Issues

• gpqa: when no option letter can be extracted, a single-letter gold label is substring-matched against the whole reply, so a reply that never answers can score 1.0

  THUDM/slime#2409 · Filter by author shaurya416 opened on Sep 24, 2026

• [Bug] Fully-async callback admits nested ABORTED agent results to the completed queue

  THUDM/slime#2402 · Filter by author heyxingxing opened on Sep 22, 2026

• [Bug] Per-sample train_metadata is lost during conversion and DP transport

  THUDM/slime#2400 · Filter by author heyxingxing opened on Sep 22, 2026

• [Bug] Fully-async worker failures never reach the training consumer

  THUDM/slime#2399 · Filter by author heyxingxing opened on Sep 22, 2026

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

  bug
  THUDM/slime#2388 · Filter by author dirtyDan0 opened on Sep 16, 2026

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

  question
  THUDM/slime#2387 · Filter by author hkxxxxx opened on Sep 15, 2026

• Verify evals on Papers with Code

  THUDM/slime#2386 · Filter by author NielsRogge opened on Sep 14, 2026

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

  bug
  THUDM/slime#2384 · Filter by author JayPritchet opened on Sep 14, 2026

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

  THUDM/slime#2374 · Filter by author kugua2025aigc-code opened on Sep 9, 2026

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

  THUDM/slime#2370 · Filter by author kugua2025aigc-code opened on Sep 8, 2026
  1

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

  bug
  THUDM/slime#2344 · Filter by author looput opened on Sep 2, 2026

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

  question
  THUDM/slime#2339 · Filter by author zhuyijie88 opened on Aug 31, 2026
  2

• [Question] realign 使用本轮 response 长度判断

  question
  THUDM/slime#2338 · Filter by author stivory0 opened on Aug 31, 2026
  1

• [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)

  THUDM/slime#2336 · Filter by author yszhli opened on Aug 28, 2026

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

  bug
  THUDM/slime#2332 · Filter by author bcol23 opened on Aug 26, 2026
  1

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

  question
  THUDM/slime#2329 · Filter by author ycoliver opened on Aug 26, 2026

• 如何才能快速训练GLM 5.2

  question
  THUDM/slime#2324 · Filter by author zyh190507 opened on Aug 24, 2026
  1

• [Question] 什么时候支持ling3.0呢？

  question
  THUDM/slime#2314 · Filter by author jmgytc opened on Aug 24, 2026

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

  THUDM/slime#2299 · Filter by author EazyReal opened on Aug 21, 2026

• [Agent RL] Qwen3.5 history re-templating causes false trajectory forks

  THUDM/slime#2288 · Filter by author zy20031230 opened on Aug 19, 2026

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

  question
  THUDM/slime#2285 · Filter by author zhuyijie88 opened on Aug 19, 2026
  3

• [Bug] Qwen3.5 MoE online weight update sends ETP-sharded TEGroupedLinear weights without all-gather

  bug
  THUDM/slime#2280 · Filter by author Shiyang980713 opened on Aug 17, 2026

• Per-comm memory probe (available_memory/memory_stats) is O(#allocator segments) and slows weight sync & training under expandable_segments

  THUDM/slime#2268 · Filter by author yszhli opened on Aug 13, 2026
  1

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

  bug
  THUDM/slime#2253 · Filter by author yc32768 opened on Aug 5, 2026
  1

• [Question] Proposal: Add Train–Inference Consistency Data Collection to slime

  question
  THUDM/slime#2244 · Filter by author zyang6 opened on Jul 27, 2026