Upgrade Torch and it's dependencies to v2.5.x for video-llama (opea-project#1551)

Signed-off-by: Abolfazl Shahbazi <[email protected]>

Author: ashahba

comps/third_parties/video-llama/src/Dockerfile

@@ -1,7 +1,7 @@
 # Copyright (C) 2024 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 
-FROM python:3.9-slim
+FROM python:3.11-slim
 
 ENV LANG=C.UTF-8
 
@@ -30,9 +30,12 @@ RUN git clone ${VIDEO_LLAMA_REPO} Video-LLaMA && \
     mv video_llama ../ && \
     cd ../ && rm -rf Video-LLaMA
 
+# Modify the degradations.py file to import rgb_to_grayscale from torchvision.transforms.functional
+RUN sed -i 's/from torchvision.transforms.functional_tensor import rgb_to_grayscale/from torchvision.transforms.functional import rgb_to_grayscale/' /usr/local/lib/python3.11/site-packages/pytorchvideo/transforms/augmentations.py && \
+    sed -i 's/torchvision.transforms.functional_tensor/torchvision.transforms.v2.functional/' /usr/local/lib/python3.11/site-packages/pytorchvideo/transforms/augmentations.py
+
 USER user
 
 ENV PYTHONPATH=/home/user
 
-
 ENTRYPOINT ["bash", "start.sh"]

comps/third_parties/video-llama/src/README.md

@@ -24,7 +24,7 @@ export ip_address=$(hostname -I | awk '{print $1}')
 export no_proxy=$no_proxy,${ip_address}
 export LVM_ENDPOINT=http://${ip_address}:9009
 # Start service
-docker compose -f comps/third_parties/llama-vision/deployment/docker_compose/docker_compose.yaml up -d
+docker compose -f comps/third_parties/video-llama/src/docker_compose_vllama.yaml up -d
 # it should take about 1.5 hours for the model to download in the video-llama server, assuming a maximum download speed of 100 Mbps
 until docker logs lvm-video-llama 2>&1 | grep -q "Uvicorn running on"; do
     sleep 5m

comps/third_parties/video-llama/src/docker_compose_vllama.yaml

@@ -14,6 +14,11 @@ services:
       https_proxy: ${https_proxy}
       no_proxy: ${no_proxy}
       llm_download: "True"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:9009/v1/health_check || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 10
     volumes:
       - "/home/$USER/.cache:/home/user/.cache" # RECOMMENDED: use cache to avoid download
       - video-llama-model:/home/user/model

comps/third_parties/video-llama/src/requirements.txt

@@ -12,7 +12,7 @@ iopath
 langchain
 langchain-community
 langchain-core
-numpy==1.26.4
+numpy
 omegaconf
 opencv-python-headless
 opentelemetry-api
@@ -26,12 +26,11 @@ sentence-transformers
 sentencepiece
 shortuuid
 timm
-torch==1.13.1 --index-url https://download.pytorch.org/whl/cpu
-torchaudio==0.13.1 --index-url https://download.pytorch.org/whl/cpu
-torchvision==0.14.1 --index-url https://download.pytorch.org/whl/cpu
-transformers==4.47.1
+torch==2.5.1 --index-url https://download.pytorch.org/whl/cpu
+torchaudio~=2.5.1 --index-url https://download.pytorch.org/whl/cpu
+torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cpu
+transformers
 uvicorn
 validators
 webdataset
 werkzeug
-

tests/lvms/test_lvms_video_llama.sh

@@ -15,20 +15,19 @@ function build_docker_images() {
     cd $WORKPATH
     echo $(pwd)
     docker build --no-cache --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -t opea/lvm-video-llama:comps --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f comps/third_parties/video-llama/src/Dockerfile .
-    if $? ; then
+    if [ $? -ne 0 ]; then
         echo "opea/lvm-video-llama built fail"
         exit 1
     else
         echo "opea/lvm-video-llama built successful"
     fi
     docker build --no-cache --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -t opea/lvm:comps --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy  -f comps/lvms/src/Dockerfile .
-    if $? ; then
+    if [ $? -ne 0 ]; then
         echo "opea/lvm built fail"
         exit 1
     else
         echo "opea/lvm built successful"
     fi
-
 }
 
 function start_service() {
@@ -73,7 +72,6 @@ function main() {
 
     stop_docker
     echo y | docker system prune
-
 }
 
 main