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
Prompt queue: row button labels do not take the next row's clicks (#15383)
be64e0d Daniel Han committed 26 days ago
Studio video: refuse NaN / inf decodes for every family before a PIL / uint8 cast or clamp hides them (#15374)
ad79849 Daniel Han committed 26 days ago
Studio: price per-layer embedding and Engram lookup tables as host or disk memory, not VRAM (#14465)
2b1476f Daniel Han committed 26 days ago
Merge branch 'main' of https://github.com/unslothai/unsloth
0be4c5d Daniel Han committed 26 days ago
TP Trainer parity tests: run AdamW with eps 1e-4 so reduction-order noise is not amplified (#15381)
7b14417 Daniel Han committed 26 days ago
Bump install.sh / install.ps1 pins to unsloth>=2026.10.9 (#15380)
06853a0 Daniel Han committed 26 days ago
Studio Images: no toast when the prompt upsampler is not available (#15365)
16672fb Daniel Han committed 26 days ago
Risky loader baseline: record the FSDP2 export's store handoff (#15379)
208d46b Daniel Han committed 26 days ago
Studio: prompt edit box, menu heading chips, hub Run, files panel, loaded row menu (#15375)
bb65bde Michael Han committed 26 days ago
Studio diffusion: fuse the int8 SwiGLU MLP on block-streamed Qwen-Image-2.1 denoisers (#15370)
aa9b152 Daniel Han committed 26 days ago
FSDP2: run GGUF, torchao, OpenVINO and GGML exports on rank 0 instead of refusing them (#15353)
d08984c Daniel Han committed 26 days ago
Studio: one request per chat model load, sent to both /validate and /load (#15139)
3df124b oobabooga committed 26 days ago
Studio video: keep the untiled decode gate visible under the stream encoder, skip the fp32 VAE rerun on non-finite latents, refuse NaN clips clearly (#15339)
975e864 Daniel Han committed 26 days ago
Studio: one owner for every request to llama-server, so a crashed server recovers on every API route (#15138)
c74438a oobabooga committed 26 days ago
Studio: keep a row's control by its title when a description sits under it (#15349)
f7a9616 Michael Han committed 26 days ago
Studio: reserve the DSpark/DFlash sidecar in --fit-target and recover from a Metal out-of-memory (#14431)
215039a Daniel Han committed 26 days ago
Block compile: keep each compiled block whole under DDP (DeepSeek-OCR torchrun) (#15359)
839b06f Daniel Han committed 26 days ago
Studio diffusion: cast-cache writes read block-streamed tensors from their host copy (#15357)
a2db307 Daniel Han committed 26 days ago
Studio diffusion: the prompt encode streams only the free-memory shortfall (#15348)
2b48056 Daniel Han committed 26 days ago
Studio: safer saved run settings, Always load, and API tab tidy-up (#15338)
5628964 Michael Han committed 26 days ago
FP8 fp16 dequant: keep autograd for a trainable weight or scale (#15354)
d9b50da Daniel Han committed 26 days ago
CI: rank apt mirrors by speed so a slow Azure mirror cannot time out apt steps (#15350)
1cf32bf Daniel Han committed 26 days ago
Studio: stop the macOS sandbox probe from leaking processes that spin forever (#15345)
6285c11 Michael Han committed 26 days ago
Do not serve inference_mode packed metadata to a training forward (#15347)
fa46d96 Daniel Han committed 26 days ago
Studio: exclude diffusion models from Chat's On Device list (#15336)
8cc9e01 Daniel Han committed 26 days ago