Repository navigation
ggml kernels: GQ_FUSED_ROWS in every build, the CUDA chain's host side checked without AVX2 #4425
ci.yml
on: pull_request
Engine (Linux, CPU)
17s
Container image
15s
Sanitizers (ASan + UBSan)
17s
Vulkan (Lavapipe, software)
17s
musl libc (Alpine)
15s
CUDA syntax check
17s
Qwen3.6 tiny oracle (token-exact + ASan/UBSan)
17s
OLMoE tiny oracle (token-exact + dashboard HITS)
17s
DeepSeek V4.1 tiny oracle (engram + DSA + fp4 experts + vision)
17s
Qwen3.8 tiny oracle (GDN + QSA + PLE + MoE)
17s
Laya tiny oracle (vs the laya package) and POST /v1/systemone end to end
17s
GLiNER2.5-Decide tiny oracle (vs the gliner2 package) and POST /v1/systemone end to end
17s
Clef tiny oracle (vs joint_schema_model.py) and POST /v1/systemone on qwen36
17s
TypeSafe's SDKs (Python, TypeScript) against coli serve, base URL changed only
17s
Qwen-Image tiny oracle (text encoder, DiT, scheduler, VAE vs diffusers)
17s
Inkling oracle (token-exact vs transformers)
17s
Efficiency suite (tiny oracle)
17s
Cluster parity gate (fmt6/fmt4, token-exact)
17s
DeepSeek V4.1 cluster parity gate (token-exact)
17s
HIP syntax check
17s
CUDA build (Windows, MSVC host)
17s
Web UI
17s
Python tests
16s
GGUF reader + converter tests
17s
Python tests (Windows focused)
17s
ARM (engines + NEON kernel exactness + tiny oracle)
17s
Matrix: engines-all-platforms
Matrix: vulkan-engines
Annotations
109 errors