diff --git a/Dockerfile b/Dockerfile
index 5e4efd1..59b3943 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -10,7 +10,7 @@ RUN ldconfig /usr/local/cuda-13.0/compat/
 
 # Install vLLM with FlashInfer - use CUDA 130 PyTorch wheels
 RUN uv pip install --system "packaging>=24.2" && \
-    uv pip install --system "vllm[flashinfer]==0.20.2" && \
+    uv pip install --system "vllm[flashinfer]==0.21.0" && \
     uv pip install --system git+https://github.com/deepseek-ai/DeepGEMM.git@714dd1a4a980f7937a74343d19a8eba4fe321480 --no-build-isolation
 
 # Install additional Python dependencies (after vLLM to avoid PyTorch version conflicts)
diff --git a/builder/requirements.txt b/builder/requirements.txt
index 19fceca..b05ff13 100644
--- a/builder/requirements.txt
+++ b/builder/requirements.txt
@@ -3,7 +3,7 @@ pandas
 pyarrow
 runpod==1.9.0
 huggingface-hub
-lmcache==0.4.5
+lmcache==0.4.6
 packaging>=24.2
 typing-extensions>=4.8.0
 pydantic