@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 980796cd07
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

[CI/Build][CPU] Fix flaky CI image build failure and unexpected warnings (#44852)

Signed-off-by: jiang1.li <jiang1.li@intel.com>

Li, Jiang committed Jun 8, 2026 at 19:10 UTC 980796cd075cb8e2591352a725e8adb668152aab
2 files changed +5 -4
docker/Dockerfile.cpu
+3 -3
@@ -118,9 +118,9 @@ ENV CARGO_BUILD_JOBS=4
118 # Build the release binary. Cache cargo registry/git and target/, but copy the
119 # binary out of the target/ cache mount so it persists into the image layer
120 # for later COPY --from=rust-build.
121 -RUN --mount=type=cache,target=/root/.cargo/registry \
122 - --mount=type=cache,target=/root/.cargo/git \
123 - --mount=type=cache,target=/workspace/rust/target \
121 +RUN --mount=type=cache,target=/root/.cargo/registry,sharing=locked \
122 + --mount=type=cache,target=/root/.cargo/git,sharing=locked \
123 + --mount=type=cache,target=/workspace/rust/target,sharing=locked \
124 VLLM_RS_TARGET_PATH=/workspace/vllm-rs bash build_rust.sh
125
126 ######################### BUILD IMAGE #########################
vllm/_custom_ops.py
+2 -1
@@ -3537,10 +3537,11 @@ class CPUDNNLGEMMHandler:
3537 self.handler_tensor: torch.Tensor | None = None
3538 self.n = -1
3539 self.k = -1
3540 + self.dtor = torch.ops._C.release_dnnl_matmul_handler
3541
3542 def __del__(self):
3543 if self.handler_tensor is not None:
3543 - torch.ops._C.release_dnnl_matmul_handler(self.handler_tensor.item())
3544 + self.dtor(self.handler_tensor.item())
3545
3546
3547 _supports_onednn = bool(hasattr(torch.ops._C, "create_onednn_mm_handler"))