Issues

[vllm][torch 2.16][vllm-side] CUDA image build fails: DeepEP build NCCL override (nvidia-nccl-cu13==2.30.7) breaks torch nightly import with undefined symbol ncclReduceConfig

pytorch/test-infra#9142 · opened

on Oct 16, 2026

[GreenLight policy] Wrong NO_LAND verdict on pytorch/pytorch#201160 (57c7ecc)

pytorch/test-infra#9140 · opened

on Oct 16, 2026

Handle Windows timeouts for reports

pytorch/test-infra#9133 · opened

on Oct 15, 2026

GreenLight - review 3 reported PRs and update the skill to widen the accept cases

pytorch/test-infra#9123 · opened

on Oct 14, 2026

GreenLight - Optimize review start time

pytorch/test-infra#9122 · opened

on Oct 14, 2026

[vllm][torch 2.16][vllm-side] Triton launcher warmup test calls AsyncCompileMode.submit() without new cleanup_fn argument

pytorch/test-infra#9121 · opened

on Oct 14, 2026

[vllm][torch 2.16][vllm-side] Intel BMG: online-quantization test runs a modelopt_fp4 case that XPU does not support

pytorch/test-infra#9120 · opened

on Oct 14, 2026

1

[vllm][torch 2.16][vllm-side] nixl_ep has no extension built for torch 2.16 (ModuleNotFoundError nixl_ep_cu13.nixl_ep_cpp_torch216)

pytorch/test-infra#9119 · opened

on Oct 14, 2026

[test hud] map C++ test binary and test file representations

pytorch/test-infra#9108 · opened

on Oct 13, 2026

1

[test hud] make stdout/stderr easily accessible in jsonl

pytorch/test-infra#9107 · opened

on Oct 13, 2026

[test hud][tentative] batch jsonl land and inserts to lower ClickHouse write pressure OR reduce ClickHouse memory pressure and dropped writes

pytorch/test-infra#9106 · opened

on Oct 13, 2026

[test hud] eliminate rocm alternative upload path to avoid duplicate insertions

pytorch/test-infra#9105 · opened

on Oct 13, 2026

[test hud] roll out jsonl reporter beyond pull/trunk targets and increase sampling

pytorch/test-infra#9104 · opened

on Oct 13, 2026

2

[vllm][torch 2.15] B200 DeepEP low-latency: NVSHMEM RC queue-pair setup fails with errno 22 (Invalid argument) on mlx5_12, intermittently

pytorch/test-infra#9101 · opened

on Oct 13, 2026

1

[torch 2.16] vLLM CI failures - torch nightly triage

pytorch/test-infra#9099 · opened

on Oct 13, 2026

[GreenLight policy] Wrong NO_LAND verdict on pytorch/pytorch#200725 (91b5f9e)

pytorch/test-infra#9098 · opened

on Oct 13, 2026

1

Use ci.owners for flakybot

pytorch/test-infra#9097 · opened

on Oct 12, 2026

[GreenLight policy] Wrong NO_LAND veredict on pytorch/pytorch#200126

pytorch/test-infra#9093 · opened

on Oct 12, 2026

1

[test hud] add nvidia-ml-py to PyTorch CI to probe device capabilities during test time

pytorch/test-infra#9079 · opened

on Oct 9, 2026

[vllm][torch 2.15][vllm-side] test_with_ngram_gpu_spec_decoding: preemption configs flake at the 5% acceptance-drop floor (also on main)

pytorch/test-infra#9075 · opened

on Oct 8, 2026

1

[vllm][torch 2.15][vllm-side] Triton 3.9 AsyncCompileMode.submit() requires cleanup_fn; vLLM JIT-warmup test still uses the old signature

pytorch/test-infra#9074 · opened

on Oct 8, 2026

2

[vllm][torch 2.15][needs-investigation] Nightly CI image missing from ECR; agent environment hook fails before any test runs

pytorch/test-infra#9055 · opened

on Oct 7, 2026

2

[vllm][torch 2.15] vLLM startup free-memory check fails: GPU already partly occupied when the engine starts

pytorch/test-infra#9052 · opened

on Oct 7, 2026

2

[vllm][torch 2.15] Process aborts with nanobind 'duplicate key PAD_ZERO' when tokenspeed_triton loads next to Triton 3.9 libtriton

pytorch/test-infra#9050 · opened

on Oct 7, 2026

6

[vllm][torch 2.16] Qwen3.5-397B NVFP4 DEP2-MTP GSM8K collapses to 0.0 after worker CUDA OOM on B200 under torch nightly

pytorch/test-infra#9049 · opened

on Oct 7, 2026

4

Navigation Menu

Repository navigation

Issues

[vllm][torch 2.16][vllm-side] CUDA image build fails: DeepEP build NCCL override (nvidia-nccl-cu13==2.30.7) breaks torch nightly import with undefined symbol ncclReduceConfig

[GreenLight policy] Wrong NO_LAND verdict on pytorch/pytorch#201160 (57c7ecc)

Handle Windows timeouts for reports

GreenLight - review 3 reported PRs and update the skill to widen the accept cases

GreenLight - Optimize review start time

[vllm][torch 2.16][vllm-side] Triton launcher warmup test calls AsyncCompileMode.submit() without new cleanup_fn argument

[vllm][torch 2.16][vllm-side] Intel BMG: online-quantization test runs a modelopt_fp4 case that XPU does not support

[vllm][torch 2.16][vllm-side] nixl_ep has no extension built for torch 2.16 (ModuleNotFoundError nixl_ep_cu13.nixl_ep_cpp_torch216)

[test hud] map C++ test binary and test file representations

[test hud] make stdout/stderr easily accessible in jsonl

[test hud][tentative] batch jsonl land and inserts to lower ClickHouse write pressure OR reduce ClickHouse memory pressure and dropped writes

[test hud] eliminate rocm alternative upload path to avoid duplicate insertions

[test hud] roll out jsonl reporter beyond pull/trunk targets and increase sampling

[vllm][torch 2.15] B200 DeepEP low-latency: NVSHMEM RC queue-pair setup fails with errno 22 (Invalid argument) on mlx5_12, intermittently

[torch 2.16] vLLM CI failures - torch nightly triage

[GreenLight policy] Wrong NO_LAND verdict on pytorch/pytorch#200725 (91b5f9e)

Use ci.owners for flakybot

[GreenLight policy] Wrong NO_LAND veredict on pytorch/pytorch#200126

[test hud] add nvidia-ml-py to PyTorch CI to probe device capabilities during test time

[vllm][torch 2.15][vllm-side] test_with_ngram_gpu_spec_decoding: preemption configs flake at the 5% acceptance-drop floor (also on main)

[vllm][torch 2.15][vllm-side] Triton 3.9 AsyncCompileMode.submit() requires cleanup_fn; vLLM JIT-warmup test still uses the old signature

[vllm][torch 2.15][needs-investigation] Nightly CI image missing from ECR; agent environment hook fails before any test runs

[vllm][torch 2.15] vLLM startup free-memory check fails: GPU already partly occupied when the engine starts

[vllm][torch 2.15] Process aborts with nanobind 'duplicate key PAD_ZERO' when tokenspeed_triton loads next to Triton 3.9 libtriton

[vllm][torch 2.16] Qwen3.5-397B NVFP4 DEP2-MTP GSM8K collapses to 0.0 after worker CUDA OOM on B200 under torch nightly