-
-
Notifications
You must be signed in to change notification settings - Fork 3.6k
Pull requests: lyogavin/airllm
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Carry rope_theta from the config in the MLX Llama path (fixes #368)
#369
opened Sep 17, 2026 by
Hotragn
Loading…
Pick the best available device instead of assuming CUDA (fixes #365, #355, #153)
#366
opened Sep 16, 2026 by
Hotragn
Loading…
fix: accept PyTorch tokenizer inputs in MLX generate
#363
opened Sep 9, 2026 by
Vaishnavi220506
Loading…
Fix delete_original crash for regular checkpoint files
#362
opened Sep 9, 2026 by
Nikhi00718
Loading…
feat: support prequantized BitsAndBytes streaming with Transformers 5
#347
opened Aug 23, 2026 by
DevElCuy
Loading…
Don't use per-expert streaming when shards are compressed
#344
opened Aug 21, 2026 by
Hotragn
Loading…
fix: robust layer-count parsing for nested checkpoint prefixes (complements #336)
#339
opened Aug 9, 2026 by
HEETMEHTA18
Loading…
fix: handle missing files gracefully in remove_real_and_linked_file (#297)
#337
opened Aug 7, 2026 by
bhumitschaudhry
Loading…
fix: critical bugs — unsafe torch.load (RCE), UnboundLocalError crash, thread leak, broken tests
#333
opened Aug 5, 2026 by
Sunil56224972
Loading…
docs: list Kimi K3 in Supported Models (#316)
#327
opened Aug 4, 2026 by
evanclan
Loading…
3 of 4 tasks
feat: AIRLLM_FULL_LOAD bypass — skip layer streaming when model fits in VRAM
#324
opened Aug 3, 2026 by
he1074937561-a11y
Loading…
Use the model's real generation config for custom architectures
#313
opened Jul 24, 2026 by
Hotragn
Loading…
Fix 4-bit disk-space check demanding ~3.5x the model size
#309
opened Jul 23, 2026 by
Hotragn
Loading…
Default to trust_remote_code=False, fall back only when required (addresses #293)
#306
opened Jul 20, 2026 by
Hotragn
Loading…
Prefer safetensors over pickle .bin when splitting model layers (fixes #296)
#305
opened Jul 20, 2026 by
Hotragn
Loading…
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.