Issues

[vllm][torch 2.15] Elastic EP scaling: process group not initialized in the world group map under torch nightly

pytorch/test-infra#9040 · opened

on Oct 6, 2026

[vllm][torch 2.15] B200 V1 Attention shards time out under torch nightly (~62 min on baseline)

pytorch/test-infra#9039 · opened

on Oct 6, 2026

[vllm][torch 2.15] CUDA driver initialization failed in worker processes across ~100 jobs under torch nightly

pytorch/test-infra#9038 · opened

on Oct 6, 2026

1

[vllm][torch 2.15][vllm-side] Triton autotuner rejects Kimi K3 KDA kernel key 'NC'

pytorch/test-infra#9028 · opened

on Oct 6, 2026

[vllm][torch 2.15][vllm-side] Triton removed tl.make_block_ptr; vLLM Triton kernels fail to compile

pytorch/test-infra#9027 · opened

on Oct 6, 2026

[CRCR] Health degrades when pytorch/crcr-test exceeds the callback rate limit

pytorch/test-infra#9019 · opened

on Oct 6, 2026

1

[vllm][torch 2.15] Triton bundled with torch nightly removed tl.make_block_ptr

pytorch/test-infra#9017 · opened

on Oct 6, 2026

[vllm][torch 2.15] NCCL communicator init fails with 'unhandled cuda error' on the new ProcessGroupNCCL2 creation path

pytorch/test-infra#9006 · opened

on Oct 5, 2026

5

[test hud] Update run_tests to output test run rows with platform info at runtime.

pytorch/test-infra#9005 · opened

on Oct 5, 2026

[bug] fix double counting in all_test_runs ingestion

pytorch/test-infra#9004 · opened

on Oct 5, 2026

Request macOS 27 bare-metal runners for MPS tests (macos-m2-27)

pytorch/test-infra#8964 · opened

on Oct 1, 2026

10

[CRCR] Keep qualifying nightly SHA rows complete across the selected time range

pytorch/test-infra#8960 · opened

on Oct 1, 2026

Limit doctests to one job

pytorch/test-infra#8950 · opened

on Sep 30, 2026

GreenLight - Mark pr with labels to make easy to search by evaluated, approved, etc

pytorch/test-infra#8948 · opened

on Sep 30, 2026

GreenLight - give users the possibility to join GL beta by adding themselves to an ci-infra issue

pytorch/test-infra#8947 · opened

on Sep 30, 2026

[DO NOT CLOSE] [GreenLight] Beta Users List

pytorch/test-infra#8945 · opened

on Sep 30, 2026

6

[vllm][torch 2.15] vLLM worker startup fails free-memory check: ~2.9 GiB already used on the GPU before init under torch nightly

pytorch/test-infra#8940 · opened

on Sep 29, 2026

2

[vllm][torch 2.15][vllm-side] MiniMax reduce-RMS kernels: cudaIpcMemHandle_t has no 'reserved' attribute

pytorch/test-infra#8931 · opened

on Sep 28, 2026

4

Review the policy based on 26-09-21 meeting feedback

pytorch/test-infra#8930 · opened

on Sep 28, 2026

[vllm][torch 2.15][vllm-side] nixl_ep has no build for the torch 2.15 ABI (nixl_ep_cpp_torch215 missing)

pytorch/test-infra#8899 · opened

on Sep 25, 2026

8

add doc to track various killswitches for devinfra

pytorch/test-infra#8891 · opened

on Sep 24, 2026

[vllm][torch 2.15] DeepEP low-latency workers SIGABRT after NVSHMEM IBGDA transport init fails on H100

pytorch/test-infra#8890 · opened

on Sep 24, 2026

[GreenLight policy] Wrong NO_LAND verdict on pytorch/pytorch#198501 (adffa15)

pytorch/test-infra#8885 · opened

on Sep 24, 2026

1

[CRCR][L3] Group CRCR downstream failures by repo in the Dr. CI comment

pytorch/test-infra#8880 · opened

on Sep 24, 2026

[vllm][torch 2.15] AsyncTP fusion finds zero matches on the first SP compile-range entries on torch nightly

pytorch/test-infra#8870 · opened

on Sep 23, 2026

12

Navigation Menu

Repository navigation

Issues

[vllm][torch 2.15] Elastic EP scaling: process group not initialized in the world group map under torch nightly

[vllm][torch 2.15] B200 V1 Attention shards time out under torch nightly (~62 min on baseline)

[vllm][torch 2.15] CUDA driver initialization failed in worker processes across ~100 jobs under torch nightly

[vllm][torch 2.15][vllm-side] Triton autotuner rejects Kimi K3 KDA kernel key 'NC'

[vllm][torch 2.15][vllm-side] Triton removed tl.make_block_ptr; vLLM Triton kernels fail to compile

[CRCR] Health degrades when pytorch/crcr-test exceeds the callback rate limit

[vllm][torch 2.15] Triton bundled with torch nightly removed tl.make_block_ptr

[vllm][torch 2.15] NCCL communicator init fails with 'unhandled cuda error' on the new ProcessGroupNCCL2 creation path

[test hud] Update run_tests to output test run rows with platform info at runtime.

[bug] fix double counting in all_test_runs ingestion

Request macOS 27 bare-metal runners for MPS tests (macos-m2-27)

[CRCR] Keep qualifying nightly SHA rows complete across the selected time range

Limit doctests to one job

GreenLight - Mark pr with labels to make easy to search by evaluated, approved, etc

GreenLight - give users the possibility to join GL beta by adding themselves to an ci-infra issue

[DO NOT CLOSE] [GreenLight] Beta Users List

[vllm][torch 2.15] vLLM worker startup fails free-memory check: ~2.9 GiB already used on the GPU before init under torch nightly

[vllm][torch 2.15][vllm-side] MiniMax reduce-RMS kernels: cudaIpcMemHandle_t has no 'reserved' attribute

Review the policy based on 26-09-21 meeting feedback

[vllm][torch 2.15][vllm-side] nixl_ep has no build for the torch 2.15 ABI (nixl_ep_cpp_torch215 missing)

add doc to track various killswitches for devinfra

[vllm][torch 2.15] DeepEP low-latency workers SIGABRT after NVSHMEM IBGDA transport init fails on H100

[GreenLight policy] Wrong NO_LAND verdict on pytorch/pytorch#198501 (adffa15)

[CRCR][L3] Group CRCR downstream failures by repo in the Dr. CI comment

[vllm][torch 2.15] AsyncTP fusion finds zero matches on the first SP compile-range entries on torch nightly