v0.1.814-beta

You can now run Qwen-Image-2.1 locally with Unsloth! This release also includes custom Agent Skills, and easier chat/project management. It also brings 2x faster reasoning blocks (60 FPS vs 30 FPS), more reliable training, and improved Linux installs and updates.  [Qwen Image 2.1 Guide](https://unsloth.ai/docs/models/qwen-image-2.1)

## Highlights
* Qwen-Image-2.1 fixes - diffusers update + GGUF fix
* Qwen-Image-2.1 works for **image editing** and image gen!
* Add custom Agent Skills to guide models through specific tasks.
* Chats are now draggable/customizable in sidebar.
* 2x faster long reasoning blocks (60 FPS vs 30 FPS before). Thinking UI/UX reworked.
* In-app Debian updates and an Ubuntu 24.04+ installer for ARM64.
* Projects are easier to edit and organize.

## September 22nd Update
* Added **image editing** to Qwen-Image-2.1
* Fixed diffusers updating issues and fixed GGUFs not loading for Qwen-Image
* Changed default for GGUFs to Q4_K_M from F16 / BF16
* Fixed diffusion black artifacts for A100 and consumer GPUs

## Agent Skills
* Add custom skills to guide models through specific tasks.
* Reuse skills from your existing Claude Code and `.agents` folders.
* Manage your skills and use `@` to select one in chat.

## Chats + projects
* Drag chats to reorder them, pin them or move them into a project.
* Better model loading behavior, chat properly remembers settings
* Thinking UI/UX reworked for cleaner modern experience.
* Edit project names, instructions and folders directly from the Projects page.
* Long code replies stream more smoothly while keeping syntax highlighting.

## Models + storage
* Vision models can now see images returned by MCP tools.
* Other accounts can connect to local model servers when the owner allows it.
* Chat can reuse downloaded models instead of fetching a second copy.
* CLI chats preserve loaded models’ precision and use recommended settings.
* See which caches take up disk space and clear them from Settings.

## Training + exports
* More reliable training resumes and dataset preparation.
* Model uploads to the Hub leave out files from earlier exports.
* Hugging Face datasets now work in recipes, past training runs can be resumed, resumed runs show the correct remaining time, dataset columns are mapped correctly, and selected splits/subsets are respected.

## Desktop + hardware
* Linux/Desktop distribution got substantially better - there are now in-app updates for Debian installs and a native ARM64 Linux .deb.
* New AMD ROCm Docker image with Unsloth Studio and JupyterLab.
* Larger text now has more room, and contrast settings make buttons and menus easier to see.
* Filter logs, wrap long lines and jump to the latest entries without scrolling through everything.

## What's Changed
* Bump install.sh / install.ps1 pin to unsloth>=2026.9.7 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11279
* Carry over five temp-scanner fixes that landed after #11217 merged by @danielhanchen in https://github.com/unslothai/unsloth/pull/11280
* fix the Hub README's tag and GPU support claims by @NilayYadav in https://github.com/unslothai/unsloth/pull/11131
* Give the page-fault stall test a window the scheduler cannot close by @danielhanchen in https://github.com/unslothai/unsloth/pull/11284
* Stop three tests stranding a waiting request in a module-level dict by @danielhanchen in https://github.com/unslothai/unsloth/pull/11287
* Assert the companion scan's thread, not how fast the machine is by @danielhanchen in https://github.com/unslothai/unsloth/pull/11288
* studiobench: count the attachments the composer actually renders by @danielhanchen in https://github.com/unslothai/unsloth/pull/11302
* Route today's new PowerShell test calls through the shared runner by @danielhanchen in https://github.com/unslothai/unsloth/pull/11285
* Studio: run one repair at a time, however many times Retry is pressed by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11249
* Route the last direct pwsh spawns through the shared runner by @danielhanchen in https://github.com/unslothai/unsloth/pull/11303
* Studio: keep a LoRA push private when the Hub repo already exists by @NilayYadav in https://github.com/unslothai/unsloth/pull/11229
* Studio: preserve Claude Code prompt cache with custom output styles by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11121
* Disengage FBCache on the word of whoever owns it by @danielhanchen in https://github.com/unslothai/unsloth/pull/11304
* Studio: preserve partial state for canceled GGUF companions by @Etherll in https://github.com/unslothai/unsloth/pull/11018
* Studio: restore an exported chat backup as separate chats by @NilayYadav in https://github.com/unslothai/unsloth/pull/11231
* Studio: fetch a remote image URL instead of forwarding it to llama-server by @oobabooga in https://github.com/unslothai/unsloth/pull/11160
* Studio: tell the model which tool calls were skipped by the per turn limit by @NilayYadav in https://github.com/unslothai/unsloth/pull/11154
* fix(desktop): explain updates blocked by an existing Studio server by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11273
* Desktop: support SQF script attachments by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11274
* Studio: preserve local model inventory after cache filesystem errors by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11272
* Studio: train a gated upstream model through its public Unsloth copy by @oobabooga in https://github.com/unslothai/unsloth/pull/11214
* Stop one test's app settings answering another test's read by @danielhanchen in https://github.com/unslothai/unsloth/pull/11309
* Make the WSL compiler-intermediate control say why it failed by @danielhanchen in https://github.com/unslothai/unsloth/pull/11312
* Studio: let models see images returned by MCP tools by @NilayYadav in https://github.com/unslothai/unsloth/pull/10088
* Studio: trim the composer "+" menu top level by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11314
* Studio: optically centre the model selector label by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11316
* Put back the Docker quickstart paragraph the README lost by @danielhanchen in https://github.com/unslothai/unsloth/pull/11317
* Re-approve three unsloth-zoo exec sites, pinned, and say which reviews are which by @danielhanchen in https://github.com/unslothai/unsloth/pull/11318
* Serve the decoded pixels on a multi-image turn, not the base64 they arrived as by @danielhanchen in https://github.com/unslothai/unsloth/pull/11320
* Clear the idle-reload stash between tests by @danielhanchen in https://github.com/unslothai/unsloth/pull/11322
* Desktop clean machine: pick the x64 installer, and stop blaming the .deb for the mirror by @danielhanchen in https://github.com/unslothai/unsloth/pull/11323
* Acknowledge the Colab oracle drift, and stop the diff hiding entries by @danielhanchen in https://github.com/unslothai/unsloth/pull/11324
* Give the torchcodec placeholder a real __spec__ by @danielhanchen in https://github.com/unslothai/unsloth/pull/11325
* Strip distro .devN markers from the Colab pins by @danielhanchen in https://github.com/unslothai/unsloth/pull/11326
* Give the CUDA spoof the raw-stream handle unsloth reads at import by @danielhanchen in https://github.com/unslothai/unsloth/pull/11328
* Studio: report a backend that is provably gone without sleeping out the retry ladder by @danielhanchen in https://github.com/unslothai/unsloth/pull/11213
* Name an AMD device node this account cannot open, instead of reading it as no GPU by @danielhanchen in https://github.com/unslothai/unsloth/pull/10473
* Studio: keep a connected pin that could not be written by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11269
* Studio setup: reuse the uv a previous run installed instead of downloading it again by @danielhanchen in https://github.com/unslothai/unsloth/pull/10659
* Point the interrupt legs at the phase label the installer prints by @danielhanchen in https://github.com/unslothai/unsloth/pull/11342
* Studio: tidy run settings controls and settings page titles by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11331
* Cancel superseded pull request runs, and guard that they stay cancelled by @danielhanchen in https://github.com/unslothai/unsloth/pull/11345
* Format the superseded-runs guard the way the hook does by @danielhanchen in https://github.com/unslothai/unsloth/pull/11348
* Studio: tidy the run settings panel by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11346
* Studio: let recipes with a Hugging Face dataset run by @NilayYadav in https://github.com/unslothai/unsloth/pull/11300
* Studio: bring back the Resume button for past training runs by @NilayYadav in https://github.com/unslothai/unsloth/pull/11301
* Studio: map context, question and answer columns to the right roles by @NilayYadav in https://github.com/unslothai/unsloth/pull/11292
* Studio: show the right time left after resuming training by @NilayYadav in https://github.com/unslothai/unsloth/pull/11297
* Say which locale catalog failed instead of dying in an eval by @danielhanchen in https://github.com/unslothai/unsloth/pull/11344
* Studio: push only the exported files to the Hugging Face Hub by @NilayYadav in https://github.com/unslothai/unsloth/pull/11299
* Retry a browser driver once before failing the leg by @danielhanchen in https://github.com/unslothai/unsloth/pull/11355
* Take the Linux subdirectory-watch race out of the compiler watcher test by @danielhanchen in https://github.com/unslothai/unsloth/pull/11362
* Docker ROCm image: reach the GPU through WSL2's DXG bridge, not only /dev/kfd by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11212
* Studio: turn on Code when Full access is picked by @oobabooga in https://github.com/unslothai/unsloth/pull/11283
* Studio: chat with a cached upstream model instead of downloading its Unsloth copy by @oobabooga in https://github.com/unslothai/unsloth/pull/11307
* studio: add in-app updates for debian installs by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11338
* Let the uv-safe requirements test pass where 8.3 names exist by @danielhanchen in https://github.com/unslothai/unsloth/pull/11365
* studio: protect authenticated release redirects by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11339
* Studio: give a project one home in the sidebar, and fix the row menus around it by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11370
* Studio: reopen dismissed Deep Research plan dialogs by @AzizMuminov in https://github.com/unslothai/unsloth/pull/10802
* desktop: publish a native arm64 linux deb by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11356
* Studio: export and save only the reply you are looking at by @NilayYadav in https://github.com/unslothai/unsloth/pull/11298
* fix(studio): track Marked's def rule in the link definition probe by @HughScott2002 in https://github.com/unslothai/unsloth/pull/9645
* Studio: skip duplicate weight files when downloading a model from the Hub by @NilayYadav in https://github.com/unslothai/unsloth/pull/11227
* Docker ROCm image: ship Unsloth Studio, as the CUDA image already does by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11218
* Keep the regenerable caches inside the Studio root by @danielhanchen in https://github.com/unslothai/unsloth/pull/10597
* Studio: show Vulkan iGPU memory as shared availability by @oobabooga in https://github.com/unslothai/unsloth/pull/11305
* Studio: put the sidebar nav validator signature on the line the formatter wants by @danielhanchen in https://github.com/unslothai/unsloth/pull/11374
* Studio: drop rows the chat template can't format and tell the user by @NilayYadav in https://github.com/unslothai/unsloth/pull/11293
* Let the pinned formatter have the line it wants in settings.py by @danielhanchen in https://github.com/unslothai/unsloth/pull/11371
* Give the desktop release tests an explicit utf-8 encoding by @danielhanchen in https://github.com/unslothai/unsloth/pull/11372
* Correct why a superseded probe matrix is expensive: slots, not minutes by @danielhanchen in https://github.com/unslothai/unsloth/pull/11347
* Say when a failing pull_request job is testing a stale merge ref by @danielhanchen in https://github.com/unslothai/unsloth/pull/11386
* Read the routing invariance counters as a steady state, not one request by @danielhanchen in https://github.com/unslothai/unsloth/pull/11389
* Hold the diffusion load worker on an event instead of a sleep by @danielhanchen in https://github.com/unslothai/unsloth/pull/11392
* Studio: drag sidebar rows into place, and make a project row do more by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11373
* Studio: use only the chosen split and subset for Hugging Face recipe data by @NilayYadav in https://github.com/unslothai/unsloth/pull/11294
* Count a unified-memory APU's window as shared host memory, not dedicated VRAM by @danielhanchen in https://github.com/unslothai/unsloth/pull/11366
* Studio: use copy-01 for every copy button by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11395
* Studio: treat bash `coproc` as a command boundary in the terminal safety scanners by @danielhanchen in https://github.com/unslothai/unsloth/pull/11369
* Drain worker threads without racing their start by @danielhanchen in https://github.com/unslothai/unsloth/pull/11400
* Studio: read the local model cache locally, and keep host paths off the API by @danielhanchen in https://github.com/unslothai/unsloth/pull/11026
* fix(studio): run built-in text search as a tiered engine allowlist by @alkinun in https://github.com/unslothai/unsloth/pull/11242
* Give a mixed NVIDIA plus AMD host a way to ask for the ROCm torch stack by @danielhanchen in https://github.com/unslothai/unsloth/pull/10474
* Click the reasoning trigger by slot, not by its wording by @danielhanchen in https://github.com/unslothai/unsloth/pull/11406
* Studio: use one expand icon and one download icon everywhere by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11411
* Studio: one white and one grey for dark mode thread text by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11410
* Studio: round the right corners of scrollable tool result and thinking panes by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11409
* Studio: rename the edit composer's Update button to Send, and resend an unchanged prompt by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11407
* Studio: answer a recipe check the backend cannot read the seed for by @danielhanchen in https://github.com/unslothai/unsloth/pull/11363
* Repair the two frontend CI breaks main inherited from #11373 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11416
* Studio: bind the Data Recipe export link to the account that minted it by @danielhanchen in https://github.com/unslothai/unsloth/pull/11401
* Windows installer: stop an unresolvable 8.3 alias aborting the rollback by @umran666 in https://github.com/unslothai/unsloth/pull/11319
* Studio: make the rule closing a thinking trace readable by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11424
* Studio: one rounded outline on the guided tour card by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11421
* Studio: brighten tool rows on hover, like the Thinking trigger by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11422
* Studio: use the scroll icon for Agent Skills everywhere by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11419
* Studio: say what auto-compaction actually does by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11415
* Studio: put a working chat row's spinner back on the trailing column, and tighten the pin gap by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11408
* Studio: icon-only Copy and Download in the transcript, with tooltips by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11425
* Use one speech bubble for a chat everywhere by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11428
* Studio: call the product Unsloth in the app's own copy by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11427
* Scale the settings hint icon with the UI font size by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11429
* Studio: fall back to the Vulkan sd.cpp build when the ROCm one cannot run by @danielhanchen in https://github.com/unslothai/unsloth/pull/11068
* Read only the selector shapes the sheet test can prove, and refuse the rest by @danielhanchen in https://github.com/unslothai/unsloth/pull/11089
* Make the GRPO canary run through Unsloth's own batch sampler by @danielhanchen in https://github.com/unslothai/unsloth/pull/11126
* Studio: call it Skills, and lowercase files in Chat with files by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11426
* Stop the barrel guard counting package imports as a way back in by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11431
* Reject a new studio union offender even when the recorded count is unchanged by @danielhanchen in https://github.com/unslothai/unsloth/pull/9856
* Derive the chat UI wall-clock watchdog from the turn timeout by @danielhanchen in https://github.com/unslothai/unsloth/pull/9025
* Studio: scope the GGUF converter pin to the conversion by @danielhanchen in https://github.com/unslothai/unsloth/pull/11417
* Studio auth and bootstrap gates: throttle desktop-login, narrow the tunnel CORS origin, identify Colab's proxy by @danielhanchen in https://github.com/unslothai/unsloth/pull/11398
* fix(studio): widen the link reference probe to match the definition probe by @danielhanchen in https://github.com/unslothai/unsloth/pull/11375
* remove gguf shard controls introduced by #6107 by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11315
* Fail Source lint on a duplicate TypeScript import binding by @danielhanchen in https://github.com/unslothai/unsloth/pull/8548
* A failed presence check in the indicator driver should say what the page looked like by @danielhanchen in https://github.com/unslothai/unsloth/pull/10435
* Studio: the drag-and-drop switches go, the gesture has one behaviour by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11441
* Studio: drop the "When context fills" setting and always use the server's policy by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11444
* Studio: read the saved credential and its presence in one query by @danielhanchen in https://github.com/unslothai/unsloth/pull/11420
* Studio: require a published digest on upstream llama.cpp prebuilts too by @danielhanchen in https://github.com/unslothai/unsloth/pull/11402
* fix(sft): honor max_length below the model context limit by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/11266
* Studio: three-state Thinking and Tool calls visibility in Display by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11433
* Refuse a compiler cache path shlex cannot parse, not only one with a space by @danielhanchen in https://github.com/unslothai/unsloth/pull/11414
* [perf] Qwen3.5-9B LoRA SFT with unsloth-cli.py on a B200: 0.84 s to 0.77 s per step at 1 GPU, the model-independent half of #10744 by @TarzanZhao in https://github.com/unslothai/unsloth/pull/11238
* Offer FlexAttention when head_dim puts every flash kernel out of reach by @danielhanchen in https://github.com/unslothai/unsloth/pull/11102
* Studio: edit a project from the Projects page, and sort it from the Updated column by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11443
* Repair four checks that are red on main itself by @danielhanchen in https://github.com/unslothai/unsloth/pull/11456
* studio: preserve standalone gguf settings in compare mode by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11311
* Stop SFTTrainer downcasting configs that subclass SFTConfig by @danielhanchen in https://github.com/unslothai/unsloth/pull/11440
* Point the two compiler-cache fallback tests at a temporary root Windows accepts by @danielhanchen in https://github.com/unslothai/unsloth/pull/11455
* Docker ROCm Studio image: ship JupyterLab with the notebooks, as the CUDA image does by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11286
* Studio: prefer the discrete AMD GPU when an iGPU enumerates first on Linux by @danielhanchen in https://github.com/unslothai/unsloth/pull/11437
* Studio: let the owner allow managed accounts private provider base URLs by @danielhanchen in https://github.com/unslothai/unsloth/pull/11436
* Studio: keep an AMD dGPU and a Vulkan choice through install and update by @danielhanchen in https://github.com/unslothai/unsloth/pull/11442
* Apply the logit scale on the fused cross entropy path by @vineethsaivs in https://github.com/unslothai/unsloth/pull/11277
* Studio: stop suggesting bnb-4bit models MLX cannot load on Mac by @NilayYadav in https://github.com/unslothai/unsloth/pull/9722
* Read the sidebar and reasoning contracts as claims, not as literals by @danielhanchen in https://github.com/unslothai/unsloth/pull/11404
* Studio: isolate the sandbox traversal test from the real home directory by @danielhanchen in https://github.com/unslothai/unsloth/pull/11471
* Stop four POSIX-only test modules from failing to collect on Windows by @danielhanchen in https://github.com/unslothai/unsloth/pull/11457
* Studio: keep the account confinement tests off the interpreter walk by @danielhanchen in https://github.com/unslothai/unsloth/pull/11488
* Studio: pin the hardware verdict in the two export capability tests that depend on it by @danielhanchen in https://github.com/unslothai/unsloth/pull/11490
* studio: copy full prompts from the api monitor by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11340
* Studio: pin the worker RSS in the two xet ledger tests that stand in for a worker by @danielhanchen in https://github.com/unslothai/unsloth/pull/11492
* Give the GPU summary probe test the AMD helpers its extracted block calls by @danielhanchen in https://github.com/unslothai/unsloth/pull/11472
* Studio: read the 16-bit pixels the way the rest of that test file already does by @danielhanchen in https://github.com/unslothai/unsloth/pull/11493
* Studio: give back the continuation lease when the user stops the preflight by @danielhanchen in https://github.com/unslothai/unsloth/pull/9560
* Studio installers: add --no-rollback, and report a full disk as a full disk by @danielhanchen in https://github.com/unslothai/unsloth/pull/11364
* Studio: spell the owner's sidebar name Unsloth, not unsloth by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11275
* fix(studio): bundle hf-xet for Desktop large Hub downloads by @Souravrajvi0 in https://github.com/unslothai/unsloth/pull/10849
* Studio: anchor the worker source read on the test file, not the working directory by @danielhanchen in https://github.com/unslothai/unsloth/pull/11495
* Studio: make the contrast slider reach surfaces and controls by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11459
* Studio: scale spacing with the UI font size by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11458
* Studio: make the drag preferences test compile, and ask the stricter question by @danielhanchen in https://github.com/unslothai/unsloth/pull/11503
* Stop unsloth chat and unsloth inference ignoring the model's recommended settings by @NilayYadav in https://github.com/unslothai/unsloth/pull/11478
* Studio: keep <think> text visible in the reply when Thinking is off by @NilayYadav in https://github.com/unslothai/unsloth/pull/11481
* Studio: keep a streaming code fence highlighted, and window its spans by @danielhanchen in https://github.com/unslothai/unsloth/pull/11423
* fix(studio): track the repository resolved by the training loader by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11462
* Studio: promote -ts into tensor_split before manual strip by @indrajeetapache in https://github.com/unslothai/unsloth/pull/11334
* studio/frontend: keep shutdown dialog open when server-stop fails by @danielhanchen in https://github.com/unslothai/unsloth/pull/5637
* Repair a quarantined llama.cpp runtime at launch instead of failing at model load by @danielhanchen in https://github.com/unslothai/unsloth/pull/10494
* Stop the installers importing the download stack just to be imported by @danielhanchen in https://github.com/unslothai/unsloth/pull/11446
* Studio: show what the caches cost and let them be cleared by @danielhanchen in https://github.com/unslothai/unsloth/pull/10528
* Stop unsloth chat and unsloth inference reloading a 16-bit model in 4-bit by @NilayYadav in https://github.com/unslothai/unsloth/pull/11477
* Unsloth Studio: make a newly created API key selectable when the clipboard is unavailable by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11489
* Cut CI queue: scope push triggers, and rebalance Repo tests (CPU) on measured runner time by @danielhanchen in https://github.com/unslothai/unsloth/pull/11504
* Fix the update card clipping its own buttons at the 20px interface size by @danielhanchen in https://github.com/unslothai/unsloth/pull/11510
* Studio: read the scaled lengths and colours the UI contracts were written against by @danielhanchen in https://github.com/unslothai/unsloth/pull/11505
* Keep credentials out of cached directories, and audit the org against the miri cache disclosure by @danielhanchen in https://github.com/unslothai/unsloth/pull/11517
* Stop dividing the GRPO eval loss by the accumulation steps by @vineethsaivs in https://github.com/unslothai/unsloth/pull/11337
* studio: revamp the logs tab by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11508
* Studio: resolve upstream llama.cpp releases, assets and digests without the GitHub API by @danielhanchen in https://github.com/unslothai/unsloth/pull/5886
* Build and publish signed prebuilt CUDA 13 wheels for torch 2.13 and 2.14 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11515
* Let the cache row's path label follow the interface font size by @danielhanchen in https://github.com/unslothai/unsloth/pull/11520
* Give the Docker step harnesses the environment those steps now declare by @danielhanchen in https://github.com/unslothai/unsloth/pull/11521
* Skip TRL's redundant kbit prep for models Unsloth already prepared by @danielhanchen in https://github.com/unslothai/unsloth/pull/11494
* Keep generated RL configs on the reentrant gradient checkpoint path by @danielhanchen in https://github.com/unslothai/unsloth/pull/11519
* Install the import-time Python dependencies before smoke testing a prebuilt wheel by @danielhanchen in https://github.com/unslothai/unsloth/pull/11527
* Do not auto-enable padding-free for a model whose forward cannot take it by @danielhanchen in https://github.com/unslothai/unsloth/pull/11468
* Find a GGUF companion published after the weights on the chat load route by @danielhanchen in https://github.com/unslothai/unsloth/pull/11448
* Report a Linux ROCm APU host-backed pool of zero as measured, not unknown by @danielhanchen in https://github.com/unslothai/unsloth/pull/11451
* Studio: validate the reasoning controls instead of coercing them by @Apoze in https://github.com/unslothai/unsloth/pull/9595
* Studio tool loop: close two gaps in the static screens that gate model-written code by @danielhanchen in https://github.com/unslothai/unsloth/pull/11399
* Fixes to XPU offloading during training by @ChowChowSonic in https://github.com/unslothai/unsloth/pull/9534
* Studio: chat with a full fine-tune in 16-bit instead of a 4-bit copy by @NilayYadav in https://github.com/unslothai/unsloth/pull/11479
* Docker: pass JUPYTER_PORT and the notebook skip options through run.sh by @NilayYadav in https://github.com/unslothai/unsloth/pull/11484
* Let a pre-quant checkpoint be safetensors instead of a pickle by @danielhanchen in https://github.com/unslothai/unsloth/pull/11394
* Rework Studio dark mode onto a single surface and text scale by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11502
* Let a hosted pre-cast text encoder be safetensors, not only a pickle by @danielhanchen in https://github.com/unslothai/unsloth/pull/11454
* Studio: warn when the disk is running low, and point at the caches by @danielhanchen in https://github.com/unslothai/unsloth/pull/11439
* Add the Qwen-Image-2.1 family, and the Diffusers it needs by default by @danielhanchen in https://github.com/unslothai/unsloth/pull/11405
* Route Qwen-Image-2.1 GGUFs to the sd.cpp engine by @danielhanchen in https://github.com/unslothai/unsloth/pull/11507
* Resolve the auto class for omni checkpoints, and guard embedding lookups by @danielhanchen in https://github.com/unslothai/unsloth/pull/11467
* Studio: stop one audio chat from blocking every other chat by @NilayYadav in https://github.com/unslothai/unsloth/pull/11483
* fix(grpo): autocast with DEVICE_TYPE_TORCH instead of a pinned "cuda" by @li-lizhe in https://github.com/unslothai/unsloth/pull/11461
* Warn when the installed transformers drops a pre-quantized checkpoint's bnb quant state by @danielhanchen in https://github.com/unslothai/unsloth/pull/11450
* Install a prebuilt wheel's declared runtime dependencies before smoke importing it by @danielhanchen in https://github.com/unslothai/unsloth/pull/11536
* Studio: replace reasoning pagination with a continuous transcript by @wasimysaid in https://github.com/unslothai/unsloth/pull/11522
* Keep the compiler on for native architectures loaded with trust_remote_code = True by @danielhanchen in https://github.com/unslothai/unsloth/pull/11516
* Studio: load a 16-bit Continued Pretraining run in 16-bit in Chat by @NilayYadav in https://github.com/unslothai/unsloth/pull/11480
* Load prequantized multimodal checkpoints correctly on transformers 5.4 and 5.5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11452
* Make main formatter-clean so pre-commit stops rewriting every PR by @danielhanchen in https://github.com/unslothai/unsloth/pull/11540
* Export into a directory the temp-file test owns by @danielhanchen in https://github.com/unslothai/unsloth/pull/11532
* Read the packing blocker's branch by behaviour, not by its spelling by @danielhanchen in https://github.com/unslothai/unsloth/pull/11534
* Make the GPU probe harness name the helper it failed to extract by @danielhanchen in https://github.com/unslothai/unsloth/pull/11538
* Make the thinking trace read as a trace, not as the answer by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11541
* Load the published Qwen-Image-2.1 INT8 checkpoint, and four more follow-ups by @danielhanchen in https://github.com/unslothai/unsloth/pull/11535
* Studio: take the hosted pre-cast fp8 text encoder by default on Qwen-Image-2.1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11539
* Pin the Dynamic 2.0 4-bit text encoder for Qwen-Image-2.1's native route by @danielhanchen in https://github.com/unslothai/unsloth/pull/11542
* Avoid 8-bit training optimizer on XPU by @ousamabenyounes in https://github.com/unslothai/unsloth/pull/10213
* Lift the Classic light muted token to clear AA by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11544
* Fix push_to_hub_merged for full fine-tunes on transformers 5 by @NilayYadav in https://github.com/unslothai/unsloth/pull/11482
* Studio: reuse one database connection per thread for durable chat generation runs by @danielhanchen in https://github.com/unslothai/unsloth/pull/11525
* feat(chat): make agentic turns survive tab-close by @CelesteHeartsong in https://github.com/unslothai/unsloth/pull/10365
* Qwen-Image-2.1 fails every render at 20+ steps: FBCache cannot run on a prefix KV cache by @danielhanchen in https://github.com/unslothai/unsloth/pull/11546
* Unsloth Studio: give the memory planner the companion split on a pipeline image load by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11548
* Bump install.sh / install.ps1 pin to unsloth>=2026.9.8 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11550
* Stub the low-disk check in the tool-approval harness by @danielhanchen in https://github.com/unslothai/unsloth/pull/11552
* Studio: drag sidebar rows in the desktop app, and drop the drag hint by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11561
* Studio: install the pinned Diffusers commit from a zip when there is no git by @danielhanchen in https://github.com/unslothai/unsloth/pull/11565
* Studio: finish an update with the installer it installs by @oobabooga in https://github.com/unslothai/unsloth/pull/11566
* Bump install.sh / install.ps1 pin to unsloth>=2026.9.9 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11570
* Keep weight_decay when embedding_learning_rate is set by @vineethsaivs in https://github.com/unslothai/unsloth/pull/11470
* Studio: hold the sidebar drag escape click guard until the button comes up by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11568
* Repair the four a-k checks the Qwen-Image-2.1 wave left red by @danielhanchen in https://github.com/unslothai/unsloth/pull/11573
* Let the composite renaming tests run inside the transformers 5.4 window by @danielhanchen in https://github.com/unslothai/unsloth/pull/11572
* Keep scrolling to the end of the reasoning transcript in the thread settings driver by @danielhanchen in https://github.com/unslothai/unsloth/pull/11575
* Let the stdout guard read writes, not flushes by @danielhanchen in https://github.com/unslothai/unsloth/pull/11576
* Baseline the seven unsloth-zoo 2026.9.7 findings after re-review by @danielhanchen in https://github.com/unslothai/unsloth/pull/11577
* Expect Dense (bf16) after a declined encoder precision by @danielhanchen in https://github.com/unslothai/unsloth/pull/11580
* Compare a full pass to a skipping one without the click both record as unmet by @danielhanchen in https://github.com/unslothai/unsloth/pull/11578
* Do not scan a dep that only an extra installs, platform marker or not by @danielhanchen in https://github.com/unslothai/unsloth/pull/11582
* Studio: fork a chat from its row menu by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11564
* Studio: load Qwen-Image-2.1 GGUFs through diffusers on GPU hosts by @danielhanchen in https://github.com/unslothai/unsloth/pull/11604
* Import the row fork's modules statically, which puts main back under the startup budget by @danielhanchen in https://github.com/unslothai/unsloth/pull/11588
* Probe Studio before reading the deadline in unsloth-studio-update by @danielhanchen in https://github.com/unslothai/unsloth/pull/11590

## New Contributors
* @AzizMuminov made their first contribution in https://github.com/unslothai/unsloth/pull/10802
* @taking-lying-flat made their first contribution in https://github.com/unslothai/unsloth/pull/11266
* @TarzanZhao made their first contribution in https://github.com/unslothai/unsloth/pull/11238
* @ChowChowSonic made their first contribution in https://github.com/unslothai/unsloth/pull/9534

**Full Changelog**: https://github.com/unslothai/unsloth/compare/v0.1.811-beta...v0.1.814-beta

## 📥 Install
Unsloth can be used in three ways: **[Unsloth Desktop](https://unsloth.ai/download)**, the desktop app; **[Unsloth Studio](https://unsloth.ai/docs/new/studio/)**, the web UI; or **Unsloth Core**, the code based version.

### Unsloth Desktop (recommended)

<table>
  <tr>
    <td><b>Platform</b></td>
    <td><b>Link</b></td>
  </tr>
  <tr>
    <td><b>Windows</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Windows.exe'>Download</a></td>
  </tr>
  <tr>
    <td><b>macOS</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-MacOS.dmg'>Download</a></td>
  </tr>
  <tr>
    <td><b>Linux x64 / Ubuntu (deb)</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Ubuntu.deb'>Download</a></td>
  </tr>
  <tr>
    <td><b>Linux ARM64 / Ubuntu 24.04+ (deb)</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Ubuntu-ARM64.deb'>Download</a></td>
  </tr>
  <tr>
    <td><b>Linux x64 (AppImage)</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Linux.AppImage'>Download</a></td>
  </tr>
  <tr>
    <td><b>Windows ARM64</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/download/v0.1.811-beta/Unsloth-Desktop-Windows-ARM64.exe'>Download</a></td>
  </tr>
</table>