip install mosaicml-turbo megablocks==0.5.1 --upgrade  # see docker_build_script_ubuntu.sh for x86 prebuilt wheel on s3
 pip install fschat==0.2.34 ray pandas gputil==1.4.0 uvicorn[standard]
 # optional:
 pip install flash-attn==2.4.2