p

Signed-off-by: khluu <khluu000@gmail.com>

.buildkite/scripts/generate-and-upload-nightly-index.sh

@@ -53,8 +53,10 @@ echo "Uploading indices to $S3_COMMIT_PREFIX"
 aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "$S3_COMMIT_PREFIX"
 
 # copy to /nightly/ only if it is on the main branch and not a PR
-echo "Uploading indices to overwrite /nightly/"
-aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "s3://$BUCKET/nightly/"
+if [[ "$BUILDKITE_BRANCH" == "main" && "$BUILDKITE_PULL_REQUEST" == "false" ]]; then
+    echo "Uploading indices to overwrite /nightly/"
+    aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "s3://$BUCKET/nightly/"
+fi
 
 # detect version from any wheel in the commit directory
 # download the first wheel we find to extract version metadata

.buildkite/test_areas/kernels.yaml

@@ -29,6 +29,7 @@ steps:
   - vllm/v1/attention
     # TODO: remove this dependency (https://github.com/vllm-project/vllm/issues/32267)
   - vllm/model_executor/layers/attention
+  - vllm/utils/flashinfer.py
   - tests/kernels/attention
   commands:
     - pytest -v -s kernels/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT

docker/Dockerfile

@@ -689,6 +689,10 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 . /etc/environment && \
 uv pip list
 
+# Pin transformers to 5.5.0, overwriting the version from dependencies
+RUN --mount=type=cache,target=/root/.cache/uv \
+    uv pip install --system "transformers==5.5.0"
+
 # Install deepgemm wheel that has been built in the `build` stage
 RUN --mount=type=cache,target=/root/.cache/uv \
     --mount=type=bind,from=build,source=/tmp/deepgemm/dist,target=/tmp/deepgemm/dist,ro \

docker/Dockerfile.rocm

@@ -330,8 +330,7 @@ RUN --mount=type=bind,from=export_vllm,src=/,target=/install \
     && uv pip install --system -r requirements/rocm.txt \
     && uv pip install --system -r requirements/rocm-test.txt \
     && pip uninstall -y vllm \
-    && uv pip install --system *.whl \
-    && uv pip install --system "transformers==5.5.0"
+    && uv pip install --system *.whl
 
 # Verify that PyTorch is the ROCm build, not CUDA
 RUN python3 -c "import torch; assert torch.version.hip is not None, \
@@ -428,8 +427,7 @@ RUN --mount=type=bind,from=export_vllm,src=/,target=/install \
     cd /install \
     && uv pip install --system -r requirements/rocm.txt \
     && pip uninstall -y vllm \
-    && uv pip install --system *.whl \
-    && uv pip install --system "transformers==5.5.0"
+    && uv pip install --system *.whl
 
 ARG COMMON_WORKDIR
 ARG BASE_IMAGE

tests/v1/kv_connector/unit/test_nixl_connector.py

@@ -1613,7 +1613,7 @@ def test_register_kv_caches(
                                 )
                             ]
                         ],
-                        cache_dtype=torch.bfloat16,
+                        cache_dtype="bfloat16",
                         device=torch.accelerator.current_device_index(),
                         kernel_block_sizes=[block_size],
                     )

vllm/tool_parsers/gemma4_tool_parser.py

@@ -38,7 +38,7 @@ from vllm.entrypoints.openai.responses.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
-from vllm.tool_parsers.abstract_tool_parser import ToolParser
+from vllm.tool_parsers.abstract_tool_parser import Tool, ToolParser
 from vllm.tool_parsers.utils import find_common_prefix
 
 logger = init_logger(__name__)
@@ -281,8 +281,8 @@ class Gemma4ToolParser(ToolParser):
     tool parsers.
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         if not self.model_tokenizer:
             raise ValueError(