new dockerfile that is using new VLLM repository as old one is depriciated

deployment/pipelines/chatqa/resources-model-hpu.yaml

@@ -240,7 +240,7 @@ modelConfigs:
       VLLM_DECODE_BLOCK_BUCKET_MAX: "8256"
       VLLM_ALLOW_LONG_MAX_MODEL_LEN: 1
       VLLM_ENABLE_EXPERT_PARALLEL: "0"
-    extraCmdArgs: ["--block-size","128","--dtype","bfloat16","--max-model-len","33024","--gpu-memory-util","0.99","--max-num-seqs","64","--max-num-prefill-seqs","16","--num_scheduler_steps","16","--use-padding-aware-scheduling"]
+    extraCmdArgs: ["--block-size","128","--dtype","bfloat16","--gpu-memory-util","0.99","--max-num-seqs","64"]
     tensor_parallel_size: "2"
     modelChatTemplate: |
       {%- if messages[0]['role'] == 'system' %}

src/comps/llms/impl/model_server/vllm/docker/hpu/Dockerfile

@@ -1,29 +1,128 @@
-# Copyright (C) 2024-2025 Intel Corporation
+# Copyright (C) 2025 Habana Labs, Ltd. an Intel Company
 # SPDX-License-Identifier: Apache-2.0
 
-FROM vault.habana.ai/gaudi-docker/1.22.1/ubuntu22.04/habanalabs/pytorch-installer-2.7.1:latest
+# ------------------------------
+# Base image parameters
+# ------------------------------
+ARG DOCKER_URL=vault.habana.ai/gaudi-docker
+ARG VERSION=1.22.2
+ARG BASE_NAME=ubuntu24.04
+ARG PT_VERSION=2.7.1
+ARG REVISION=latest
+ARG REPO_TYPE=habanalabs
+ARG TORCH_TYPE_SUFFIX
 
+FROM ${DOCKER_URL}/${VERSION}/${BASE_NAME}/${REPO_TYPE}/pytorch-${TORCH_TYPE_SUFFIX}installer-${PT_VERSION}:${REVISION}
+
+# Use bash and strict mode in RUN
+SHELL ["/bin/bash", "-euo", "pipefail", "-c"]
+
+# ------------------------------
+# General environment
+# ------------------------------
+ENV OMPI_MCA_btl_vader_single_copy_mechanism=none
+ENV HABANA_VISIBLE_DEVICES=all
 ENV LANG=en_US.UTF-8
 
-# create user and folders
-RUN useradd -u 1000 -m -s /bin/bash user
-ENV PATH="$PATH:/home/user/.local/bin"
-WORKDIR /home/user/
+# Ensure `python` resolves to Python 3
+RUN ln -sf /usr/bin/python3 /usr/bin/python
+
+# Optional: avoid dash for /bin/sh scripts
+RUN echo "dash dash/sh boolean false" | debconf-set-selections && \
+    DEBIAN_FRONTEND=noninteractive dpkg-reconfigure dash
+
+# Build-time system dependencies
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+      git gettext moreutils jq ca-certificates && \
+    rm -rf /var/lib/apt/lists/*
+
+# ------------------------------
+# Create non-root user for runtime
+# ------------------------------
+ARG APP_USER=appuser
+ENV HOME=/home/${APP_USER}
+ENV PATH="${HOME}/.local/bin:${PATH}"
+
+# ------------------------------
+# Prepare directories inside the image
+# ------------------------------
+# /opt/vllm-project and /opt/vllm-gaudi hold cloned sources.
+# /opt/app/scripts will host app-level scripts copied from vllm-gaudi/.cd
+RUN mkdir -p /opt/vllm-project /opt/vllm-gaudi /opt/app/scripts
+
+# ------------------------------
+# vLLM + vllm-gaudi clone & install
+# ------------------------------
+ENV VLLM_PATH=/opt/vllm-project
+ENV VLLM_PATH2=/opt/vllm-gaudi
+ARG VLLM_GAUDI_COMMIT=v0.11.2
+# Leave empty to auto-resolve from vllm-gaudi marker
+ARG VLLM_PROJECT_COMMIT=v0.11.2
 
-ENV VLLM_TARGET_DEVICE="hpu"
-# VLLM doesn't share any prebuilt HPU packages (https://docs.vllm.ai/en/latest/getting_started/installation/cpu.html#pre-built-wheels)
-# So it is impossible to easily freeze the whole vllm environment with uv
-RUN pip install --upgrade pip==25.0.1
+RUN \
+  # Clone vllm-gaudi
+  git clone https://github.com/vllm-project/vllm-gaudi.git "${VLLM_PATH2}" && \
+  cd "${VLLM_PATH2}" && \
+  git checkout "${VLLM_GAUDI_COMMIT}" && \
+  # Fetch marker branch that contains last-good commit file
+  git fetch origin "vllm/last-good-commit-for-vllm-gaudi" || true && \
+  # Resolve vLLM commit (explicit or last-good) and trim whitespace
+  if [[ -z "${VLLM_PROJECT_COMMIT}" ]]; then \
+    VLLM_PROJECT_COMMIT="$(git show 'origin/vllm/last-good-commit-for-vllm-gaudi:VLLM_STABLE_COMMIT' 2>/dev/null | tr -d ' \t\r\n')"; \
+    echo "Using last-good vLLM commit: ${VLLM_PROJECT_COMMIT}"; \
+  else \
+    echo "Using explicit vLLM commit: ${VLLM_PROJECT_COMMIT}"; \
+  fi && \
+  [[ -n "${VLLM_PROJECT_COMMIT}" ]] || (echo "ERROR: could not resolve VLLM_PROJECT_COMMIT" && exit 1) && \
+  # Clone vllm
+  git clone https://github.com/vllm-project/vllm.git "${VLLM_PATH}" && \
+  cd "${VLLM_PATH}" && \
+  git fetch origin --tags || true && \
+  git checkout "${VLLM_PROJECT_COMMIT}" && \
+  # Install vLLM build dependencies excluding torch
+  pip install --no-cache-dir -r <(sed '/^torch/d' requirements/build.txt) && \
+  # Install vLLM itself (empty target to avoid device-specific builds)
+  VLLM_TARGET_DEVICE=empty pip install --no-cache-dir --no-build-isolation . && \
+  # Install vllm-gaudi plugin for HPU
+  cd "${VLLM_PATH2}" && \
+  VLLM_TARGET_DEVICE=hpu pip install -v --no-cache-dir . --no-build-isolation
 
-RUN git clone -b v0.9.0.1+Gaudi-1.22.0 --single-branch https://github.com/HabanaAI/vllm-fork.git
-WORKDIR /home/user/vllm-fork
-RUN pip install --upgrade pip && \
-    pip install -v -r requirements-hpu.txt
+# ------------------------------
+# Copy app-like assets from vllm-gaudi/.cd (inside the image)
+# We DON'T use Docker COPY here, because these folders come from the repo we cloned during build.
+# ------------------------------
+RUN \
+  # Create target subfolders
+  mkdir -p /opt/app/scripts/templates \
+           /opt/app/scripts/entrypoints \
+           /opt/app/scripts/server \
+           /opt/app/scripts/benchmark && \
+  # Copy from vllm-gaudi's internal .cd folder into our app path
+  cp -r "${VLLM_PATH2}/.cd/templates/"   /opt/app/scripts/templates/ && \
+  cp -r "${VLLM_PATH2}/.cd/entrypoints/" /opt/app/scripts/entrypoints/ && \
+  cp -r "${VLLM_PATH2}/.cd/server/"      /opt/app/scripts/server/ && \
+  cp -r "${VLLM_PATH2}/.cd/benchmark/"   /opt/app/scripts/benchmark/
 
+# Make scripts importable
+ENV PYTHONPATH=/opt/app/scripts:${PYTHONPATH}
+WORKDIR /opt/app/scripts
 
-RUN VLLM_TARGET_DEVICE=hpu python3 setup.py install
+# ------------------------------
+# Optional: install extra Python packages you need
+# ------------------------------
+RUN pip install --no-cache-dir datasets pandas
 
-USER user
-WORKDIR /home/user/
+# ------------------------------
+# Switch to non-root user for runtime
+# ------------------------------
+USER ${APP_USER}
+ENV PYTHONDONTWRITEBYTECODE=1
 
+# ------------------------------
+# Entrypoint
+# If you have a specific server module in /opt/app/scripts (from .cd),
+# adjust this to what you actually want to run. Otherwise, keep a shell.
+# ------------------------------
+# ENTRYPOINT ["python3", "-m", "entrypoints.entrypoint_main"]
 CMD ["/bin/bash"]