All issues

Search results

pr-size: main tracks ten unclassified .clientlog evidence files, so its own whole-tree sweep is red

Status: Open.

#3131 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 11, 2026

test_cpu_threadpool: the 100x dispatch ratio is calibrated on 20 cores and flakes on 4-vCPU CI runners

Status: Open.

#3130 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 11, 2026

fix(BACKEND-ROCM-QUANT-GATHER): label the comparison scope from the run, not a literal

Status: Open.

#3126 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 10, 2026

EXL3 GEMM: implement reconstruct+cuBLAS path for M>144 to close prefill rate gap

Status: Open.

#3124 In mudler/vllm.cpp;

· localai-org-maint-bot opened on Sep 10, 2026

test_backend_cross_device: MoeRouterTopK flakes intermittently on gfx1200

Status: Open.

#3121 In mudler/vllm.cpp;

· joral opened on Sep 10, 2026

test_backend_cross_device: causal conv1d fwd/update flakes intermittently on gfx1200

Status: Open.

#3119 In mudler/vllm.cpp;

· joral opened on Sep 10, 2026

fix(BACKEND-ROCM-BF16-MOE): mirror the primary BF16 LM-head output boundary

Status: Open.

#3116 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 9, 2026

fix(BACKEND-ROCM-BF16-MOE): match decode attention and Q/K preamble arithmetic

Status: Open.

#3115 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 9, 2026

fix(BACKEND-ROCM-QUANT-GATHER): reject unqualified model captures

Status: Open.

#3113 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 9, 2026

fix(BACKEND-ROCM-ATTN-GATE-SPLIT): identify the actual F16 head in the forward test

Status: Open.

#3112 In mudler/vllm.cpp;

· VikashLoomba opened on Sep 9, 2026

fix(BACKEND-GATE-ROCM-SGLANG): diagnose Qwen3-4B first-inference GPU hang on Strix

Status: Open.

#3111 In mudler/vllm.cpp;

· localai-org-maint-bot opened on Sep 9, 2026

test(BACKEND-GATE-ROCM-SGLANG): publish lifecycle diagnostic aggregate once

Status: Open.

#3110 In mudler/vllm.cpp;

· localai-org-maint-bot opened on Sep 9, 2026