Issues

• Distributed group_mean casts group ids to float32, losing precision vs non-distributed path

  OpenPipe/ART#1102 · Filter by author OnePunchMonk opened on Oct 3, 2026

• Nonfused column LoRA misses input-gradient SUM when TP>1 and sequence parallelism is disabled

  OpenPipe/ART#1091 · Filter by author bradhilton opened on Oct 3, 2026

• Clarify STOP authority on the native-token fast path without changing rendering

  OpenPipe/ART#961 · Filter by author bradhilton opened on Sep 25, 2026
  3

• TrainerRank CP2/EP1 normal admission reaches CUDA OOM in compiled MoE LoRA forward

  OpenPipe/ART#949 · Filter by author bradhilton opened on Sep 24, 2026
  5

• TrainerRank oversized admission reaches CUDA OOM during Qwen3.6-35B-A3B forward

  OpenPipe/ART#948 · Filter by author bradhilton opened on Sep 24, 2026

• Optional EvalPort adapter for Trajectory/TrajectoryGroup (portable eval-result interchange)

  OpenPipe/ART#935 · Filter by author adhabnr-ux opened on Sep 22, 2026

• Tokenization throughput: process pool capped at 4 workers, GIL-bound thread path (benchmark: caladan 059 batches)

  enhancement
  OpenPipe/ART#926 · Filter by author bradhilton opened on Sep 19, 2026

• Support combined expert and data parallelism in TrainerRank

  OpenPipe/ART#917 · Filter by author bradhilton opened on Sep 16, 2026

• Support pipeline parallelism in TrainerRank without exposing stage scheduling to callers

  OpenPipe/ART#916 · Filter by author bradhilton opened on Sep 16, 2026

• Investigate output and gradient variability between matched same-source Qwen TrainerRank actors

  OpenPipe/ART#902 · Filter by author bradhilton opened on Sep 15, 2026
  2

• Investigate large hidden-state differences when independent requests are co-packed (Qwen3.6, no-grad)

  OpenPipe/ART#901 · Filter by author bradhilton opened on Sep 15, 2026
  5

• Retained sampled suffix can carry logprobs under a different causal prefix

  OpenPipe/ART#883 · Filter by author bradhilton opened on Sep 10, 2026
  2

• TrainerRank reusable-cache admission can leave insufficient physical CUDA memory for backward library allocations

  OpenPipe/ART#870 · Filter by author bradhilton opened on Sep 9, 2026
  9

• Determine whether a large dynamics RL batch refusal is necessary or avoidable by execution layout

  OpenPipe/ART#869 · Filter by author bradhilton opened on Sep 9, 2026
  3

• Qwen3.5-35B-A3B at EP1 with CP>1 segfaults in Transformer Engine's grouped GEMM on real-data routing (likely empty local experts)

  OpenPipe/ART#851 · Filter by author bradhilton opened on Sep 4, 2026

• TrainerRank MoE memory admission: backward-recompute OOM and profile calibration failures

  OpenPipe/ART#848 · Filter by author bradhilton opened on Sep 4, 2026
  28

• Dedicated LocalBackend vLLM server can become unreachable during LoRA adapter reload

  OpenPipe/ART#678 · Filter by author arcticfly opened on May 12, 2026
  1

• Regression: forked GRPO run can produce malformed checkpoint outputs on newer ART ref

  bug
  OpenPipe/ART#672 · Filter by author arcticfly opened on May 8, 2026

• LoRA adapters are not properly syncing when using LocalBackend - stale adapters are being used in rollouts

  OpenPipe/ART#661 · Filter by author gmanlan opened on Apr 24, 2026

• LocalBackend fork_checkpoint doesn't update vLLM's initial LoRA

  OpenPipe/ART#651 · Filter by author arcticfly opened on Apr 13, 2026

• Add from_entity parameter to _experimental_fork_checkpoint

  OpenPipe/ART#649 · Filter by author arcticfly opened on Apr 10, 2026

• auto_rl example uses

  OpenPipe/ART#641 · Filter by author joshuajnoble opened on Apr 3, 2026

• ServerlessBackend creates a new W&B run per training step, causing "Run finished" alerts to fire on every step

  OpenPipe/ART#638 · Filter by author go5paopao opened on Apr 2, 2026

• [Security] Hardcoded API Key in vLLM Server Configuration Allows Authentication Bypass

  OpenPipe/ART#628 · Filter by author YLChen-007 opened on Mar 23, 2026

• Running out of memory with version v0.5.17

  OpenPipe/ART#617 · Filter by author Aleksis99 opened on Mar 16, 2026