unslothai/unsloth v0.1.2-beta

First Release post Unsloth Studio!

on GitHub

Hey guys, this is our first release since we launched Unsloth Studio last week. From now on you can directly access all our updates through our changelog here: https://unsloth.ai/docs/new/changelog

You can now update Unsloth Studio! Just use: unsloth studio update. Please update to use all the newest fixes and features.

• Tool calling improved. Better llama.cpp parsing, no raw tool markup in chat, faster inference, a new Tool Outputs panel, timers.
• Windows CPU or GPU now works seamlessly. Please reinstall!
• App shortcuts. Once installed, you can now launch in Windows, MacOS and Linux via a shortcut icon in the Start / Launch and Desktop.
• Pre-compiled llama.cpp binaries and mamba_ssm for finetuning - 6x faster installs! Also <300MB in size for binaries.
• 50% reduced installation sizes (-7GB or more savings), 2x faster installs and faster resolving. 50% smaller pypi sizes.
• Colab with free T4 GPUs with Unsloth Studio now fixed! Try it here. Due to pre-compiled binaries, it's also 20x faster!
• You can now properly use old GGUFs from Hugging Face or LM Studio
• MacOS and CPU now have Data Recipes enabled with multi-file uploading.
• AMD support preliminary for Linux only machines - auto detects.
• Settings sidebar redesign. Settings are now grouped into Model, Sampling, Tools, and Preferences
• Context length now adjustable. Keep in mind this is not needed as llama.cpp smartly uses the exact context you need via --fit on
• Persistent system prompts and presets. Custom system prompts and chat presets now persist across reloads and page changes.
• Multi-file upload. Data recipes now support multiple drag-and-drop uploads for PDF, DOCX, TXT, and MD, with backend extraction, saved uploads, and improved previews.
• Better chat observability. Studio now shows llama-server timings and usage, a context-window usage bar, and richer source hover cards.
• Better UX overall - clickable links, better LaTeX parsing, tool / code / web tooltips for default cards and much more!
• LiteLLM - Unsloth Studio and Unsloth were NOT affected by the recent LiteLLM compromise. Nemo Data Designer used LiteLLM only up to 1.80, not the affected 1.82.7 or 1.82.8, and has since removed it entirely.
• We now have a new one line install command, just run: Copycurl -fsSL https://unsloth.ai/install.sh | sh

Fixes:

• Windows/setup improvements. Fixed silent Windows exits, Anaconda/conda-forge startup crashes, broken non-NVIDIA Windows installs, and missing early CUDA/stale-venv setup checks.
• System prompts fixed. They work again for non-GGUF text and vision inference.
• GGUF export expanded. Full fine-tunes, not just LoRA/PEFT, can now export to GGUF. Base model resolution is more reliable, and unsupported export options are disabled in the UI.
• Chat scroll/layout fixes. Fixed scroll-position issues during generation, thinking-panel layout shift, and viewport jumps when collapsing reasoning panels.
• Smarter port conflict detection. Studio now detects loopback conflicts, can identify the blocking process when possible, and gives clearer fallback-port messages.

Example of automatic parameter settings for context length etc:

What's Changed

• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #4542
• fix: store embedding_learning_rate on self in UnslothTrainingArguments by @GoldenGrapeGentleman in #4531
• studio: persist system prompt and preset settings across navigation by @Imagineer99 in #4538
• studio: stop scroll hijack during generation and fix thinking panel layout shift by @Imagineer99 in #4543
• Fix Studio port conflict detection for loopback addresses by @danielhanchen in #4532
• fix(studio): show Windows-specific reset-password command by @Shine1i in #4529
• fix(studio): restore scroll lock on reasoning panel collapse by @danielhanchen in #4545
• fix: always show chat tool icons by @Shine1i in #4525
• fix: system prompt ignored in unsloth inference by @Shine1i in #4528
• fix: handle prompt/completion datasets in slow-path BOS detection by @danielhanchen in #4548
• fix: give @0xKushwaha git history credit for completion_only_loss fix by @danielhanchen in #4552
• ⚠️Remove quarantined litellm for precaution -- Unsloth Studio NOT affected by @danielhanchen in #4553
• fix: pin unsloth>=2026.3.11 in install scripts by @danielhanchen in #4556
• Regroup chat settings sidebar into focused sections by @Shine1i in #4551
• Add GRPO resume vLLM cleanup guard by @MagellaX in #4411
• fix: prevent UnicodeEncodeError on Windows CP1252 consoles in studio setup by @Krishnachaitanyakc in #4563
• studio: windows desktop shortcut launcher by @Imagineer99 in #4558
• Remove duplicate frontend assets from wheel (~31 MB savings) by @danielhanchen in #4567
• feat(studio): training history persistence and past runs viewer by @Shine1i in #4501
• fix: remove auto wandb.finish() after train() to allow post-training evaluate() by @Krishnachaitanyakc in #4564
• feat: Implement Q-GaLore optimizer and custom embedding learning rate… by @OnePunchMonk in #4511
• Bump Data Designer to 0.5.4 (removes litellm dependency) by @danielhanchen in #4569
• feat(chat): cleaner tool UI, inline LaTeX, clickable links by @Shine1i in #4561
• [Studio] Try installing causal-conv1d from prebuilt wheels if avialable by @Datta0 in #4547
• Feature/add dependabot and codeql security checks by @pkloehn1 in #4479
• build(deps): bump the actions group with 2 updates by @dependabot[bot] in #4570
• build(deps): bump oxc-parser from 0.116.0 to 0.121.0 in /studio/backend/core/data_recipe/oxc-validator in the npm-oxc-validator group by @dependabot[bot] in #4571
• Remove advanced CodeQL workflow (conflicts with default setup) by @danielhanchen in #4584
• Add macOS and Linux desktop shortcuts to install.sh by @danielhanchen in #4568
• perf(studio): upgrade to Vite 8 + auto-install bun for faster frontend builds by @Etherll in #4522
• feat(tokenizer): add get_tokenizer_info() diagnostic helper by @cz-03 in #4436
• Add ROCm (AMD GPU) support to studio setup by @danielhanchen in #4585
• Consolidate dual venvs and separate install from update by @rolandtannous in #4530
• studio: stabilize reasoning panel scroll behavior and prevent composer overlap by @Imagineer99 in #4587
• Use prebuilt llama.cpp for unsloth studio setup by @mmathew23 in #4562
• fix(studio): add -ngl flag for GPU offloading in llama-server by @danielhanchen in #4588
• fix(studio): add pip nvidia CUDA libs to LD_LIBRARY_PATH for llama-server by @danielhanchen in #4590
• fix(studio): validate bun install and retry from official source on failure by @danielhanchen in #4589
• fix(studio): clear bun cache on failure and retry before falling back to npm by @danielhanchen in #4594
• Pin torch>=2.4,<2.11.0 in Studio installers by @danielhanchen in #4595
• fix(studio): source-build fallback prefers Unsloth's tested tag over upstream latest by @danielhanchen in #4593
• fix(studio): add bun cache validation to Windows setup.ps1 by @danielhanchen in #4596
• feat: multi-source model discovery (HF default, legacy cache, LM Studio) by @rolandtannous in #4591
• Add unsloth to User PATH on Windows after install by @danielhanchen in #4597
• Add PID file tracking and unsloth studio stop command by @danielhanchen in #4598
• feat(studio): editable context length with Apply/Reset for GGUF settings by @danielhanchen in #4592

New Contributors

• @MagellaX made their first contribution in #4411
• @Krishnachaitanyakc made their first contribution in #4563
• @OnePunchMonk made their first contribution in #4511
• @pkloehn1 made their first contribution in #4479
• @dependabot[bot] made their first contribution in #4570
• @cz-03 made their first contribution in #4436

Full Changelog: b8475...v0.1.2-beta