-
Notifications
You must be signed in to change notification settings - Fork 1.5k
Pull requests: Lightning-AI/litgpt
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(thunder): remove duplicate _configure_launcher that dropped the external-launch guard
#2324
opened Sep 8, 2026 by
Anai-Guo
Loading…
fix(merge_lora): use CPU-safe precision to avoid false AMP warning
#2322
opened Sep 7, 2026 by
ShamikOfficial
Loading…
1 task done
fix: keep pretrain validate lockstep on empty ranks (#2319)
#2321
opened Sep 4, 2026 by
ShamikOfficial
Loading…
5 of 6 tasks
Handle alternate CLI entry points when saving checkpoints
#2318
opened Aug 31, 2026 by
Marchematics
Loading…
feat: add MultiturnJSON datamodule + full multi-turn SFT support
#2317
opened Aug 31, 2026 by
ysjprojects
Collaborator
•
3/3
Loading…
feat: multiturn support for ChatML
#2314
opened Aug 30, 2026 by
ysjprojects
Collaborator
•
1/3
Loading…
feat: make cross_entropy_chunk_size tunable, with an auto memory budget (#2190) [3/3]
#2312
opened Aug 28, 2026 by
OnePunchMonk
Contributor
Loading…
1 task done
profile: measure the chunked_cross_entropy memory spike and find its source (#2190) [2/3]
#2311
opened Aug 28, 2026 by
OnePunchMonk
Contributor
Loading…
1 task done
fix: raise a clear error on stale KV cache, avoid double allocation in chat (#2190) [1/3]
#2310
opened Aug 28, 2026 by
OnePunchMonk
Contributor
Loading…
1 task done
fix(generate): reject top_k below 1 instead of masking every logit
#2309
opened Aug 27, 2026 by
Anai-Guo
Loading…
Fix first token handling in speculative decoding
#2308
opened Aug 25, 2026 by
Excelius-Wang
Loading…
Add in-memory data support to the JSON DataModule
#2304
opened Aug 22, 2026 by
Excelius-Wang
Loading…
feat: optional torchembed fused Triton RoPE kernel (~3-4x training speedup)
#2280
opened Jul 6, 2026 by
py-ai-dev
Loading…
Handle tokenizer configs without tokenizer_class
#2278
opened Jun 25, 2026 by
CodingFeng101
Loading…
ci: add Lightning registry fallback for gated tokenizer tests
#2271
opened Jun 24, 2026 by
bhimrazy
Collaborator
Loading…
Fix malformed <|im_start> token in Phi4Reasoning prompt style
#2267
opened Jun 16, 2026 by
JSap0914
Loading…
fix(finetune): all_reduce val loss across devices for initial and final validation
#2266
opened Jun 13, 2026 by
discobot
Loading…
fix: make convert_from_litgpt output loadable with AutoModel.from_pretrained
#2265
opened Jun 13, 2026 by
discobot
Loading…
Previous Next
ProTip!
Add no:assignee to see everything that’s not assigned.