[staging CI] unslothai/unsloth#6747 - #26

Closed

oobabooga wants to merge 34 commits into

mainfrom

Commits on Jun 29, 2026

Studio: add DFlash speculative decoding support
oobabooga
committed
2da0f28
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
579444b

Commits on Jul 3, 2026

Merge branch 'main' into dflash-spec-decoding
oobabooga
committed
fe1a5d0

Commits on Jul 8, 2026

Speed up Studio startup path (#6899)

wasimysaid
and
danielhanchen
authored
49d1fb3
Polish assistant message actions menu (#6962)
shimmyshimmer
authored
e7e6a0f
Move New badge to System settings tab (#6963)

shimmyshimmer
and
pre-commit-ci[bot]
authored
7f9964f
Fix opencode Unsloth provider selection (#6906)

authored
393d7e9
Fix Hermes install hint on Windows (#6903)

authored
baacbd0
Studio: heal DiffusionGemma tool calls into structured tool_calls (#6851)

authored
a113f89
Fix case-variant model matching and GGUF cache reuse in unsloth start (#6900)

authored
df6b5a5
Studio: show Hugging Face address on hover for Hub and online model rows (#6382) (#6928)

danielhanchen
and
wasimysaid
authored
f1a2621
Studio: fix currency and indentation edge cases in LaTeX rendering (#6957)
danielhanchen
authored
de60a3a
Add MLX backend support for CLI unsloth train (#6709)

authored
38dacb8
feat(cli): support MLX distributed inference (#6845)

authored
2a6abe2
feat(mlx): route trainer callbacks (#6929)
Lyxot
authored
934f879
(GRPO) Fix PEFT replacement for TRL >= 1.7.0, add missing compute_aux_loss for TRL >= 1.7.0 (#6904)

authored
07c8bbb
version-compat CI: fake CPU training runs for SFT/GRPO/DPO (#6965)

danielhanchen
and
pre-commit-ci[bot]
authored
0e1ed88
Fix OpenClaw start default to local TUI (#6937)

authored
6ef0936
feat: detect installed coding agent CLIs in Studio settings (#6909)

authored
e86b787
Studio: don't pin transformers before the training worker activates the 5.x sidecar (#6968)

authored
41dd95e
Studio: source CPU llama.cpp prebuilts from unslothai/llama.cpp (#6311)

authored
fcb1152
fix(studio/hub): apply repo_id length limit per segment, not whole string (#6946) (#6953)

authored
d0c8d55
MoE LoRA: auto-target per-expert Linear experts (gpt-oss 4bit) instead of leaving them frozen (#6936)

danielhanchen
and
pre-commit-ci[bot]
authored
62a6eb2
Studio: fix flash-attn and torchao install on Blackwell (sm_100+) GPUs (Closes #6961) (#6970)

authored
03cbe21
Versioning
danielhanchen
committed
38ea267
Fix the tensor-parallel probe test's stale spec-drop fallback anchor
oobabooga
committed
bb5b202
Add has_blackwell_gpu to the mlx worker test's wheel_utils stub (#6980)
danielhanchen
authored
3d41e58
Studio: allow CPU-only DiffusionGemma by granting the diffusion runner the CPU device (#6979)
danielhanchen
authored
116ce48
Bump install.sh / install.ps1 pins to unsloth>=2026.7.2 and unsloth-zoo>=2026.7.2 (#6981)
danielhanchen
authored
1a274c4
Studio: render thinking blocks for safetensors inference with prefilled <think> templates (#6816)

authored
5c2e536
Remove API menu new badge (#6983)
shimmyshimmer
authored
7a9fb44
Merge branch 'main' into dflash-spec-decoding
oobabooga
committed
49be96a
Retry extra-args DFlash without spec, skip HF DFlash reload dedup
oobabooga
committed
132b61c
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
d2b20b8