All issues

Search results

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

enhancementNew feature or requestNew feature or request

Status: Open.

#926 In openpipe/art;

· bradhilton opened on Sep 19, 2026

Support combined expert and data parallelism in TrainerRank

Status: Open.

#917 In openpipe/art;

· bradhilton opened on Sep 16, 2026

Support pipeline parallelism in TrainerRank without exposing stage scheduling to callers

Status: Open.

#916 In openpipe/art;

· bradhilton opened on Sep 16, 2026

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

Status: Open.

#902 In openpipe/art;

· bradhilton opened on Sep 15, 2026

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

Status: Open.

#901 In openpipe/art;

· bradhilton opened on Sep 15, 2026

Retained sampled suffix can carry logprobs under a different causal prefix

Status: Open.

#883 In openpipe/art;

· bradhilton opened on Sep 10, 2026

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

Status: Open.

#870 In openpipe/art;

· bradhilton opened on Sep 9, 2026

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

Status: Open.

#869 In openpipe/art;

· bradhilton opened on Sep 9, 2026

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

Status: Open.

#851 In openpipe/art;

· bradhilton opened on Sep 4, 2026

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

Status: Open.

#848 In openpipe/art;

· bradhilton opened on Sep 4, 2026

Dedicated LocalBackend vLLM server can become unreachable during LoRA adapter reload

Status: Open.

#678 In openpipe/art;

· arcticfly opened on May 12, 2026

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

bugSomething isn't workingSomething isn't working

Status: Open.

#672 In openpipe/art;

· arcticfly opened on May 8, 2026