v0.1.808-beta

This is a large performance and reliability + bug fix release for Unsloth

## Highlights
* 1.2-1.7x faster diffusion. AMD 20% perf boost vs ROCM via Vulkan
* 2x faster updating, remove SAC + AV false positives for Windows
* Blender MCP, detect Hermes, AMD gibberish fixed (reported to AMD)
* Over 250+ bug fixes, 60% smaller binaries and performance improvements
* Strix iGPU BIOS popup - 3x faster inference if more VRAM for iGPU
* Updated PyTorch to 2.11 from 2.10 - 2.14 will be soon. MLX improvements
* Default to gpt-6-astra for Codex logins and improve OpenAI API processing
* MLX fixed when self healing / updating, causing slowdowns for inf + training
* Fix AppImage being very laggy. Fixed RAG upload issues stuck at 28% and API issues
* New Docker image published at https://hub.docker.com/r/unsloth/unsloth for Studio and notebooks

## Performance Boosts
- Diffusion is 1.2x to 1.7x faster for INT8 / FP8 pathways - all models accelerated.
- 23% faster prompt processing and 8% faster generation on Strix Halo.
- Gated-delta models now train up to 25% faster on Apple Silicon.
- Quantized MLX KV caches use up to 74% less prompt memory than before.

## AMD + Windows

- Strix Halo and Strix Point now default to Vulkan for faster inference.
- AMD iGPUs without ROCm now use Vulkan instead of CPU on Linux.
- Windows llama.cpp binaries are now signed to reduce Smart App Control blocks.
- Windows now clearly explains when code integrity blocks model loading.
- Reinstalling Unsloth on Windows keeps your supported PyTorch version.

## MLX + Apple Silicon

- Fresh Mac installs keep MLX training and exports working.
- Fine-tune with DoRA and more DPO loss types on Apple Silicon.
- Batched MLX generation now streams and samples each chat independently.
- More multimodal models can be fine-tuned using text-only datasets.

## Studio + API

- RAG and document uploads are faster, show clearer progress and no longer appear stuck.
- Studio menus, Find and Settings now open more smoothly.
- GPT-6 Astra is available for Codex logins with Low through Max reasoning controls.
- Improved OpenAI-compatible API streaming, audio input and model loading.
- AppImage builds are more reliable with a pinned release toolchain.

## Installs + Docker

- PyTorch 2.11 is now the default across supported installers.
- New NVIDIA Docker images for training and Studio, from Turing to Blackwell.
- Native AMD64 and ARM64 images are available from Docker Hub.
- The Unsloth Python package is now >60% smaller.

Docker: https://hub.docker.com/r/unsloth/unsloth

## Download Unsloth Desktop

Unsloth Desktop is **free and open source**. Download it for:

- **Windows**
- **macOS**
- **Linux**

**[🦥 Download Unsloth Desktop](https://unsloth.ai/download)**

## What's Changed
* Publish only the wheel to PyPI, not the sdist by @danielhanchen in https://github.com/unslothai/unsloth/pull/10202
* Bump install.sh / install.ps1 pin to unsloth>=2026.9.2 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10203
* Add DeepSeek Harness to unsloth start by @NilayYadav in https://github.com/unslothai/unsloth/pull/10097
* Studio: stop generating when the client goes away by @NilayYadav in https://github.com/unslothai/unsloth/pull/10096
* Studio: keep what you typed with an image when a chat is shortened by @NilayYadav in https://github.com/unslothai/unsloth/pull/10165
* Order the Ollama stop-token scan instead of leaving it to a set by @vineethsaivs in https://github.com/unslothai/unsloth/pull/10101
* Studio: replay a web_search result the way the client sent it by @Lyxot in https://github.com/unslothai/unsloth/pull/10131
* Studio: say when a chat is too long instead of showing a raw error by @NilayYadav in https://github.com/unslothai/unsloth/pull/10090
* Restore the comment tails the trim cut off by @danielhanchen in https://github.com/unslothai/unsloth/pull/10205
* Studio: accept subtitle, caption and the rest of the common attachment formats by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/9738
* Studio: count only this adapter's engines in the Windows AMD GPU utilization by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/9317
* Fix diffusion offload host RAM retention by @wasimysaid in https://github.com/unslothai/unsloth/pull/10184
* Let audio models load into CPU RAM instead of the GPU by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10054
* Stop the zoo digest pin living in two places by @danielhanchen in https://github.com/unslothai/unsloth/pull/10206
* Format test_video_backend.py the way the pinned ruff wants by @danielhanchen in https://github.com/unslothai/unsloth/pull/10235
* Security audit: re-approve the four torch findings that 2.14.0 opened by @danielhanchen in https://github.com/unslothai/unsloth/pull/10233
* unsloth_cli tests: stop the mklink kwargs assertion breaking on an unrelated keyword by @danielhanchen in https://github.com/unslothai/unsloth/pull/10231
* Pin the multi-turn determinism probe to a backend that can be deterministic by @danielhanchen in https://github.com/unslothai/unsloth/pull/10234
* CI: stop grep -c || echo 0 producing a two-line count by @danielhanchen in https://github.com/unslothai/unsloth/pull/10232
* Security audit: re-approve the huggingface-hub backoff loop for 1.30.0 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10244
* Stop the stall-window comparison being decided by one noisy sample by @danielhanchen in https://github.com/unslothai/unsloth/pull/10246
* Studio: report a cut off /v1/responses turn as incomplete by @NilayYadav in https://github.com/unslothai/unsloth/pull/10225
* Studio: keep a still-advancing llama.cpp prefill from timing out by @oobabooga in https://github.com/unslothai/unsloth/pull/10172
* Studio: bring the Llasa top_p default inside the range the API accepts by @danielhanchen in https://github.com/unslothai/unsloth/pull/9194
* CLI: pass --hf-token to the export checkpoint load by @NilayYadav in https://github.com/unslothai/unsloth/pull/10218
* Pin AppImage plugin to a versioned release by @wasimysaid in https://github.com/unslothai/unsloth/pull/10201
* Studio: thread hf_token through load_and_format_dataset by @KrishPatel1905 in https://github.com/unslothai/unsloth/pull/8924
* Studio: stop the update banners floating off the corner by @oobabooga in https://github.com/unslothai/unsloth/pull/10229
* fix(studio): stop handing a wildcard bind's public IP to LAN peers by @amanyagami in https://github.com/unslothai/unsloth/pull/9912
* Studio: report free memory on Mac as what is actually free by @NilayYadav in https://github.com/unslothai/unsloth/pull/10224
* Studio: match the update banner tests on classes, not on file substrings by @danielhanchen in https://github.com/unslothai/unsloth/pull/10267
* Studio: size a Deep Research run to the model it actually runs on by @NilayYadav in https://github.com/unslothai/unsloth/pull/10220
* Studio: fall back when a page declares a charset we do not know by @NilayYadav in https://github.com/unslothai/unsloth/pull/10221
* Studio: pin which rejection site the confirm-stream tests assert by @danielhanchen in https://github.com/unslothai/unsloth/pull/10284
* Studio: turn the generation prompt off on the final answer continuation by @NilayYadav in https://github.com/unslothai/unsloth/pull/10219
* Studio: let the API load settings panel forget an entry by @NilayYadav in https://github.com/unslothai/unsloth/pull/10160
* Windows torch release preservation and a torch 2.11 default across installers by @danielhanchen in https://github.com/unslothai/unsloth/pull/7256
* Add cu128/cu126/cu130 torch 2.11.0 extras (torch2110) with xformers 0.0.35 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6955
* Studio: assert the write ordering, not what the write resolves to by @danielhanchen in https://github.com/unslothai/unsloth/pull/10289
* Kaggle: fix the nightly, which has never once got as far as Kaggle by @danielhanchen in https://github.com/unslothai/unsloth/pull/10270
* Keep probing for a usable nvidia-smi before falling back to ROCm on Windows by @danielhanchen in https://github.com/unslothai/unsloth/pull/4958
* Trim comments in studio/frontend chat, model picker and media features by @danielhanchen in https://github.com/unslothai/unsloth/pull/10114
* Add a dispatch-only Docker Hub credential probe by @danielhanchen in https://github.com/unslothai/unsloth/pull/10291
* Fix Studio find-in-page correctness and typing by @wasimysaid in https://github.com/unslothai/unsloth/pull/10209
* Studio: stop exports timing out while they are still working by @NilayYadav in https://github.com/unslothai/unsloth/pull/10258
* Stop unsloth chat from reloading a model the server already has by @NilayYadav in https://github.com/unslothai/unsloth/pull/10257
* Studio: stop hiding every model pulled with ollama pull by @NilayYadav in https://github.com/unslothai/unsloth/pull/10222
* Anchor studio source slices on declarations, not on comment text by @danielhanchen in https://github.com/unslothai/unsloth/pull/10297
* Studio: fix Restore settings breaking preset saving by @NilayYadav in https://github.com/unslothai/unsloth/pull/10256
* Lazy-load Studio entry surfaces by @wasimysaid in https://github.com/unslothai/unsloth/pull/10236
* Restore persisted Studio resource monitor by @wasimysaid in https://github.com/unslothai/unsloth/pull/10237
* fix(frontend): render data-URI images in chat markdown by @CelesteHeartsong in https://github.com/unslothai/unsloth/pull/10269
* Fix Hub GGUF fit estimates for companion files by @Imagineer99 in https://github.com/unslothai/unsloth/pull/10249
* Pin the auth mount gate to its behaviour, not one spelling of the guard by @danielhanchen in https://github.com/unslothai/unsloth/pull/10307
* Trim comments in tests, scripts, CI helpers and unsloth_cli by @danielhanchen in https://github.com/unslothai/unsloth/pull/10113
* Add Unsloth Docker images (base + Studio) for any NVIDIA GPU host, Ampere through Blackwell by @danielhanchen in https://github.com/unslothai/unsloth/pull/5748
* Restore the facts the frontend comment trim reworded away by @danielhanchen in https://github.com/unslothai/unsloth/pull/10292
* Install the cublas wheel NVIDIA actually publishes for the bundle's CUDA major by @danielhanchen in https://github.com/unslothai/unsloth/pull/10324
* Studio: fix whisper prebuilt pairing on linux-arm64 with a GPU llama runtime by @danielhanchen in https://github.com/unslothai/unsloth/pull/10278
* Docker publish: verify the merged manifest against the arches' children by @danielhanchen in https://github.com/unslothai/unsloth/pull/10333
* Docker: describe the shipped images on Docker Hub and in the README by @danielhanchen in https://github.com/unslothai/unsloth/pull/10334
* Stop asserting the titlebar navigation padding from source by @danielhanchen in https://github.com/unslothai/unsloth/pull/10326
* Restore facts the comment trim cut from the unsloth package by @danielhanchen in https://github.com/unslothai/unsloth/pull/10328
* Restore facts the comment trim cut from the Studio backend by @danielhanchen in https://github.com/unslothai/unsloth/pull/10329
* Restore facts the comment trim cut from the studiobench harness by @danielhanchen in https://github.com/unslothai/unsloth/pull/10331
* fix(studio): align desktop titlebar buttons by @mahiatlinux in https://github.com/unslothai/unsloth/pull/10321
* Docker publish: sync the Hub page with DOCKER_API_KEY on the namespace route by @danielhanchen in https://github.com/unslothai/unsloth/pull/10343
* Revert "Studio: keep a Downloads entry when the list is empty" by @Etherll in https://github.com/unslothai/unsloth/pull/10298
* Make the credential probe actually delete its throwaway tag by @danielhanchen in https://github.com/unslothai/unsloth/pull/10294
* Docker: print the Studio admin password in docker logs, add UNSLOTH_STUDIO_PASSWORD by @danielhanchen in https://github.com/unslothai/unsloth/pull/10344
* Docker: retry the deadsnakes PPA add so a Launchpad 504 does not fail the build by @danielhanchen in https://github.com/unslothai/unsloth/pull/10363
* Docker publish: keep the tag list to core, latest, studio, releases and dated nightly pins by @danielhanchen in https://github.com/unslothai/unsloth/pull/10364
* Studio: take GGML_CUDA_ENABLE_UNIFIED_MEMORY only when the weights outgrow the APU carve-out by @danielhanchen in https://github.com/unslothai/unsloth/pull/10351
* Studio: prefer the self-contained MTP head so llama-server's --fit can measure it by @danielhanchen in https://github.com/unslothai/unsloth/pull/10342
* Studio: measure the real size of attached documents instead of guessing by @NilayYadav in https://github.com/unslothai/unsloth/pull/10318
* Bump the npm-frontend-security group across 1 directory with 5 updates by @dependabot[bot] in https://github.com/unslothai/unsloth/pull/10194
* Chat: keep the payload when a provider streams arguments as a decoded object by @lonexreb in https://github.com/unslothai/unsloth/pull/10275
* deps: bump setuptools build pins and raise the ip-address override to 10.3.1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10138
* studio: stop find in page matching part way through a grapheme by @danielhanchen in https://github.com/unslothai/unsloth/pull/10200
* Studio: unsloth chat loads the model you picked, not another GGUF in the same folder by @NilayYadav in https://github.com/unslothai/unsloth/pull/10357
* Studio: keep the overlay rail honest about what it reserves by @danielhanchen in https://github.com/unslothai/unsloth/pull/10372
* Kaggle T4: fail a batched row that generates nothing while its single does not by @danielhanchen in https://github.com/unslothai/unsloth/pull/10367
* Docker: one command sets up the NVIDIA Container Toolkit on the host by @danielhanchen in https://github.com/unslothai/unsloth/pull/10366
* Kaggle T4: describe the batched-generation expectation as a precision property, not a pending fix by @danielhanchen in https://github.com/unslothai/unsloth/pull/10377
* Studio: keep an API key from borrowing the operator's Hugging Face token (#10126) by @harshaygadekar in https://github.com/unslothai/unsloth/pull/10142
* Use hybrid uv cache migration for Studio installs by @wasimysaid in https://github.com/unslothai/unsloth/pull/10204
* feat(chat): render on-disk sandbox images in chat markdown by @CelesteHeartsong in https://github.com/unslothai/unsloth/pull/10361
* Point unsloth start openclaw memory search at the Unsloth embeddings API by @NilayYadav in https://github.com/unslothai/unsloth/pull/10320
* Studio: serve /v1/embeddings from the configured embedding model when the loaded GGUF cannot by @NilayYadav in https://github.com/unslothai/unsloth/pull/10315
* Studio: stop converting the model twice when exporting GGUF to the Hub by @NilayYadav in https://github.com/unslothai/unsloth/pull/10317
* Studio: return an error when tools are sent to a model that cannot use them by @NilayYadav in https://github.com/unslothai/unsloth/pull/10314
* Local Agent Guides CI: judge a finished turn the CLI would not exit from by @danielhanchen in https://github.com/unslothai/unsloth/pull/10399
* Docker toolkit installer: stop pipefail turning a grep match into a miss by @danielhanchen in https://github.com/unslothai/unsloth/pull/10402
* Fix composer send and stop icon centering across display scales by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10407
* Give push_to_ollama the arguments create_ollama_modelfile takes by @vineethsaivs in https://github.com/unslothai/unsloth/pull/10304
* Kaggle CI: dispatch the kernel and collect it later, instead of holding a runner by @danielhanchen in https://github.com/unslothai/unsloth/pull/10183
* Studio: make the chat and sidebar menus open without freezing the page by @NilayYadav in https://github.com/unslothai/unsloth/pull/10262
* Studio: unbreak the frontend suite and the icon alignment check by @danielhanchen in https://github.com/unslothai/unsloth/pull/10413
* Studio smoke: report a transport failure instead of crashing on one by @danielhanchen in https://github.com/unslothai/unsloth/pull/10420
* Give `unsloth studio update` the same uv cache the backend already uses by @danielhanchen in https://github.com/unslothai/unsloth/pull/10386
* Studio installer: torch 2.11 on Linux CPU, Vulkan for AMD without ROCm, gfx1033 gate, and three install fixes by @danielhanchen in https://github.com/unslothai/unsloth/pull/8412
* Studio: four load and update messages that told the user the wrong thing by @danielhanchen in https://github.com/unslothai/unsloth/pull/10383
* Studio: keep the llama-server prompt cache on a shared-memory GPU by @danielhanchen in https://github.com/unslothai/unsloth/pull/10382
* Studio: retry as one sequence when llama.cpp refuses a unified KV cache by @danielhanchen in https://github.com/unslothai/unsloth/pull/10371
* Publish Docker on the daily schedule, and move layer cache off the Actions quota by @danielhanchen in https://github.com/unslothai/unsloth/pull/10429
* Studio: report a throughput the engine could have produced by @danielhanchen in https://github.com/unslothai/unsloth/pull/10384
* Stop shipping the frontend source tree and desktop crate in the wheel by @danielhanchen in https://github.com/unslothai/unsloth/pull/10426
* Studio: optimize video galleries by removing per-card media pipelines by @oobabooga in https://github.com/unslothai/unsloth/pull/10228
* Studio: do not crash a model load when shutdown tears the server down under it by @danielhanchen in https://github.com/unslothai/unsloth/pull/10369
* Studio: route AMD integrated GPUs to the Vulkan llama.cpp prebuilt by @danielhanchen in https://github.com/unslothai/unsloth/pull/10381
* Studio: give a local GGUF one variant identity, so an API load applies its saved settings by @Lyxot in https://github.com/unslothai/unsloth/pull/10238
* Studio: reuse MLX vision-language prompt prefixes between turns by @Lyxot in https://github.com/unslothai/unsloth/pull/10239
* Studio: share one Hub model_info read across the model-config probes by @Lyxot in https://github.com/unslothai/unsloth/pull/10230
* Keep the LoRA GGUF and compressed-tensors converters off the operator's Hugging Face token by @Souravrajvi0 in https://github.com/unslothai/unsloth/pull/10263
* Studio: load the speech model an audio API request asks for by @NilayYadav in https://github.com/unslothai/unsloth/pull/10217
* Studio: stop rejecting provider API keys longer than 190 bytes by @Lyxot in https://github.com/unslothai/unsloth/pull/10421
* Studio: dial the bound address for in-process API calls, not a loopback literal by @Lyxot in https://github.com/unslothai/unsloth/pull/10140
* Studio: admit a bearer that carries no token under keyless API access by @Lyxot in https://github.com/unslothai/unsloth/pull/10423
* Studio: convert pasted images when tools are turned on by @NilayYadav in https://github.com/unslothai/unsloth/pull/10094
* Fix GGUF catalog snapshot selection by @Imagineer99 in https://github.com/unslothai/unsloth/pull/10210
* Studio: accept the standard input_audio message part by @NilayYadav in https://github.com/unslothai/unsloth/pull/10255
* Studio: use the eval dataset for TTS and codec audio training by @NilayYadav in https://github.com/unslothai/unsloth/pull/10259
* studio: use the hermes and deepseek logos in the agents picker by @mahiatlinux in https://github.com/unslothai/unsloth/pull/10266
* Studio: use every thinking level a local model offers by @NilayYadav in https://github.com/unslothai/unsloth/pull/10458
* Bind the amp helpers on the MLX runtime too by @vineethsaivs in https://github.com/unslothai/unsloth/pull/9447
* OpenAI stream: gate UI control frames behind X-Unsloth-Events opt-in by @denys-bushuliak in https://github.com/unslothai/unsloth/pull/10362
* Clean up Hub runtime/config coupling  by @Sneakr in https://github.com/unslothai/unsloth/pull/9886
* Stop unsloth start codex and openclaw from inheriting provider API keys by @NilayYadav in https://github.com/unslothai/unsloth/pull/10316
* Studio: add GPT-6 Astra to the fallback catalog and reasoning controls by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10475
* Studio: let Deep Research write longer reports on saved connections by @NilayYadav in https://github.com/unslothai/unsloth/pull/10254
* Fix the red Cross-platform parity run: decode PowerShell stdout as UTF-8 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10476
* Studio: name the llama.cpp backend in the install log, and stop labelling Windows ROCm torch as CPU by @danielhanchen in https://github.com/unslothai/unsloth/pull/10370
* notebook_validator: read shell the way bash does, and state the torch/torchcodec contract by @danielhanchen in https://github.com/unslothai/unsloth/pull/10414
* Docker Studio: build the labextension with @jupyter/builder so it never reaches for GitHub by @danielhanchen in https://github.com/unslothai/unsloth/pull/10483
* Drop late inference responses after their request is released by @ianderse in https://github.com/unslothai/unsloth/pull/10388
* Studio: stop re-fetching every llama.cpp release from the GitHub API by @NilayYadav in https://github.com/unslothai/unsloth/pull/10461
* Do not self-heal MLX into a --no-torch install by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10409
* Pin tokenizers with transformers so Apple Silicon keeps Train and Export by @danielhanchen in https://github.com/unslothai/unsloth/pull/10431
* Make the venv-hardening decode strict, and say which codepoints differed by @danielhanchen in https://github.com/unslothai/unsloth/pull/10432
* Stop unsloth train from ignoring config keys it does not know by @NilayYadav in https://github.com/unslothai/unsloth/pull/10452
* Apply the load settings unsloth start dsh is given by @NilayYadav in https://github.com/unslothai/unsloth/pull/10451
* Studio: keep an Ollama Modelfile SYSTEM prompt in force when the date line is on by @HughScott2002 in https://github.com/unslothai/unsloth/pull/10463
* Studio: fix sending an image with no caption on Anthropic by @NilayYadav in https://github.com/unslothai/unsloth/pull/10455
* Studio: keep the uploaded eval dataset for Whisper and audio VLM runs too by @danielhanchen in https://github.com/unslothai/unsloth/pull/10482
* torchcodec: cover torch 2.11 in the compatibility guard and pin per torch minor by @danielhanchen in https://github.com/unslothai/unsloth/pull/7474
* fix(tokenizer): enable add_bos_token for Gemma 4 base models by @Souravrajvi0 in https://github.com/unslothai/unsloth/pull/10312
* Studio: make the Min P and Repetition Penalty sliders work by @NilayYadav in https://github.com/unslothai/unsloth/pull/10456
* Studio: show mlx-community models as MLX in the model picker by @NilayYadav in https://github.com/unslothai/unsloth/pull/10457
* Recognise Windows code integrity blocks, and require the runtime we ship by @danielhanchen in https://github.com/unslothai/unsloth/pull/10404
* Assert the release publishing path uploads the wheel only by @danielhanchen in https://github.com/unslothai/unsloth/pull/10419
* Studio: harden the codec-audio eval config added in #10259 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10477
* Scan release bundles for potentially unwanted applications too by @danielhanchen in https://github.com/unslothai/unsloth/pull/10478
* Refactor setup.sh into modular functions and restructure main execution flow by @xerrni in https://github.com/unslothai/unsloth/pull/5408
* fix: private dataset splits/metadata not loading in Studio UI by @Shivamjohri247 in https://github.com/unslothai/unsloth/pull/4965
* Feat: add unsloth skill, fix cli by @TimPietrusky in https://github.com/unslothai/unsloth/pull/4443
* bump trl version by @danielhanchen in https://github.com/unslothai/unsloth/pull/4248
* Add persistent wiki-RAG pipeline with vendored Graphify, lint-driven enrichment, and debug/maintenance APIs by @zohairshafi in https://github.com/unslothai/unsloth/pull/5103
* Fix Windows setup access-denied on existing llama-server ACLs by @aiSynergy37 in https://github.com/unslothai/unsloth/pull/4881
* [WIP] Apple Silicon (MPS/Metal) Support by @danielhanchen in https://github.com/unslothai/unsloth/pull/4259
* docs: improve documentation on exporting models from Colab by @danielhanchen in https://github.com/unslothai/unsloth/pull/4247
* Add Qwen3-Omni Support with Optimized MTP Fine-Tuning by @danielhanchen in https://github.com/unslothai/unsloth/pull/4235
* Studio: allow DoRA training on Apple Silicon by @Lyxot in https://github.com/unslothai/unsloth/pull/7508
* Studio: ask for a password on a raw `-H 0.0.0.0` bind too by @danielhanchen in https://github.com/unslothai/unsloth/pull/10485
* fix(studio): refresh profile stats and mode-aware activity summary by @Souravrajvi0 in https://github.com/unslothai/unsloth/pull/9402
* Studio: keep queued prompts when generation is stopped by @claxman in https://github.com/unslothai/unsloth/pull/10445
* Fix the red parity gate: run_pwsh has to set the writing end of the pipe too by @danielhanchen in https://github.com/unslothai/unsloth/pull/10484
* Studio: say when a file cannot be read instead of ignoring it by @NilayYadav in https://github.com/unslothai/unsloth/pull/10261
* Studio: recover saved context when switching back to a snapshot path by @claxman in https://github.com/unslothai/unsloth/pull/10447
* Studio: install the supported MLX stack on fresh macOS installs by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10403
* Cache janitor: sweep every two hours, drop dead buildkit blobs on sight, keep one CodeQL overlay base by @danielhanchen in https://github.com/unslothai/unsloth/pull/10488
* Studio: stop installing flash-linear-attention and tilelang, unsloth_zoo vendors the GDN kernels by @danielhanchen in https://github.com/unslothai/unsloth/pull/10487
* Docker Hub README: xformers is amd64 only, GB10 runs through PTX, canonical docs link by @danielhanchen in https://github.com/unslothai/unsloth/pull/10492
* Docker Studio: keep the base image's CUDA llama.cpp instead of the CPU prebuilt by @danielhanchen in https://github.com/unslothai/unsloth/pull/10495
* Make the formatter refuse a ruff it is not pinned to, and re-format the four files that drifted by @danielhanchen in https://github.com/unslothai/unsloth/pull/10512
* Tests: let the xFormers pin guards hold on both main and pip by @danielhanchen in https://github.com/unslothai/unsloth/pull/10514
* Studio: fix document upload stalls, races and text decoding by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10496
* Desktop: fix NVIDIA X11 WebKit fence exhaustion by @oobabooga in https://github.com/unslothai/unsloth/pull/10214
* Studio: quiet image and video generation logs  by @oobabooga in https://github.com/unslothai/unsloth/pull/9885
* Studio: record how a Windows install was launched and what it produced by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/8066
* Bump install.sh / install.ps1 pin to unsloth>=2026.9.3 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10521
* Fix main's red encoding guard, and the ruff drift in the same files by @danielhanchen in https://github.com/unslothai/unsloth/pull/10522
* Restore the executable bit on the formatter hook by @danielhanchen in https://github.com/unslothai/unsloth/pull/10519
* Studio: give the document upload simulations one deadline by @danielhanchen in https://github.com/unslothai/unsloth/pull/10517
* Let the torch tag check tell the handoff restore from a conditional report by @danielhanchen in https://github.com/unslothai/unsloth/pull/10524
* Keep scripts/run_ruff_format.py executable by @danielhanchen in https://github.com/unslothai/unsloth/pull/10518
* Unbreak CI on main: name the encoding on four checked-in file reads, and make the ruff format hook executable by @danielhanchen in https://github.com/unslothai/unsloth/pull/10523
* Pin the accelerator index for torchao too, and let a starved index pin fall back by @danielhanchen in https://github.com/unslothai/unsloth/pull/10490
* Release: do not fail the Windows Defender gate when the runner refuses real-time protection by @danielhanchen in https://github.com/unslothai/unsloth/pull/10527
* Studio: make three RAG tests hold on Windows by @danielhanchen in https://github.com/unslothai/unsloth/pull/10530
* Keep the loggers stub a package so tests/studio can be collected in one process by @danielhanchen in https://github.com/unslothai/unsloth/pull/10531
* Format two files that drifted from the pinned ruff on main by @danielhanchen in https://github.com/unslothai/unsloth/pull/10532
* Make build.sh executable so ./build.sh publish works from a clone by @danielhanchen in https://github.com/unslothai/unsloth/pull/10534
* CI: assert the installed unsloth CLI runs on the virgin Windows lane by @danielhanchen in https://github.com/unslothai/unsloth/pull/10535
* Stop unsloth start from killing a model download that is still running by @NilayYadav in https://github.com/unslothai/unsloth/pull/10453
* Studio: add opt-in first-use Blender MCP setup by @wasimysaid in https://github.com/unslothai/unsloth/pull/10443
* fix(desktop): trust the OS certificate store on the Linux desktop (#9218) by @yzxcj797 in https://github.com/unslothai/unsloth/pull/9240
* Fix Studio port fallback on Windows by @imherro in https://github.com/unslothai/unsloth/pull/9449
* Studio: recover compare-pane settings through a snapshot path by @claxman in https://github.com/unslothai/unsloth/pull/10505
* Studio: advise raising the iGPU memory allocation when a model overflows it by @danielhanchen in https://github.com/unslothai/unsloth/pull/10486
* Unbreak main: catch the install.ps1 handoff harness up with the block it drives by @danielhanchen in https://github.com/unslothai/unsloth/pull/10538
* Studio: clear dumpability before every crash the RAG probe child takes by @danielhanchen in https://github.com/unslothai/unsloth/pull/10547
* Bound RAG scope retirement by folder identity, not by the clock by @danielhanchen in https://github.com/unslothai/unsloth/pull/10561
* Studio: record what was on screen when an upload simulation fails by @danielhanchen in https://github.com/unslothai/unsloth/pull/10541
* Studio: only treat a real link definition as one by @NilayYadav in https://github.com/unslothai/unsloth/pull/10223
* Key the recreated-project delete tests to state, not to a call count by @danielhanchen in https://github.com/unslothai/unsloth/pull/10571
* Studio: detect models downloaded by Hermes by @NilayYadav in https://github.com/unslothai/unsloth/pull/10327
* Make run_ruff_format.py reject arguments it cannot honour instead of silently dropping them by @danielhanchen in https://github.com/unslothai/unsloth/pull/10578
* Collapse repeated fixture construction in the studio install tests by @danielhanchen in https://github.com/unslothai/unsloth/pull/10501
* Share the frontend source readers and the shell test harness by @danielhanchen in https://github.com/unslothai/unsloth/pull/10500
* Studio: decode a Windows device id before comparing it by @danielhanchen in https://github.com/unslothai/unsloth/pull/10536
* Studio: stop reading a success line like "Errors: 0" as a failed tool result by @NilayYadav in https://github.com/unslothai/unsloth/pull/10554
* Studio: remove OS tabs from generated agent command by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10586
* Resize Studio images and share Model Hub provider logos by @danielhanchen in https://github.com/unslothai/unsloth/pull/10427
* Share builders and setup helpers in the studio backend tests, part 2 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10498
* Keep model.device on the decoder while the embedding is offloaded to RAM by @danielhanchen in https://github.com/unslothai/unsloth/pull/10569
* Studio: keep torchao quantisation from mutating inductor config, and make its int8 GEMM capture-safe by @danielhanchen in https://github.com/unslothai/unsloth/pull/10525
* Studio: verify Hub credentials before cache-backed reads by @Souravrajvi0 in https://github.com/unslothai/unsloth/pull/10264
* Never import sentencepiece on Windows by @danielhanchen in https://github.com/unslothai/unsloth/pull/10593
* Attach dispatch hooks on a num_labels load that asked for fast_inference by @danielhanchen in https://github.com/unslothai/unsloth/pull/10509
* Bound the /validate calls one autoload cascade can send by @danielhanchen in https://github.com/unslothai/unsloth/pull/10508
* Trim the comments in scripts/uninstall.sh and scripts/uninstall.ps1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10596
* Studio: run document uploads off the event loop by @NilayYadav in https://github.com/unslothai/unsloth/pull/10552
* Re-run the pinned ruff formatter over two test files by @danielhanchen in https://github.com/unslothai/unsloth/pull/10601
* Apply the pinned formatter to test_hub_token_caller_identity.py by @danielhanchen in https://github.com/unslothai/unsloth/pull/10602
* Studio: clamp an unavailable reasoning effort to the nearest rung, not the weakest by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/10575
* fix: improve NVIDIA GPU detection and VS toolchain handling on Windows by @AronRubin in https://github.com/unslothai/unsloth/pull/5336
* Studio: find the AMD Vulkan driver through the Windows device registrations by @oobabooga in https://github.com/unslothai/unsloth/pull/10564
* Studio: skip torchcodec for torch below the compatibility table by @RudraBhaskar9439 in https://github.com/unslothai/unsloth/pull/10506
* Studio: avoid hidden PowerShell arguments in CLI setup and updates by @oobabooga in https://github.com/unslothai/unsloth/pull/10560
* Guard main against drifting out of its own formatting hook by @danielhanchen in https://github.com/unslothai/unsloth/pull/10603
* Ask for the NPP runtime by the name that has wheels on CUDA 13 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10587
* Studio: report a Claude reply cut off by the context window as truncated by @NilayYadav in https://github.com/unslothai/unsloth/pull/10557
* Share builders and setup helpers in the studio backend tests, part 1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/10497
* [Fix] Studio: show "Using tool" for active tool calls by @debojyoti-dev in https://github.com/unslothai/unsloth/pull/10548
* Studio: give the conversation archive's order a final tiebreaker by @danielhanchen in https://github.com/unslothai/unsloth/pull/10572
* Studio: apply EXIF orientation to uploaded images by @NilayYadav in https://github.com/unslothai/unsloth/pull/10553
* Studio: capture the diffusion denoiser step into a CUDA graph by @danielhanchen in https://github.com/unslothai/unsloth/pull/10537
* Studio: remove the background staged desktop update and return to the classic update by @danielhanchen in https://github.com/unslothai/unsloth/pull/10604
* Studio: try every address a site resolves to when fetching a page by @NilayYadav in https://github.com/unslothai/unsloth/pull/10319
* Windows installer: emit the native path helper instead of compiling it by @danielhanchen in https://github.com/unslothai/unsloth/pull/10540

## New Contributors
* @KrishPatel1905 made their first contribution in https://github.com/unslothai/unsloth/pull/8924
* @amanyagami made their first contribution in https://github.com/unslothai/unsloth/pull/9912
* @harshaygadekar made their first contribution in https://github.com/unslothai/unsloth/pull/10142
* @denys-bushuliak made their first contribution in https://github.com/unslothai/unsloth/pull/10362
* @ianderse made their first contribution in https://github.com/unslothai/unsloth/pull/10388
* @HughScott2002 made their first contribution in https://github.com/unslothai/unsloth/pull/10463
* @xerrni made their first contribution in https://github.com/unslothai/unsloth/pull/5408
* @Shivamjohri247 made their first contribution in https://github.com/unslothai/unsloth/pull/4965
* @TimPietrusky made their first contribution in https://github.com/unslothai/unsloth/pull/4443
* @zohairshafi made their first contribution in https://github.com/unslothai/unsloth/pull/5103
* @claxman made their first contribution in https://github.com/unslothai/unsloth/pull/10445
* @imherro made their first contribution in https://github.com/unslothai/unsloth/pull/9449
* @AronRubin made their first contribution in https://github.com/unslothai/unsloth/pull/5336
* @RudraBhaskar9439 made their first contribution in https://github.com/unslothai/unsloth/pull/10506
* @debojyoti-dev made their first contribution in https://github.com/unslothai/unsloth/pull/10548

**Full Changelog**: https://github.com/unslothai/unsloth/compare/v0.1.806-beta...v0.1.808-beta