-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Pull requests: THUDM/slime
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[Fix] Support origin HF weight backfilling in parallel converter
#2270
opened Aug 13, 2026 by
albaNnaksqr
Contributor
Loading…
Fix O(#segments) per-comm memory probe that slows weight sync & training under expandable_segments
#2269
opened Aug 13, 2026 by
yszhli
Loading…
Fix model convert when use latest megatron
#2267
opened Aug 12, 2026 by
alexqdh
Contributor
Loading…
fix: load critic from policy checkpoints without value head
#2259
opened Aug 9, 2026 by
Dodojordi
Loading…
fix(rollout): guard zero-length response against x[-0:] returning the whole sequence
#2255
opened Aug 6, 2026 by
hobostay
Contributor
Loading…
fix: pool speculative-decoding metrics instead of averaging per-sample ratios
#2240
opened Jul 26, 2026 by
keepkeen
Contributor
Loading…
[Multimodal][Model] Make Qwen3.5-VL work with packed sequences
#2233
opened Jul 25, 2026 by
TobyYang7
Loading…
fix: avoid in-place mutation of param in MTP weight conversion (related to #2131)
#2227
opened Jul 21, 2026 by
botbikamordehai2-sketch
Loading…
Fix ref weights being broadcast on the first async update after resume
#2224
opened Jul 20, 2026 by
ishzgu
Loading…
[docker] backport output-gate slicing when query groups < TP
#2221
opened Jul 20, 2026 by
LLMShark
Loading…
fix(rollout): make dynamic refills granular
#2218
opened Jul 18, 2026 by
EazyReal
Contributor
Loading…
Support NemotronH hybrid Mamba+Attention+MoE training (e.g. Nemotron Nano-30B-A3B)
#2211
opened Jul 16, 2026 by
HelloWorldLTY
Contributor
Loading…
[Rollout] Add opt-in group-scoped session affinity
#2206
opened Jul 14, 2026 by
chengcuiping
Loading…
fix: normalize rewards by explicit sample groups
#2204
opened Jul 14, 2026 by
morluto
Contributor
Loading…
fix: validate DAPO integer labels without float coercion
#2203
opened Jul 14, 2026 by
morluto
Contributor
Loading…
Add critic load fallback for resuming training
#2200
opened Jul 13, 2026 by
coding-famer
Contributor
Loading…
fix: guard zero rollout temperature logprob scaling
#2197
opened Jul 12, 2026 by
EazyReal
Contributor
Loading…
fix(rollout): honor per-call args in generate_rollout_async (GenerateState caches first args)
#2196
opened Jul 10, 2026 by
sdc17
Loading…
fix(update_weight): preserve grouped MoE expert axis during GLU rechunk
#2193
opened Jul 10, 2026 by
LLMShark
Loading…
fix(update_weight): restore FlashInfer MoE layout after BF16 hot updates
#2192
opened Jul 10, 2026 by
LLMShark
Loading…
Previous Next
ProTip!
Follow long discussions with comments:>50.