fix for new requirements structure

Dockerfile

@@ -213,7 +213,7 @@ RUN curl -fsL https://patch-diff.githubusercontent.com/raw/vllm-project/vllm/pul
          echo "PR 35568 already applied, skipping."; \
        else \
          echo "Applying PR 35568..."; \
-         git apply -v pr35568.diff; \
+         git apply -v --exclude="tests/*" pr35568.diff; \
        fi \
     && rm pr35568.diff
 
@@ -231,9 +231,9 @@ RUN curl -fsL https://patch-diff.githubusercontent.com/raw/vllm-project/vllm/pul
 RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
     python3 use_existing_torch.py && \
     sed -i "/flashinfer/d" requirements/cuda.txt && \
-    sed -i '/^triton\b/d' requirements/test.txt && \
-    sed -i '/^fastsafetensors\b/d' requirements/test.txt && \
-    uv pip install -r requirements/build.txt
+    sed -i '/^triton\b/d' requirements/test/cuda.txt && \
+    sed -i '/^fastsafetensors\b/d' requirements/test/cuda.txt && \
+    uv pip install -r requirements/build/cuda.txt
 
 # Apply Patches
 # TEMPORARY PATCH for fastsafetensors loading in cluster setup - tracking https://github.com/vllm-project/vllm/issues/34180