From 88490e9551025b7725d63fd128864b46112dd822 Mon Sep 17 00:00:00 2001
From: velaraptor-runpod <chris.vela@runpod.io>
Date: Thu, 9 Jul 2026 18:22:40 -0500
Subject: [PATCH] feat: update to 0.23.0

---
 Dockerfile                 | 3 ++-
 configs/qwen/qwen3_8b.yaml | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/Dockerfile b/Dockerfile
index 9c56e20..9b99034 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -16,7 +16,8 @@ ENV PATH="/root/.local/bin:$PATH"
 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" && \
+RUN --mount=type=cache,target=/root/.cache/uv \
+    uv pip install --system "packaging>=24.2" && \
     uv pip install --system "vllm[flashinfer]==0.23.0" && \
     uv pip install --system git+https://github.com/deepseek-ai/DeepGEMM.git@714dd1a4a980f7937a74343d19a8eba4fe321480 --no-build-isolation && \
     uv pip install --system --force-reinstall --no-deps nixl-cu13
diff --git a/configs/qwen/qwen3_8b.yaml b/configs/qwen/qwen3_8b.yaml
index 0be2b36..8d6dea5 100644
--- a/configs/qwen/qwen3_8b.yaml
+++ b/configs/qwen/qwen3_8b.yaml
@@ -7,4 +7,6 @@ quantization: fp8
 kv-cache-dtype: fp8
 enforce-eager: false
 enable-prefix-caching: true
+vllm-release: v2.22.5 
+compilation-config: '{"cudagraph_mode": "PIECEWISE"}'
 speculative-config: '{"model":"RedHatAI/Qwen3-8B-speculator.eagle3","method":"eagle3","num_speculative_tokens":3}'