slime · Issues· 498 open

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

[Bug] Qwen-VL patch bypasses SGLang token-ID preservation and breaks multi-turn TITO

[Question] Would an adaptive in-reward KL controller fit Slime's PPO/RLHF scope?

[Bug] Resume advances Megatron scheduler with rollout_id instead of train step

Verify evals on Papers with Code

[Bug] Broadcasting full rollout_routed_experts data to every Megatron training rank will cause host OOM

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

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

[Bug] shipped dynamic-sampling filters crash on fan-out groups (list[list[Sample]]) from multi-turn agent rollouts

[BUG] 稳定训一半OOM，已开cp

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

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

[Question] realign 使用本轮 response 长度判断

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

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