Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Commits on main
Studio: time a continued reply as the whole answer, not just its last pass (#16451)
92aaa26 oobabooga committed 18 days ago
Studio: keep the live generation line from moving the branch picker, and update it during short replies (#16449)
0102e22 oobabooga committed 18 days ago
Studio: estimate the prompt's time left from how the cost per token grows, not the rate so far (#16450)
ac72298 oobabooga committed 18 days ago
Studio: list every Gemini chat model (follow nextPageToken, allow new families) (#16433)
6e3890b Debojyoti Banerjee committed 18 days ago
Preserve external chat import recovery across checkpoint failures (#16458)
f9cecc9 Daniel Han committed 18 days ago
Backend CI: run the context parallel compose gloo suite in its own shard (#16447)
a848037 Daniel Han committed 18 days ago
Studio: add a Time format setting and drop the redundant desktop update note (#16448)
432344e Michael Han committed 18 days ago
PP env GA test: compare the trained adapter at 5e-5 (#16457)
d4dcba7 Daniel Han committed 18 days ago
Studio: turn navigator at the chat's left edge, building on #11668 (#16300)
d56738f Michael Han committed 18 days ago
Studio: read lookup tables lazily when needed to enable the MoE cache (#16444)
df96104 oobabooga committed 18 days ago
Repair the three CI checks #16407 left red on main (#16446)
ee6640a Daniel Han committed 18 days ago
Studio: replace stray Lucide icons and settle duplicate icons (#16445)
2f5ff22 Michael Han committed 19 days ago
UNSLOTH_PARALLEL: one flag and one root mesh for dp / tp / pp / cp, pipeline parallel from the environment, and a capability registry (#16407)
9341351 Daniel Han committed 19 days ago
Profile stats tests: seed at noon, so a run just before midnight keeps the reply on today (#16440)
725d823 Daniel Han committed 19 days ago
Multi-account Playwright: record and retry once when the post-logout /login load never fires (#16437)
3a4406b Daniel Han committed 19 days ago
Studio: shorten the Hugeicons trash and bookmark comments (#16439)
a7935da Michael Han committed 19 days ago
Studio tests: chat width driver picks the presets by position, not by their copy (#16431)
681111a Daniel Han committed 19 days ago
Studio tests: take the hub source baseline when the first test starts, not at collection (#16430)
4824fcc Daniel Han committed 19 days ago
Risky loader baseline: record the float8 training precision fallback (#16428)
674b243 Daniel Han committed 19 days ago
Studio tests: best of five for the two single-sample nested-header linearity checks (#16426)
5819e6c Daniel Han committed 19 days ago
Studio sandbox: remap the Windows spelling of /mnt/data-style paths (#16412)
66f1bed Daniel Han committed 19 days ago
Tests: gpt-oss dense-expert reload fixture drops replaced experts' quant state (#16246)
58cf8e9 Daniel Han committed 19 days ago
Pipeline parallel: carry the Qwen3-MoE router auxiliary loss across stages (#16406)
31dac50 Daniel Han committed 19 days ago
Block compile: install the router_logits capture hooks before the first block compiles (#16410)
5a1a522 Daniel Han committed 19 days ago
Studio: run multi-GPU compile tests on real CUDA, fail on eager fallback (#16224)
a106493 Daniel Han committed 19 days ago
Full finetuning notice: say float32 for forced-float32 models on GPUs without bfloat16 (#16405)
bb5a3a3 Daniel Han committed 19 days ago
Studio tests: a separate engine port block for every xdist worker past 40 (#16245)
3a4600d Daniel Han committed 19 days ago
Studio: use the Hugeicons trash and bookmark-01 icons (#16409)
d36c202 Michael Han committed 19 days ago
Studio tests: skip the CUDA compiled-linear checks where production stays eager (#16237)
5fcbc73 Daniel Han committed 19 days ago