From e633b23b8c71935a31cb727be4e2ba84b1c2ed5d Mon Sep 17 00:00:00 2001
From: Siddharth Venkatesan <siddhave@amazon.com>
Date: Mon, 14 Oct 2024 13:14:27 -0700
Subject: [PATCH] [docker] update dependencies in lmi dockerfile (#2439)

---
 serving/docker/lmi.Dockerfile            | 20 ++++++++++----------
 serving/docker/scripts/install_python.sh |  2 +-
 2 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/serving/docker/lmi.Dockerfile b/serving/docker/lmi.Dockerfile
index 6e552b08e..434e8723a 100644
--- a/serving/docker/lmi.Dockerfile
+++ b/serving/docker/lmi.Dockerfile
@@ -22,22 +22,22 @@ ARG pydantic_version=2.8.2
 ARG djl_converter_wheel="https://publish.djl.ai/djl_converter/djl_converter-0.30.0-py3-none-any.whl"
 # HF Deps
 ARG protobuf_version=3.20.3
-ARG transformers_version=4.43.2
-ARG accelerate_version=0.32.1
-ARG bitsandbytes_version=0.43.1
-ARG optimum_version=1.21.2
+ARG transformers_version=4.45.2
+ARG accelerate_version=1.0.0
+ARG bitsandbytes_version=0.44.1
+ARG optimum_version=1.23.0
 ARG auto_gptq_version=0.7.1
-ARG datasets_version=2.20.0
+ARG datasets_version=3.0.1
 ARG autoawq_version=0.2.5
-ARG tokenizers_version=0.19.1
+ARG tokenizers_version=0.20.1
 # LMI-Dist Deps
 ARG vllm_version=0.6.1.post2
 ARG flash_attn_2_wheel="https://github.com/vllm-project/flash-attention/releases/download/v2.6.1/vllm_flash_attn-2.6.1-cp310-cp310-manylinux1_x86_64.whl"
-ARG flash_infer_wheel="https://github.com/flashinfer-ai/flashinfer/releases/download/v0.1.2/flashinfer-0.1.2+cu121torch2.4-cp310-cp310-linux_x86_64.whl"
+ARG flash_infer_wheel="https://github.com/flashinfer-ai/flashinfer/releases/download/v0.1.6/flashinfer-0.1.6+cu124torch2.4-cp310-cp310-linux_x86_64.whl"
 # %2B is the url escape for the '+' character
 ARG lmi_dist_wheel="https://publish.djl.ai/lmi_dist/lmi_dist-11.0.0%2Bnightly-py3-none-any.whl"
 ARG seq_scheduler_wheel="https://publish.djl.ai/seq_scheduler/seq_scheduler-0.1.0-py3-none-any.whl"
-ARG peft_version=0.11.1
+ARG peft_version=0.13.1
 
 EXPOSE 8080
 
@@ -133,5 +133,5 @@ LABEL com.amazonaws.sagemaker.capabilities.multi-models="true"
 LABEL com.amazonaws.sagemaker.capabilities.accept-bind-to-port="true"
 LABEL djl-version=$djl_version
 LABEL cuda-version=$cuda_version
-# To use the 535 CUDA driver, CUDA 12.1 can work on this one too
-LABEL com.amazonaws.sagemaker.inference.cuda.verified_versions=12.2
+# To use the 535 CUDA driver, CUDA 12.4 can work on this one too
+LABEL com.amazonaws.sagemaker.inference.cuda.verified_versions=12.4
diff --git a/serving/docker/scripts/install_python.sh b/serving/docker/scripts/install_python.sh
index 5df1cd167..b61ac94b0 100755
--- a/serving/docker/scripts/install_python.sh
+++ b/serving/docker/scripts/install_python.sh
@@ -12,7 +12,7 @@ else
   DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends curl software-properties-common git
   add-apt-repository -y ppa:deadsnakes/ppa
   apt-get autoremove -y python3
-  apt-get install -y "python${PYTHON_VERSION}-dev" "python${PYTHON_VERSION}-distutils" "python${PYTHON_VERSION}-venv"
+  DEBIAN_FRONTEND=noninteractive apt-get install -y "python${PYTHON_VERSION}-dev" "python${PYTHON_VERSION}-distutils" "python${PYTHON_VERSION}-venv"
   ln -sf /usr/bin/"python${PYTHON_VERSION}" /usr/bin/python3
   ln -sf /usr/bin/"python${PYTHON_VERSION}" /usr/bin/python
   curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py