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
No-deps pair test: start PowerShell through the shared runner (#16533)
07b323f Daniel Han committed 17 days ago
Pipeline parallel: end a non-last stage without a graph break under torch.compile (#16167)
fa2a5e7 Daniel Han committed 17 days ago
No-torch install harness: model a feed that resolves the Unsloth / Zoo pair (#16532)
69f489d Daniel Han committed 17 days ago
Multi-LoRA cold compile test: read the source as utf-8 (#16531)
7f0620d Daniel Han committed 17 days ago
Scoped download tests: seed the Hub cache the offline path now pins (#16530)
0f4464e Daniel Han committed 17 days ago
FSDP2: sum fp16 replicated gradients over unsynced micro-batches in float32 (#16173)
ff3f063 Daniel Han committed 17 days ago
No-Triton import: hand the triton stub to unsloth importers without publishing it (#15576)
74bc04a Daniel Han committed 17 days ago
Save Desktop Gallery batches with one folder choice (#16521)
3978a1b Daniel Han committed 17 days ago
Parallel isolation guard: list the Hub download tests poll deadlines as benign timing (#16526)
de02164 Daniel Han committed 17 days ago
Worker stderr mirror: read once more after an empty read that raced the stop (#16525)
12c4ba5 Daniel Han committed 17 days ago
Studio settings: size the composer Beta badge and the settings search offsets from the UI scale (#16524)
e8df6d5 Daniel Han committed 17 days ago
Studio: offer only the reasoning levels a template allows and show its raise message (#16502)
139ba23 oobabooga committed 17 days ago
Studio: darken Liquid Glass 3D composers over bright wallpapers (#16477)
1e34af3 Wasim Yousef Said committed 17 days ago
Studio: keep generated chat titles in the conversation's language (#16413)
99d1bc6 Debojyoti Banerjee committed 17 days ago
PPO fused rollouts: keep TRL's scores when generate applies logits processors (#15574)
3647f33 Daniel Han committed 17 days ago
Keep fused-loss engagement counters out of traced code (#15575)
54657ed Daniel Han committed 17 days ago
Studio Hub download: keep refs/main on the last complete snapshot under failed updates and concurrent workers (#16422)
1a951a9 Daniel Han committed 17 days ago
SDPA training on SM100: keep the caller's backend policy when avoiding cuDNN (#16230)
65e4bc6 Daniel Han committed 17 days ago
Fix FSDP2 checkpoint planning across parameter swaps and accumulation
99a8bdf Daniel Han committed 17 days ago
Resolve compatible Unsloth and Zoo pairs for all no-torch installs (#16157)
95aae40 Daniel Han committed 17 days ago
Packed mRoPE attention: no graph break on warm packed rows (#16416)
9db840f Daniel Han committed 17 days ago
Multi-LoRA grouped experts: run the grouped support and parity probes only eagerly (#16415)
bf38ae0 Daniel Han committed 17 days ago
Studio diffusion: a numbered shard without its index is not a complete snapshot (#16418)
e02bc9e Daniel Han committed 17 days ago
Studio Compare: keep the negative prompt on cells the Detailer pass still uses (#16149)
71229cc Daniel Han committed 17 days ago
Desktop: recheck the install fingerprint before accepting a cached capability answer (#16134)
a12502e Daniel Han committed 17 days ago
FSDP2 QAT export: count the converted copy in the writer's host RAM check (#16166)
242f14b Daniel Han committed 17 days ago