Repository Issue Activity (beta)

#2370 [Bug] save_model fires unconditionally at the final step regardless of --save-interval; final optimizer-state write can kill the job

#1136 [BUG] 稳定训一半OOM，已开cp

#1487 Training hangs indefinitely after rollout phase completes on 8 B200 GPUs with CP=4, TP=2

#200 UX/Unintended bug: sample.reward is None on aborted sample

#2338 [Question] realign 使用本轮 response 长度判断

#397 if we use GRPO and args.kl_coef is non-zero, is the KL computation incorrect?

#1462 cannot import name 'GPU_MEMORY_TYPE_CUDA_GRAPH' from 'sglang.srt.constants'

#543 grade_answer_verl fails when rm_type="boxed_math" strips the \boxed{} wrapper upstream

#1369 【BUG】wrong implementation of zero-centered mean when normalizing advantages

#2214 Your project is on StackMap — a curated map of the AI stack

#2339 [Discussion] Support background evaluation while fully-async training rollout is running

#2344 [Bug] Fully-async multi-agent rollout crashes on nested groups

#2176 [Bug] 过采样关停的时候judge不会被杀掉

#2336 [Bug] Qwen3.5-VL grounding RL silently wrong: bundled SGLang image drops H/W rows of [3,T] M-RoPE (upstream sgl-project/sglang#35345, fix #35744)

#2324 如何才能快速训练GLM 5.2

#2332 [Bug] W&B initialization fails with wandb 0.28.2 because wandb.util.generate_id was removed

#2329 [Question] Multiple critic updates per actor update (SAO-style faster value update)?

#2285 [Question/Performance] Full message history is re-tokenized on every coding-agent turn

#2314 [Question] 什么时候支持ling3.0呢？

#1111 Add cache to process inputs

#2253 [Bug] Training/log-prob forward materializes full-vocab fp32 logits over the entire packed sequence → OOM for large-vocab, long multi-turn RL

#2188 [Bug] Colocate weight update fails with torch_memory_saver/offload because PyTorch CUDA IPC _share_cuda_ raises cudaErrorInvalidValue

#2201 GLM-5.2 SparseMLA TileLang backward returns NaN gradients from finite inputs

#2209 Delta weight sync (NCCL) produces NaN weights on Qwen3.5-122B MoE → "probability tensor contains inf/nan" crash

#2299 [Bug] Agent adapters rewrite malformed tool-call arguments before the client sees them