Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

ci: build portable llama.cpp release binaries (#3037)

zhifu gao0f655f27ea99
diff --git a/.github/workflows/build-llamacpp-binaries.yml b/.github/workflows/build-llamacpp-binaries.yml
index 1624b9292..fc28880ab 100644
--- a/.github/workflows/build-llamacpp-binaries.yml
+++ b/.github/workflows/build-llamacpp-binaries.yml
@@ -33,7 +33,23 @@ jobs:
- uses: actions/checkout@v4
- name: Configure and build
run: |
- cmake -B build -DCMAKE_BUILD_TYPE=Release
+ # Release binaries must be portable across user machines, not tuned to
+ # the ephemeral CI runner CPU. A native ggml build may emit AVX512 or
+ # AVX-VNNI instructions and then crash with SIGILL on ordinary CPUs.
+ # Keep the prebuilt x64 package conservative; optimized AVX2/AVX512
+ # builds can be published as separate assets later.
+ cmake -B build -DCMAKE_BUILD_TYPE=Release \
+ -DGGML_NATIVE=OFF \
+ -DGGML_AVX=OFF \
+ -DGGML_AVX2=OFF \
+ -DGGML_AVX_VNNI=OFF \
+ -DGGML_AVX512=OFF \
+ -DGGML_AVX512_VBMI=OFF \
+ -DGGML_AVX512_VNNI=OFF \
+ -DGGML_AVX512_BF16=OFF \
+ -DGGML_FMA=OFF \
+ -DGGML_F16C=OFF \
+ -DGGML_BMI2=OFF
cmake --build build --config Release -j 2
- name: Package
run: |