feat: update 0.16.0, add lmcache

Dockerfile

@@ -1,20 +1,21 @@
 FROM nvidia/cuda:12.9.1-base-ubuntu22.04 
 
 RUN apt-get update -y \
-    && apt-get install -y python3-pip
+    && apt-get install -y python3-pip curl \
+    && curl -LsSf https://astral.sh/uv/install.sh | sh
+
+ENV PATH="/root/.local/bin:$PATH"
 
 RUN ldconfig /usr/local/cuda-12.9/compat/
 
-# Install vLLM with FlashInfer - use CUDA 12.8 PyTorch wheels (compatible with vLLM 0.15.1)
+# Install vLLM with FlashInfer - use CUDA 12.9 PyTorch wheels
-RUN python3 -m pip install --upgrade pip && \
+RUN uv pip install --system "packaging>=24.2" && \
-    python3 -m pip install "vllm[flashinfer]==0.16.0" --extra-index-url https://download.pytorch.org/whl/cu129
+    uv pip install --system "vllm[flashinfer]==0.16.0" --extra-index-url https://download.pytorch.org/whl/cu129
-
-
 
 # Install additional Python dependencies (after vLLM to avoid PyTorch version conflicts)
 COPY builder/requirements.txt /requirements.txt
-RUN --mount=type=cache,target=/root/.cache/pip \
+RUN --mount=type=cache,target=/root/.cache/uv \
-    python3 -m pip install --upgrade -r /requirements.txt
+    uv pip install --system -r /requirements.txt
 
 # Setup for Option 2: Building the Image with the Model included
 ARG MODEL_NAME=""
@@ -24,6 +25,7 @@ ARG QUANTIZATION=""
 ARG MODEL_REVISION=""
 ARG TOKENIZER_REVISION=""
 ARG VLLM_NIGHTLY="false"
+ARG LMCACHE="false"
 
 ENV MODEL_NAME=$MODEL_NAME \
     MODEL_REVISION=$MODEL_REVISION \
@@ -45,10 +47,14 @@ ENV MODEL_NAME=$MODEL_NAME \
 
 ENV PYTHONPATH="/:/vllm-workspace"
 
+RUN if [ "${LMCACHE}" = "true" ]; then \
+    uv pip install --system lmcache; \
+fi
+
 RUN if [ "${VLLM_NIGHTLY}" = "true" ]; then \
-    pip install -U vllm --pre --index-url https://pypi.org/simple --extra-index-url https://wheels.vllm.ai/nightly && \
+    uv pip install --system -U vllm --pre --index-url https://pypi.org/simple --extra-index-url https://wheels.vllm.ai/nightly && \
     apt-get update && apt-get install -y git && rm -rf /var/lib/apt/lists/* && \
-    pip install git+https://github.com/huggingface/transformers.git; \
+    uv pip install --system git+https://github.com/huggingface/transformers.git; \
 fi
 
 COPY src /src

src/engine_args.py

@@ -401,6 +401,17 @@ def get_engine_args():
         if os.getenv("MAX_PARALLEL_LOADING_WORKERS"):
             logging.warning("Overriding MAX_PARALLEL_LOADING_WORKERS with None because more than 1 GPU is available.")
     
+    # LMCache requires HMA to be disabled
+    _kv_transfer = args.get("kv_transfer_config")
+    _kv_offload = args.get("kv_offloading_backend")
+    _lmcache_active = _kv_offload == "lmcache" or (
+        isinstance(_kv_transfer, dict)
+        and "lmcache" in str(_kv_transfer.get("kv_connector", "")).lower()
+    )
+    if _lmcache_active and not args.get("disable_hybrid_kv_cache_manager"):
+        args["disable_hybrid_kv_cache_manager"] = True
+        logging.info("LMCache detected: automatically setting disable_hybrid_kv_cache_manager=True")
+
     # Deprecated env args backwards compatibility
     if args.get("kv_cache_dtype") == "fp8_e5m2":
         args["kv_cache_dtype"] = "fp8"