-
-
Notifications
You must be signed in to change notification settings - Fork 6.9k
Pull requests: unslothai/unsloth
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Use the dynamic cache for Gemma multimodal generation
#10903
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Studio: keep the operator's package-manager policy in the installer
#10902
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Pin the --local unsloth-zoo git overlay to a resolved commit
#10901
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Give every pwsh call in the suite the private startup cache
#10900
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
installer: stop condemning a warm uv cache for a directory uv never writes
#10899
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Stop ten test assertions from measuring the runner instead of the code
#10897
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Fix three CI failures that were not the PRs they failed on
#10896
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
fix(studio): keep Studio's auth directory out of the tool loop
#10895
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Share one child interpreter, and one notebook prototype, where the work is identical
#10893
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Stop an older unsloth_zoo turning a fully masked microbatch into NaN
#10892
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Allow fast_inference for Qwen3.5
edge case#
new-feature#
non-studio#
#10891
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Studio: NVFP4 time budget, DiT VAE decode compile
important#
new-feature#
regresses#
#10889
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Studio: auto transformer precision walks int8 first on every GPU tier
regresses#
#10888
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Stop three studio meta-tests re-cutting and re-walking the same sources
#10886
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Honor user tensor_split when auto tensor-parallel planner returns an even split
important#
#10884
opened Sep 14, 2026 by
chakshu-dhannawat
Contributor
Loading…
Studio: official image picks default to the hosted FP8 or INT8 checkpoint
affects UI#
important#
new-feature#
>3000#
#10883
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Split Repo tests (CPU) into three shards, and take the shell suite off its critical path
#10882
opened Sep 14, 2026 by
danielhanchen
Member
Loading…
Invalidate packed attention caches when lengths change
edge case#
non-studio#
regresses#
#10879
opened Sep 13, 2026 by
vineethsaivs
Contributor
Loading…
Restore full-rank gradients after Q-GaLore updates
edge case#
non-studio#
#10878
opened Sep 13, 2026 by
vineethsaivs
Contributor
Loading…
Studio: support multiple resident GGUF models
new-feature#
#10876
opened Sep 13, 2026 by
MiddlePathLabs
Loading…
Pass Q-GaLore optimizer options to bitsandbytes by name
important#
non-studio#
#10874
opened Sep 13, 2026 by
vineethsaivs
Contributor
Loading…
Add approved image attachment inputs for MCP tools
affects UI#
new-feature#
regresses#
>3000#
#10871
opened Sep 13, 2026 by
Etherll
Collaborator
Loading…
fix(studio): include loaded llama extra args in active model baseline
affects UI#
edge case#
#10870
opened Sep 13, 2026 by
Souravrajvi0
Contributor
Loading…
1 task done
fix(studio): key Run settings editors by model and quant only
affects UI#
regresses#
#10869
opened Sep 13, 2026 by
Souravrajvi0
Contributor
Loading…
2 tasks done
Studio: add Greek to dictation and read-aloud language options
affects UI#
new-feature#
>3000#
#10865
opened Sep 13, 2026 by
christosdaggas
Loading…
Previous Next
ProTip!
Updated in the last three days: updated:>2026-09-10.