From 21ee57a26c20a6a2765526cc61f7f2683bb537d5 Mon Sep 17 00:00:00 2001 From: Tanmay Verma Date: Tue, 10 Oct 2023 11:37:14 -0700 Subject: [PATCH] Cherry-picking vLLM backend changes (#6404) * Update build.py to build vLLM backend (#6394) * Add Python backend when vLLM backend built (#6397) --------- Co-authored-by: dyastremsky <58150256+dyastremsky@users.noreply.github.com> --- build.py | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/build.py b/build.py index 0e8b045bb3..41846c0163 100755 --- a/build.py +++ b/build.py @@ -78,6 +78,8 @@ "2023.0.0", # Standalone OpenVINO "2.4.7", # DCGM version "py310_23.1.0-1", # Conda version + "9.1.0.1", # TRT version for building TRT-LLM backend + "12.2", # CUDA version for building TRT-LLM backend "0.2.1.post1", # vLLM version ) } @@ -1364,6 +1366,16 @@ def dockerfile_prepare_container_linux(argmap, backends, enable_gpu, target_mach TRITON_VERSION_MAP[FLAGS.version][7] ) + if "vllm" in backends: + # [DLIS-5606] Build Conda environment for vLLM backend + # Remove Pip install once vLLM backend moves to Conda environment. + df += """ +# vLLM needed for vLLM backend +RUN pip3 install vllm=={} +""".format( + TRITON_VERSION_MAP[FLAGS.version][9] + ) + df += """ WORKDIR /opt/tritonserver RUN rm -fr /opt/tritonserver/*