v0.1.900-beta

We're adding support for decision models, a unified Library for docs and media, document viewer, many Apple Silicon improvements, creation of Skills, and ~4.5× faster image and video generation.
* Run and serve **Decision Models** like Laya (open-source Jev) locally
* **Skills Editor** to create, edit, and delete Skills directly in Desktop
* **ModelScope** model downloading is now here for users who can't use HF
* **Library + Document Viewer** for PDF, Word, Excel, PowerPoint, and chats.
* You can now attach files before loading a model
* **Apple Silicon Improvements** including batched serving, structured outputs, and TurboQuant KV cache.
* Configure how much of a model each GPU receives.
* **Faster Image + Video Generation** with ~4.5x faster LTX-2.3 clips and 1.7–6.3x faster VAE decoding.
* New food-inspired **Themes**. Check them out!

## Library + Viewer

* View **PDF, Word, Excel, and PowerPoint** files directly in Unsloth, with links back to their source.
* Manage chats, images, videos, everything in our new **Library** tab.
* Improved attachment cards and support for reading more file formats.
* **Create your own sidebar sections** and drag sections to reorder them.

## Laya + decision models

Run **Laya decision models locally** to answer yes/no, multiple-choice, and scoring questions with probabilities.
* Enable the Decision API from **Settings > API**, with model selection and CPU or GPU controls.
* Supports the **TypeSafe SDK** through a Jev-compatible `/v1/systemone` endpoint.
* Runs natively on **MLX for Apple Silicon** when GPU is selected. [[Details](https://github.com/unslothai/unsloth/pull/11603)](https://github.com/unslothai/unsloth/pull/11603)

## Faster image + video generation

* **LTX-2.3 clips are ~4.5x faster** with distilled sampling, compile fixes, and hosted FP8 weights.
* Image and video VAE optimizations deliver **1.7–6.3x faster decoding**, with total generation speedups depending on the workflow.
* MiniMax-H3's first render is **up to a minute faster**, with **25–29 GiB lower peak memory** on A100, B200, and RTX PRO 6000.

## Apple Silicon + MLX

* **Batched serving** allows multiple replies to decode at once.
* Added **grammar-constrained structured outputs** through `response_format`.
* Added **TurboQuant KV cache** and KV cache quantization for sliding-window models such as Gemma 4.
* Improved memory estimates and automatic context sizing based on available memory.

## Skills + chat improvements

* **Create, edit, and delete Skills** from the Desktop Skills menu.
* Improved conversation recall across repeated auto compactions.
* Long-running tool calls continue even when they produce no output, with better protection against excessive output consuming memory.
* Auto-scroll or manual scroll your conversations
* Use **ModelScope** instead of Hugging Face for model downloads
* You can now attach as many files before loading a model, if a model doesn't accept an input, the prompt will just ignore it

## Training + inference

* Faster **block-FP8 LoRA training** and support for **LoRA on compressed-tensors W8A8 checkpoints**.
* Load **Voxtral and Qwen2-Audio** through `FastModel`.
* Export saved checkpoints from crashed or cancelled training runs.
* Important compatibility fixes for Gemma, newer TRL trainers, and FP8 on RTX 5090 and RTX PRO 6000.

## What's Changed
* Wait for the model load before reading the unsloth run banner by @danielhanchen in https://github.com/unslothai/unsloth/pull/11679
* Bump install.sh / install.ps1 pin to unsloth>=2026.9.11 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11689
* Revert "Studio: steady the Images and Video loading spinners, and tidy the progress card" by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11691
* Follow #11635's Diffusers prefetch and #11660's responsive classes in the tests by @danielhanchen in https://github.com/unslothai/unsloth/pull/11680
* Give the ROCm bf16 chain test every name _gpu_init imports from device_type by @danielhanchen in https://github.com/unslothai/unsloth/pull/11688
* Read the sidebar pin's width only where matchMedia exists, and expect the update card's scaled width by @danielhanchen in https://github.com/unslothai/unsloth/pull/11700
* Keep a test's fake pid identity out of the cached owner identity by @danielhanchen in https://github.com/unslothai/unsloth/pull/11708
* Expect the unsloth Z-Image-Turbo mirror in the image download queue drive by @danielhanchen in https://github.com/unslothai/unsloth/pull/11711
* Classify uv's new --output-format as a value flag in the pip shim by @danielhanchen in https://github.com/unslothai/unsloth/pull/11715
* Studio: do not roll back a model switch whose load got no answer by @danielhanchen in https://github.com/unslothai/unsloth/pull/11729
* studio: add responses api selection for custom providers by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11354
* Count the stripper's whole-buffer work instead of timing it by @danielhanchen in https://github.com/unslothai/unsloth/pull/11735
* Studio: keep tracking a new chat's upload when its id commits late by @danielhanchen in https://github.com/unslothai/unsloth/pull/11738
* studiobench: compare a code fence one arm scrolled past on its text by @danielhanchen in https://github.com/unslothai/unsloth/pull/11741
* Anchor the GRPO MoE aux-loss fail-fast on the assignment, not its expression by @danielhanchen in https://github.com/unslothai/unsloth/pull/11747
* Chat UI driver: wait for the Recents thread to load, and let the check fail by @danielhanchen in https://github.com/unslothai/unsloth/pull/11749
* Clean-machine trace leg: let git fetch the pinned git+ requirements, nothing else by @danielhanchen in https://github.com/unslothai/unsloth/pull/11751
* Studio: add Search Hub to the Images, Video and Audio model pickers by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11756
* Studio: keep the model name and quant whole in the Images and Video header by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11757
* Studio: drag to reorder, add to project and quick download for the image and video galleries by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11758
* Studio: make the Images, Video and Audio settings rail resizable by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11760
* Studio installer: do not run Apple's git shim to ask whether git works by @danielhanchen in https://github.com/unslothai/unsloth/pull/11759
* Data settings check: pin the legacy store outcome in the failed-delete check by @danielhanchen in https://github.com/unslothai/unsloth/pull/11763
* Update idempotency: count whisper.cpp only when the install has one by @danielhanchen in https://github.com/unslothai/unsloth/pull/11762
* Account matrix: cover the gallery move and add-to-project routes by @danielhanchen in https://github.com/unslothai/unsloth/pull/11767
* Studio: stop clipping the model description's descenders; follow the resizable rail in the layout contracts by @danielhanchen in https://github.com/unslothai/unsloth/pull/11769
* Studio: follow-ups for the media gallery, hub link and rail divider by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11765
* studiobench: do not fail UI parity on repetitions that swap two renderings by @danielhanchen in https://github.com/unslothai/unsloth/pull/11772
* MCP HTTP integration test: let the server bind its own port by @danielhanchen in https://github.com/unslothai/unsloth/pull/11777
* Export full fine-tunes in 16-bit from the CLI by default by @NilayYadav in https://github.com/unslothai/unsloth/pull/11716
* Studio: end Alpaca training samples with the end token by @NilayYadav in https://github.com/unslothai/unsloth/pull/11719
* Studio: keep Qwen-Image-2.1 int8 / fp8 compiling on torch 2.12 (CantSplit) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11677
* Studio: let an explicit FBCache request engage on Qwen-Image-2.1 and other prefix-KV models by @danielhanchen in https://github.com/unslothai/unsloth/pull/11713
* Studio: make the HunyuanImage-2.1 denoiser step capturable as a CUDA graph by @danielhanchen in https://github.com/unslothai/unsloth/pull/11753
* Studio: pin, reorder and add to project for Audio history by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11774
* Studio: Show token usage and cache stats on connected provider chats by @NilayYadav in https://github.com/unslothai/unsloth/pull/11717
* fix(studio): support preserve thinking for llama.cpp connections by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11706
* Studio: stop the max speed tier recompiling on every new prompt length by @danielhanchen in https://github.com/unslothai/unsloth/pull/11731
* Layout contract: tie the Train rail clamp to its scroller's padding by @danielhanchen in https://github.com/unslothai/unsloth/pull/11773
* Studio: fix broken Chinese, Japanese and emoji text in gpt-oss replies by @NilayYadav in https://github.com/unslothai/unsloth/pull/11720
* Studio: Keep Codex reasoning between tool calls on /v1/responses by @NilayYadav in https://github.com/unslothai/unsloth/pull/11726
* Studio: Make `unsloth start --reasoning on/off` take effect by @NilayYadav in https://github.com/unslothai/unsloth/pull/11718
* Studio: example prompt per image workflow, remember the last prompt by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11775
* Studio: dock the floating Live monitor beside Run settings by @wasimysaid in https://github.com/unslothai/unsloth/pull/11699
* Account matrix: cover the audio gallery move and add-to-project routes by @danielhanchen in https://github.com/unslothai/unsloth/pull/11789
* Studio: show the Custom size fields under Output size in unified Edit by @oobabooga in https://github.com/unslothai/unsloth/pull/11690
* Fix false ONNX rejection for models with native weights by @wasimysaid in https://github.com/unslothai/unsloth/pull/11695
* Fix GGUF vision capability selection by @wasimysaid in https://github.com/unslothai/unsloth/pull/11696
* Baseline the two huggingface_hub 1.33.0 / 2.0.0 findings after review by @danielhanchen in https://github.com/unslothai/unsloth/pull/11811
* Replace an in-flight chat model load with the latest pick by @wasimysaid in https://github.com/unslothai/unsloth/pull/11697
* Prebuilt installers: retry a download the server drops by @danielhanchen in https://github.com/unslothai/unsloth/pull/11816
* Load-replacement contract: read the rollback guard, not its one-line spelling by @danielhanchen in https://github.com/unslothai/unsloth/pull/11817
* Unsloth Studio (AMD/ROCm): don't turn on cudnn.benchmark for image and video generation by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11732
* Padding-free gate tests: state that UNSLOTH_RETURN_LOGITS is unset by @danielhanchen in https://github.com/unslothai/unsloth/pull/11863
* Eval step: restore UNSLOTH_RETURN_LOGITS even when evaluation raises by @danielhanchen in https://github.com/unslothai/unsloth/pull/11865
* Studio update launcher tests: give every test a private STUDIO_HOME by @danielhanchen in https://github.com/unslothai/unsloth/pull/11869
* Studio: a sandbox read waits out a legacy move's staging window by @danielhanchen in https://github.com/unslothai/unsloth/pull/11876
* Agent guides CI: pass optional installer flags only while the installer takes them by @danielhanchen in https://github.com/unslothai/unsloth/pull/11878
* fix(studio): keep Stop generating visible while queuing by @Biotrioo in https://github.com/unslothai/unsloth/pull/9123
* Studio: give the Windows ROCm torchao stub a version transformers 5 can parse by @danielhanchen in https://github.com/unslothai/unsloth/pull/11640
* Give vLLM a valid top_k when fast_inference enables it in the GRPO trainer by @danielhanchen in https://github.com/unslothai/unsloth/pull/11675
* Studio: stop unsloth start announcing a switch when --model names the file already loaded by @oobabooga in https://github.com/unslothai/unsloth/pull/11873
* Studio: stop holding a passthrough response 5 s when a watcher swallows its cancel by @oobabooga in https://github.com/unslothai/unsloth/pull/11859
* Studio: say the installer script is missing instead of showing the PowerShell logo by @oobabooga in https://github.com/unslothai/unsloth/pull/11862
* Studio: stop rescanning every model folder on each request that names a model not on disk by @oobabooga in https://github.com/unslothai/unsloth/pull/11872
* studiobench: image_upload closes the menu it opened when it gives up by @danielhanchen in https://github.com/unslothai/unsloth/pull/11892
* Parallel-isolation guard: exempt the resolver's back-dated staleness precondition by @danielhanchen in https://github.com/unslothai/unsloth/pull/11894
* Studio: stop opening a saved chat from re-running a Max Tokens reply by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11875
* Studio: use the browser's timezone for today's date in chat by @NilayYadav in https://github.com/unslothai/unsloth/pull/11852
* SentenceTransformer: preserve masks for patched Gemma3 attention by @Etherll in https://github.com/unslothai/unsloth/pull/11866
* Studio: initialize new chats before attaching documents by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11838
* Studio: stop inflating chat images by re-encoding every one to PNG by @oobabooga in https://github.com/unslothai/unsloth/pull/11889
* Studio: fix document search for embedding models other than the default by @NilayYadav in https://github.com/unslothai/unsloth/pull/11853
* Studio: train audio datasets on the columns the dataset check found by @NilayYadav in https://github.com/unslothai/unsloth/pull/11850
* Studio: follow the JSON format a client asks for on /v1/messages by @NilayYadav in https://github.com/unslothai/unsloth/pull/11854
* Studio: let a temporary chat be saved to history by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11901
* Studio Hub: show a running download as Downloading, not as a paused partial by @danielhanchen in https://github.com/unslothai/unsloth/pull/11797
* Accept the transformers 5.0 ignore_keys argument in validate_rope by @danielhanchen in https://github.com/unslothai/unsloth/pull/11560
* Studio: verify the Diffusers main zip against a pinned SHA-256 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11908
* Studio: restyle the save temporary chat popup as a standard dialog by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11914
* Fix text_only 4-bit load and generate for Gemma-4 and other VLM text configs by @danielhanchen in https://github.com/unslothai/unsloth/pull/11684
* Studio: add File and View menu items to the macOS desktop app by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11902
* Studio: keep the model's own chat template when the Unsloth one can't render the rows by @oobabooga in https://github.com/unslothai/unsloth/pull/11487
* Split the flash-attn prebuilt wheel build across parallel ccache jobs by @oobabooga in https://github.com/unslothai/unsloth/pull/11812
* Studio: find older chats by title in chat search by @NilayYadav in https://github.com/unslothai/unsloth/pull/11296
* Studio: fix login failures during slow startup by @NilayYadav in https://github.com/unslothai/unsloth/pull/11289
* Studio: build Qwen-Image-2.1's token layout once per render instead of every step by @danielhanchen in https://github.com/unslothai/unsloth/pull/11887
* Load a VLM through its native image-text class when the repo's auto_map class is untrusted by @danielhanchen in https://github.com/unslothai/unsloth/pull/11613
* Unsloth Studio (AMD): floor torch at 2.11 on the gfx103X-all and gfx110X-all families too by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11834
* Studio: read DOCX content controls, tracked insertions and text boxes by @L4XB in https://github.com/unslothai/unsloth/pull/11803
* Studio: start settings labels with the setting, not "Show" by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11924
* Studio: align the model selector label and truncate long project names by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11917
* Studio: style the chat scrollbar like Run settings by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11925
* Unsloth Studio (AMD/ROCm): warn about, and refuse, a GPU the installed PyTorch has no kernels for by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11571
* Studio: video auto precision keeps a resident bf16 DiT by @danielhanchen in https://github.com/unslothai/unsloth/pull/11831
* Studio: show Theme first in Settings > Appearance by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11921
* Stop the Starling, Yi-chat and LFM2 templates leaking whitespace by @Abhishek-B-R in https://github.com/unslothai/unsloth/pull/11779
* Studio: clicking a gallery item keeps the typed prompt by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11930
* Studio: move Read aloud and Edit response into the More menu by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11920
* Studio: make the composer the same width as the chat column by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11926
* UI scale contract: count the save-temporary-chat button among the chat header's 30px controls by @danielhanchen in https://github.com/unslothai/unsloth/pull/11932
* studio: allow local imatrix files for gguf export by @mahiatlinux in https://github.com/unslothai/unsloth/pull/11350
* Studio: guard export operations when the Hub is unreachable by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11466
* Studio: keep a Codex chat working after a tool returns an image on a text-only model by @NilayYadav in https://github.com/unslothai/unsloth/pull/11476
* Studio: show example prompts as placeholder hints by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11931
* Studio: keep a chat's attached files when you fork it by @NilayYadav in https://github.com/unslothai/unsloth/pull/11295
* Studio: add a multiline send shortcut and spell out what each one does by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11927
* Studio: stream durable chat runs at display frame rate by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11900
* Embed server tests: intercept only the server's own Popen by @danielhanchen in https://github.com/unslothai/unsloth/pull/11934
* Studio: build the ConvRot rotation from its definition by @danielhanchen in https://github.com/unslothai/unsloth/pull/11807
* Studio: keep all text when adding Word files to a knowledge base by @NilayYadav in https://github.com/unslothai/unsloth/pull/11725
* Studio: show when a response was written, in its More menu by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11928
* Studio: load the hosted INT8 pre-quant checkpoints on torchao 0.18 and later by @danielhanchen in https://github.com/unslothai/unsloth/pull/11884
* Studio: show context checkpoints apart from the KV cache in the memory estimate by @oobabooga in https://github.com/unslothai/unsloth/pull/11581
* Studio: drag to reorder pinned models in the model picker by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11941
* Cast fp16 leftovers to the requested dtype after a text_only pre-quantized load by @danielhanchen in https://github.com/unslothai/unsloth/pull/11692
* Studio: translate the inline Read aloud and Edit response settings by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11933
* Accept block_sequence_ids in chunked causal masks on transformers 5.17 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11693
* Studio: convert decoded video frames to uint8 on the GPU before the mp4 encode by @danielhanchen in https://github.com/unslothai/unsloth/pull/11879
* Studio: fix pinned rows in the model picker by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11943
* Studio: stop rewriting the compile-cache bundle on every warm start, and bound its disk use by @danielhanchen in https://github.com/unslothai/unsloth/pull/11874
* Studio: list the Images workflows in the phone sidebar again by @oobabooga in https://github.com/unslothai/unsloth/pull/11936
* Studio: reuse unchanged files from an older snapshot instead of re-downloading them by @danielhanchen in https://github.com/unslothai/unsloth/pull/11796
* Studio: draw one drop line per gap when dragging sidebar rows by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11942
* Studio: show the full URL while a web fetch waits for approval by @wasimysaid in https://github.com/unslothai/unsloth/pull/11694
* Rebuild byte-level tokenizers that transformers v5 loads as LlamaTokenizer by @danielhanchen in https://github.com/unslothai/unsloth/pull/11686
* Plan a device map for text_only loads of vision-language models by @danielhanchen in https://github.com/unslothai/unsloth/pull/11584
* Installer: stop picking cu126 when a slow NVIDIA driver times out CUDA detection by @danielhanchen in https://github.com/unslothai/unsloth/pull/11916
* Studio: skip xFormers when its torch requirement is unmet by @oobabooga in https://github.com/unslothai/unsloth/pull/11847
* Studio: add HTTP recording fallback to Dictate by @Etherll in https://github.com/unslothai/unsloth/pull/11075
* Studio: propagate the Ollama CUDA runtime to llama-server (replacement for #7563) by @wasimysaid in https://github.com/unslothai/unsloth/pull/11666
* Prevent setup-size flash during desktop startup by @wasimysaid in https://github.com/unslothai/unsloth/pull/11910
* Studio: return freed host memory to the OS after diffusion and video unload by @danielhanchen in https://github.com/unslothai/unsloth/pull/11795
* Walk every sub-config when deciding whether a config carries remote code by @danielhanchen in https://github.com/unslothai/unsloth/pull/11555
* Studio: fix vision training with evaluation on when there is no eval split by @NilayYadav in https://github.com/unslothai/unsloth/pull/11851
* Studio: stop a compiled Qwen-Image-2.1 render holding 2 GiB of prefix K/V it does not need by @danielhanchen in https://github.com/unslothai/unsloth/pull/11882
* Dequantize FP8 weights left raw by text_only and offloaded loads by @danielhanchen in https://github.com/unslothai/unsloth/pull/11841
* Studio: video status reports CUDA graphs off when they never engage by @danielhanchen in https://github.com/unslothai/unsloth/pull/11886
* fix(studio): show compaction notices for tool-loop checkpoints by @Imagineer99 in https://github.com/unslothai/unsloth/pull/11702
* Pin the ARM64 Arrow overlay to a commit before its wheels are signed by @danielhanchen in https://github.com/unslothai/unsloth/pull/11911
* Load 4.x remote code and config-only remote code on transformers 5 (Trinity-Large, MiniMax-M3) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11658
* Studio: serve the Jev API locally with Laya by @NilayYadav in https://github.com/unslothai/unsloth/pull/11603
* Studio: opt-in NVENC for the video mp4 export by @danielhanchen in https://github.com/unslothai/unsloth/pull/11881
* Studio: compile the diffusion denoiser on fp16 GPUs when a speed tier is picked (T4 and other pre-Ampere cards) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11899
* Studio: whole-model offload points weights back at their host tensors instead of copying them by @danielhanchen in https://github.com/unslothai/unsloth/pull/11764
* Studio: pin the kept offload weights on first onload when host RAM allows by @danielhanchen in https://github.com/unslothai/unsloth/pull/11766
* fix(studio): count rendered GGUF prompts before shared-KV admission by @Imagineer99 in https://github.com/unslothai/unsloth/pull/10794
* Point -bf16 4-bit requests at quantization_config, and warn when a bitsandbytes load quantized nothing by @danielhanchen in https://github.com/unslothai/unsloth/pull/11586
* Studio: HunyuanVideo-1.5 padded-text trim on the max speed tier by @danielhanchen in https://github.com/unslothai/unsloth/pull/11824
* Studio: release VRAM when an image model is unloaded mid-render by @danielhanchen in https://github.com/unslothai/unsloth/pull/11790
* Studio: faster MiniMax-H3 video VAE encode and decode by @danielhanchen in https://github.com/unslothai/unsloth/pull/11801
* fix(grpo): use the evaluated model's output head for log-probs by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/11877
* Studio: auto step cache (FBCache) only on the max speed tier by @danielhanchen in https://github.com/unslothai/unsloth/pull/11791
* Studio: cache text-encoder outputs so repeat prompts skip the encoder by @danielhanchen in https://github.com/unslothai/unsloth/pull/11844
* fix(studio): classify MLX requests from the applied chat template override by @Lyxot in https://github.com/unslothai/unsloth/pull/11903
* Unsloth Studio (AMD): allow INT8 / FP8 image and video precision without torchao by @danielhanchen in https://github.com/unslothai/unsloth/pull/11631
* Studio: count the hosted text encoder and keep the GGUF denoiser resident when only the encoder does not fit by @danielhanchen in https://github.com/unslothai/unsloth/pull/11802
* Studio: run an explicit int8 under offload on NVIDIA as torchao-free W8A8 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11712
* Studio: apply remembered settings for MLX and safetensors models to API loads by @Lyxot in https://github.com/unslothai/unsloth/pull/11904
* Studio: keep bf16 for auto precision on image families that cannot compile by @danielhanchen in https://github.com/unslothai/unsloth/pull/11819
* Studio: tile the VAE instead of refusing oversized upscales, and in-app remedies for image refusals by @danielhanchen in https://github.com/unslothai/unsloth/pull/11798
* Load and train remote-code multimodal wrappers: Phi-4-reasoning-vision in 4-bit and Nemotron-3-Nano-Omni by @danielhanchen in https://github.com/unslothai/unsloth/pull/11526
* Finish a 16bit load of a static per-tensor fp8 checkpoint (Mistral-Small-4) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11531
* Studio: static step skip for image models that keeps CUDA graphs by @danielhanchen in https://github.com/unslothai/unsloth/pull/11737
* Studio: static step skip for video generation by @danielhanchen in https://github.com/unslothai/unsloth/pull/11748
* Studio: Qwen-Image-2.1 RoPE in real arithmetic inside the compiled blocks by @danielhanchen in https://github.com/unslothai/unsloth/pull/11888
* Studio: Library page for files, media and fine-tunes by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11770
* Resolve remote-code model classes, reach per-expert submodules with LoRA, run the root dispatch hook on the embedding's device (Nemotron-Labs-Teacher) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11543
* Studio: Library settings, storage and sortable list columns by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11771
* Studio: open Images and Video previews in the Library viewer by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11776
* Format the two files #11526 left off the formatter's fixed point by @danielhanchen in https://github.com/unslothai/unsloth/pull/11955
* Load and fine-tune LongCat-Flash-Lite-Sparse on transformers' longcat_flash by @danielhanchen in https://github.com/unslothai/unsloth/pull/11620
* Hand a composition with no forward of its own to its thinker (Qwen3-Omni) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11523
* fix(studio): retain GGUF quants across cache-folder switches by @wasimysaid in https://github.com/unslothai/unsloth/pull/11938
* Keep the training state intact across a standalone evaluate() / predict() by @danielhanchen in https://github.com/unslothai/unsloth/pull/11860
* fix(studio): recover from a dead Metal GPU queue instead of failing every later request by @Lyxot in https://github.com/unslothai/unsloth/pull/11383
* Pin the compiled mask wrapper only when it is compiled by @danielhanchen in https://github.com/unslothai/unsloth/pull/11957
* Load NVIDIA ModelOpt FP8 checkpoints through the transformers fp8 quantizer by @danielhanchen in https://github.com/unslothai/unsloth/pull/11592
* Studio: run Qwen-Image-2.1 int8/fp8 on 24 GB cards with the transformer resident and the text encoder streamed by @danielhanchen in https://github.com/unslothai/unsloth/pull/11883
* Divide by gradient accumulation for forwards that take **kwargs but return a mean loss by @danielhanchen in https://github.com/unslothai/unsloth/pull/11898
* Load 4.x-era configs that transformers 5 strict validation rejects (Llama 4 attn_temperature_tuning) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11836
* Studio: minimal OS sandbox for Python and Terminal tools on Linux and macOS by @oobabooga in https://github.com/unslothai/unsloth/pull/11209
* Studio: add Windows MXC Preview sandboxing by @Etherll in https://github.com/unslothai/unsloth/pull/11390
* Studio: whole-model NVFP4 for the video families with hosted pre-quantized denoisers by @danielhanchen in https://github.com/unslothai/unsloth/pull/10729
* Studio: per-layer NVFP4 image policies, flashinfer FP4 backend and a gated auto row by @danielhanchen in https://github.com/unslothai/unsloth/pull/10730
* Studio: NVFP4 flashinfer backend kernel items (device guard, persistent barrier, bias path, cached dispatch) by @danielhanchen in https://github.com/unslothai/unsloth/pull/10731
* Studio: keep inline HTML text in its line when parsing an upload by @L4XB in https://github.com/unslothai/unsloth/pull/11818
* Studio: read text, Markdown and HTML uploads in the encoding they were written in by @L4XB in https://github.com/unslothai/unsloth/pull/11896
* Fix PEFT base export targets silently exporting the base model by @tweekli in https://github.com/unslothai/unsloth/pull/11781
* Studio: install flashinfer on demand for NVFP4 without moving torch by @danielhanchen in https://github.com/unslothai/unsloth/pull/11730
* Studio: steady pinned model drop line and pill-shaped project drop highlight by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11986
* Studio: compile the VAE decode for DiT families, from the NVFP4 time budget pass by @danielhanchen in https://github.com/unslothai/unsloth/pull/10889
* Studio: Library grid Sort menu, tidier toolbar menus and card icons by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11978
* Studio: cap context checkpoints by host RAM for sliding-window and SSM models by @Lyxot in https://github.com/unslothai/unsloth/pull/11918
* Studio: sort projects in the sidebar and show pinned chats once by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11988
* Studio tests: read UI labels from the en catalog, and follow the response-details action into MessageMenuTime by @danielhanchen in https://github.com/unslothai/unsloth/pull/11949
* Studio: open model row tooltips from the name only by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11985
* Studio: pin fine-tuned models in the model picker by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11984
* CI: only cancel superseded pull request runs, never dispatches or schedules by @danielhanchen in https://github.com/unslothai/unsloth/pull/11977
* CI: fix path filters that miss dependencies, and narrow three that are too broad by @danielhanchen in https://github.com/unslothai/unsloth/pull/11989
* Studio Playwright helpers: named per-step budgets, fail-fast step report, condition waits by @danielhanchen in https://github.com/unslothai/unsloth/pull/11979
* Studio Playwright chat_ui: condition waits, per-step budgets, 25 s less in the theme step by @danielhanchen in https://github.com/unslothai/unsloth/pull/11983
* Studio Playwright: condition waits and per-step budgets in extra_ui and update_banner_layout by @danielhanchen in https://github.com/unslothai/unsloth/pull/11990
* Studio Playwright: condition waits and per-step budgets in model_config and memory_estimate by @danielhanchen in https://github.com/unslothai/unsloth/pull/11987
* Studio Playwright: condition waits and per-step budgets in loaded_models_indicator and ui_font_scale by @danielhanchen in https://github.com/unslothai/unsloth/pull/11992
* Studio Playwright: condition waits and step budgets in thread_scoped_settings, mcp_arguments, chat_width by @danielhanchen in https://github.com/unslothai/unsloth/pull/11991
* Export: add save_pretrained_openvino and push_to_hub_openvino support by @goodmai in https://github.com/unslothai/unsloth/pull/11907
* Studio: line sidebar section headers up with row content by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/11996
* CLI: keep a reply's trailing '<' or '<th' once the stream has ended by @breken-ai in https://github.com/unslothai/unsloth/pull/11893
* Studio: drop Move up / Move down from sidebar row menus by @danielhanchen in https://github.com/unslothai/unsloth/pull/12000
* Studio: Help menu items and a Go menu for desktop Help search by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12002
* Studio: size Library gallery items without resolving each file by @danielhanchen in https://github.com/unslothai/unsloth/pull/12003
* Fix left-padded Online DPO scoring during training by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/11885
* feat(studio): ModelScope as a model source and a custom Hugging Face endpoint by @Lyxot in https://github.com/unslothai/unsloth/pull/11761
* Keep the Kaggle GPU harness tests independent of the caller's CUDA_VISIBLE_DEVICES by @danielhanchen in https://github.com/unslothai/unsloth/pull/12010
* OpenVINO export: trust remote code only for a remote-code model, and decode the bounds probe as UTF-8 by @danielhanchen in https://github.com/unslothai/unsloth/pull/12012
* Skip test_wait_for_settled when playwright.sync_api is only a stub by @danielhanchen in https://github.com/unslothai/unsloth/pull/12013
* feat(install): fall back to CERNET and npmmirror when package hosts are blocked or slow by @Lyxot in https://github.com/unslothai/unsloth/pull/11786
* Skip Unsloth's generated compile cache in the exec-literal lint by @danielhanchen in https://github.com/unslothai/unsloth/pull/12014
* tests: make the ROCm install suite pass on Windows, macOS and arm64 runners by @danielhanchen in https://github.com/unslothai/unsloth/pull/12005
* Studio setup: read the amd-smi index-space line without head -n 1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/12004
* Add longcat_flash_lsa to the fused-MoE conversion snapshot by @danielhanchen in https://github.com/unslothai/unsloth/pull/12018
* Record only the test thread's sleeps as Deep Research retry backoff by @danielhanchen in https://github.com/unslothai/unsloth/pull/12019
* Studio CLI: stop `unsloth run` re-exec'ing itself forever when the Studio venv is a symlink by @danielhanchen in https://github.com/unslothai/unsloth/pull/11788
* Studio: do not reapply ROCR_VISIBLE_DEVICES when picking the AMD card in setup.sh and the llama.cpp prebuilt probe by @danielhanchen in https://github.com/unslothai/unsloth/pull/11965
* Studio: run image and video denoises on one render thread so cuDNN caches are reused by @danielhanchen in https://github.com/unslothai/unsloth/pull/11843
* Studio: return an error when embeddings dimensions can't be honored by @NilayYadav in https://github.com/unslothai/unsloth/pull/11968
* Studio: keep a 0 label and blank a NaN cell when mapping columns to chat roles by @breken-ai in https://github.com/unslothai/unsloth/pull/11895
* Studio: support web_search_20260209 on /v1/messages by @NilayYadav in https://github.com/unslothai/unsloth/pull/11963
* Studio: keep a working GPU when one probe fails, and name the card behind no_gpu by @danielhanchen in https://github.com/unslothai/unsloth/pull/11944
* Studio: download only one copy of a GGUF quant by @NilayYadav in https://github.com/unslothai/unsloth/pull/11966
* Give the real-host NVIDIA probe test a budget that a busy driver can meet by @danielhanchen in https://github.com/unslothai/unsloth/pull/12026
* Unsloth Studio installer (AMD/Linux): send RDNA 4 cards (RX 9000, R9700) to AMD's gfx120X wheels by @danielhanchen in https://github.com/unslothai/unsloth/pull/11935
* Studio: keep a CSV seed's values as written when dropping its index column by @L4XB in https://github.com/unslothai/unsloth/pull/11945
* Studio media viewer: Zoom to fit at the bottom of the scale menu by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12028
* Studio: shadow the dark mode composer so it stands off the chat by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12023
* Studio: name the Git Bash MXC incompatibility and stop re-probing it by @danielhanchen in https://github.com/unslothai/unsloth/pull/12021
* Keep more VRAM headroom on Windows CUDA, and say when a hand-set context does not fit by @danielhanchen in https://github.com/unslothai/unsloth/pull/11368
* Unsloth Studio (AMD): keep export off a GPU PyTorch has no kernels for, like the iGPU by @oobabooga in https://github.com/unslothai/unsloth/pull/11946
* Read the hot-path I/O cost at its steady minimum across repeats by @danielhanchen in https://github.com/unslothai/unsloth/pull/12029
* Unsloth Studio: a companion fetch with no denoiser no longer blocks deleting its base by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11828
* Studio: stop Python tool network calls from skipping the host allowlist by @oobabooga in https://github.com/unslothai/unsloth/pull/11172
* Studio: center collapsed sidebar icons and tighten the rail by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12031
* Unsloth Studio: list a GGUF with no header metadata under On Device on the Images page by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11830
* Keep the conversion backfill's donor stub off the transformers package by @danielhanchen in https://github.com/unslothai/unsloth/pull/12034
* Unsloth Desktop: show Stopping… after Stop so a pending cancel doesn't look ignored by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11976
* Studio: stop edit_file writing a compacted-argument placeholder into files by @oobabooga in https://github.com/unslothai/unsloth/pull/11950
* Studio: decode Wan video in fp16 with channels_last_3d convs by @danielhanchen in https://github.com/unslothai/unsloth/pull/11999
* Unsloth Studio: refuse picking a hosted FP8/INT8 checkpoint repo as a pipeline before it downloads by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11829
* Studio: stop the second prompt length recompiling Qwen-Image on the max tier and with int8 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11842
* Studio: size the image memory plan at the dtype the pipeline loads in (SDXL resident on 24 GB) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11922
* Studio: stop MiniMax-H3 recompiling on the second caption and the first i2v by @danielhanchen in https://github.com/unslothai/unsloth/pull/11880
* Unsloth Studio / Desktop: log what an image load and each generation actually resolved to by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11994
* Unsloth Studio: stop offering "Continue" on a base repo that only holds a GGUF's text encoder and VAE by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11644
* Apply SFTConfig.router_aux_loss_coef to MoE models that cache it at init by @danielhanchen in https://github.com/unslothai/unsloth/pull/12006
* Load only the language model for text_only on repo-code composites by @danielhanchen in https://github.com/unslothai/unsloth/pull/11861
* Put back the unsloth_zoo modules the device map opt-in tests stub by @danielhanchen in https://github.com/unslothai/unsloth/pull/12038
* Studio: pass tool_result is_error through to the model on /v1/messages by @NilayYadav in https://github.com/unslothai/unsloth/pull/11962
* SentenceTransformer: add opt-in FP32 merged-pair ranking loss by @Etherll in https://github.com/unslothai/unsloth/pull/11867
* Studio: opt-in Hadamard rotation for Qwen-Image-2.1's int8 transformer so it matches bf16 by @danielhanchen in https://github.com/unslothai/unsloth/pull/11835
* Grouped-linear LoRA for DeepSeek-V4, remote-code shims for Step-3.7, and a real message for Mistral-format checkpoints by @danielhanchen in https://github.com/unslothai/unsloth/pull/11528
* Studio: fall back to native when SageAttention does not run on this GPU by @danielhanchen in https://github.com/unslothai/unsloth/pull/11997
* Keep flash attention off sub-models that do not support it (LFM2.5-VL SigLIP2 tower) by @danielhanchen in https://github.com/unslothai/unsloth/pull/11959
* Studio: stop trading the denoiser's quantisation away to pay for offload by @danielhanchen in https://github.com/unslothai/unsloth/pull/11558
* Unsloth Studio: report the VAE decode on the image progress bar by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11740
* Translate the Library toolbar Sort menu in every locale by @danielhanchen in https://github.com/unslothai/unsloth/pull/12045
* Read the partial safetensors delete-menu guard by operator, not verbatim by @danielhanchen in https://github.com/unslothai/unsloth/pull/12042
* Resolve the macOS app menu's chords as a Mac in its test by @danielhanchen in https://github.com/unslothai/unsloth/pull/12046
* Advertise NVFP4 diffusion on the text encoder driver's mocked host by @danielhanchen in https://github.com/unslothai/unsloth/pull/12047
* Keep peft's is_torchao_available cache API through the stale-torchao patch by @danielhanchen in https://github.com/unslothai/unsloth/pull/12049
* Run the formatter fixed-point guard's batches side by side by @danielhanchen in https://github.com/unslothai/unsloth/pull/12050
* Let the macOS tab sampler ride out a navigation still in flight after login by @danielhanchen in https://github.com/unslothai/unsloth/pull/12053
* Studio: let Deep Research finish a turn handed off from a chat generation by @MohammadHijjawi97 in https://github.com/unslothai/unsloth/pull/11923
* Studio: custom sidebar sections, section menus and drag to reorder by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12016
* Studio: link folders when creating a project by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12057
* Studio: open the user menu Help submenu upward by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12032
* Skip chordless menu items in the native chord collision check by @danielhanchen in https://github.com/unslothai/unsloth/pull/12063
* Give the health wait's working-child tests room for the worker to start by @danielhanchen in https://github.com/unslothai/unsloth/pull/12065
* Count #12016's section header among the sidebar's scaled 30px rows by @danielhanchen in https://github.com/unslothai/unsloth/pull/12066
* Turn off Dr GRPO reward scaling under TRL's "group" default by @vineethsaivs in https://github.com/unslothai/unsloth/pull/11951
* Studio: keep the sidebar menu shadow in light mode by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12062
* Reset the permission step's storage at the start of the next document by @danielhanchen in https://github.com/unslothai/unsloth/pull/12073
* Let the Projects section stand in for its row in the macOS tab walk by @danielhanchen in https://github.com/unslothai/unsloth/pull/12074
* Load #11526's text-core refusal in the save_method routing harness by @danielhanchen in https://github.com/unslothai/unsloth/pull/12077
* Studio: read personalization only once a first sign-in has changed its password by @danielhanchen in https://github.com/unslothai/unsloth/pull/12071
* Studio: train a SQuAD answer's text, not the answers dict, when mapping columns to chat roles by @breken-ai in https://github.com/unslothai/unsloth/pull/12056
* Back off and retry the update banner navigation on ERR_NO_BUFFER_SPACE by @danielhanchen in https://github.com/unslothai/unsloth/pull/12081
* Keep TRL's own RL config defaults and clamp preference max_length by @danielhanchen in https://github.com/unslothai/unsloth/pull/12069
* Unsloth Studio (AMD/Windows): report an iGPU's used VRAM when a discrete card sits beside it by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11871
* CI: CodeQL advanced setup that analyses only the languages a PR touches by @danielhanchen in https://github.com/unslothai/unsloth/pull/11998
* Give the killed formatter grandchild as long to die as it had to start by @danielhanchen in https://github.com/unslothai/unsloth/pull/12086
* Studio: rework Appearance settings and add flavor color themes by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12030
* Studio: quieter settings headings and one section gap on every page by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12064
* Keep the padding-free column test off the datasets numpy formatter by @danielhanchen in https://github.com/unslothai/unsloth/pull/12082
* Studio: put Reapply next to Generate on the image and video pages by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11974
* Unsloth Studio / Desktop: keep the Images Cancel load pill off the model panel divider by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/12079
* Studio: keep run duration on Mac when training with an eval set by @NilayYadav in https://github.com/unslothai/unsloth/pull/11967
* fix: handle strided cross entropy inputs by @MrCapricornLiu in https://github.com/unslothai/unsloth/pull/10713
* Stop unsloth chat and unsloth inference switching a loaded GGUF to a different quant by @NilayYadav in https://github.com/unslothai/unsloth/pull/11855
* Studio: tell Safari which account the Studio password belongs to by @NilayYadav in https://github.com/unslothai/unsloth/pull/11858
* Studio: find models in a Hugging Face cache folder added as a location by @NilayYadav in https://github.com/unslothai/unsloth/pull/11970
* Unsloth Studio installer (AMD/Windows): don't mistake ZLUDA for an NVIDIA GPU by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11736
* Studio: follow redirect pages when reading a web page by @NilayYadav in https://github.com/unslothai/unsloth/pull/11856
* Studio: don't stop long running tool calls that print nothing by @NilayYadav in https://github.com/unslothai/unsloth/pull/11969
* Studio: stop the dense-quant probes pinning a CUDA context on every card of a multi-GPU host by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11954
* Studio: reject echo, suffix and best_of on /v1/completions by @NilayYadav in https://github.com/unslothai/unsloth/pull/11964
* Fix the prebuilt wheel publish step and shorten its release notes by @danielhanchen in https://github.com/unslothai/unsloth/pull/12068
* Studio: stop generating when the client disconnects on a non-streaming request by @NilayYadav in https://github.com/unslothai/unsloth/pull/11961
* Studio: decode the SDXL VAE in fp16 on fp16 GPUs by @danielhanchen in https://github.com/unslothai/unsloth/pull/12036
* Studio: honor response_format on the MLX backend with grammar-constrained decoding by @Lyxot in https://github.com/unslothai/unsloth/pull/10180
* Studio: document viewer for PDF, Word, Excel and PowerPoint, with origin links in Library by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12001
* Studio: chat attachment cards, chips and the Library viewer by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12017
* Studio: keep an eagerly decoded image VAE contiguous on NVIDIA by @danielhanchen in https://github.com/unslothai/unsloth/pull/12035
* Studio: skip the cuDNN benchmark search for the MiniMax-H3 audio VAE on A100 / B200 / RTX PRO 6000 (first render up to a minute faster, 25-29 GiB lower peak) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12040
* Studio: keep a chat's start date in the system prompt, note a new date on the latest user turn by @danielhanchen in https://github.com/unslothai/unsloth/pull/12096
* Keep Llama 3.2 Vision off flash attention (vision and cross attention have no is_causal) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12033
* Load the repo AutoProcessor for AutoModel-only repo-code VLMs by @danielhanchen in https://github.com/unslothai/unsloth/pull/12037
* Unsloth Studio / Desktop: let the GPUs picker say how much of the model each card gets by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/12015
* Studio: regionally compile Lumina-2 and HiDream-I1, and re-decide their auto precision by measurement by @danielhanchen in https://github.com/unslothai/unsloth/pull/12039
* Studio: refuse a hand-set Metal context only past the GPU wired limit by @Lyxot in https://github.com/unslothai/unsloth/pull/10804
* Studio: keep a pinned context as a request limit instead of refusing KV cache quantization by @Lyxot in https://github.com/unslothai/unsloth/pull/11084
* Studio: let full-scope keyless callers auto-switch models, and explain the refusal elsewhere by @Lyxot in https://github.com/unslothai/unsloth/pull/11180
* Upload all GGUF files in one commit so create_pr opens one pull request by @NilayYadav in https://github.com/unslothai/unsloth/pull/11857
* Studio: return an error when a model can't use the tools sent to it by @NilayYadav in https://github.com/unslothai/unsloth/pull/11960
* Read settings.py as utf-8 in the palette filter test by @danielhanchen in https://github.com/unslothai/unsloth/pull/12105
* Read settings.py as UTF-8 in the unknown-palette test by @danielhanchen in https://github.com/unslothai/unsloth/pull/12101
* Carry the resolved attention implementation to nested configs a remote config baked flash attention into (Nemotron 3 Nano Omni) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12099
* Studio: share MLX VLM prompt-cache snapshot buffers and replay exact prompts by @Lyxot in https://github.com/unslothai/unsloth/pull/11659
* GRPO: default to TRL's dapo loss with beta 0, cap CISPO weights at 5.0 by @danielhanchen in https://github.com/unslothai/unsloth/pull/12088
* Patch the TRL trainers that moved to trl.experimental (ORPO, CPO, Online DPO, GKD, ...) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12097
* Studio: quantize the KV cache of sliding-window MLX models such as Gemma 4 by @Lyxot in https://github.com/unslothai/unsloth/pull/11082
* Studio: cache and coalesce nvidia-smi reads in the backend by @danielhanchen in https://github.com/unslothai/unsloth/pull/11995
* Unsloth Studio installer (AMD/Linux): explain why a newer ROCm gets ROCm 7.2 PyTorch by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11651
* Treat a reaped grandchild as dead in the formatter timeout test by @danielhanchen in https://github.com/unslothai/unsloth/pull/12107
* Trim comments in the DeepSeek-V4 grouped LoRA, remote-code shims and Mistral-format loader code by @danielhanchen in https://github.com/unslothai/unsloth/pull/12089
* Keep ORPO / CPO rows within max_length on TRL 0.29+ by @danielhanchen in https://github.com/unslothai/unsloth/pull/12115
* Keep the Triton MoE grouped GEMM in compiled graphs and index weights past 2^31 elements by @danielhanchen in https://github.com/unslothai/unsloth/pull/12114
* Studio: read more chat attachment formats, and hand the rest to the python tool by @Lyxot in https://github.com/unslothai/unsloth/pull/11379
* Unsloth Desktop: Create, Edit, and Delete Skills from the Skills Menu by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/11800
* Studio: show the Hub error for an unreadable GGUF repo instead of routing it to Transformers by @danielhanchen in https://github.com/unslothai/unsloth/pull/12117
* Count the MLX grammar engine slot in the Apple Silicon step totals by @danielhanchen in https://github.com/unslothai/unsloth/pull/12135
* Pass gradients through compressed-tensors activation quantization so W8A8 checkpoints train with LoRA by @danielhanchen in https://github.com/unslothai/unsloth/pull/11585
* Studio: keep the last full-attention layer unquantized in the MLX KV cache by @Lyxot in https://github.com/unslothai/unsloth/pull/11083
* Gate the Mllama CUDA forward test on has_real_cuda by @danielhanchen in https://github.com/unslothai/unsloth/pull/12143
* Batched serving on the MLX path: several replies decoding at once by @Lyxot in https://github.com/unslothai/unsloth/pull/10310
* Point dsh at Unsloth through a --patch overlay instead of settings.yaml by @danielhanchen in https://github.com/unslothai/unsloth/pull/12145
* fix(studio): mark /api reads no-store so an idle desktop app stops rewriting its disk cache by @alkinun in https://github.com/unslothai/unsloth/pull/12148
* Wait for every diffusion run a test started before undoing its runs dir by @danielhanchen in https://github.com/unslothai/unsloth/pull/12149
* Studio: move Managed accounts to the Accounts tab and drop Mark as unread from chat menus by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12120
* Studio: keep checkpoint compaction under --disable-tools by @danielhanchen in https://github.com/unslothai/unsloth/pull/12119
* Studio: only offer Agent Skills when Code is on by @danielhanchen in https://github.com/unslothai/unsloth/pull/12118
* Settle the permission step's reloads on the pill instead of networkidle by @danielhanchen in https://github.com/unslothai/unsloth/pull/12153
* Speed up block-FP8 LoRA training: run FP8 linears eagerly, 8 warps for 128-row GEMM tiles by @danielhanchen in https://github.com/unslothai/unsloth/pull/12027
* Fall back from FBGEMM for rowwise FP8 on GPUs it has no kernel for (RTX PRO 6000 / 5090) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12098
* Studio: TurboQuant KV cache option for MLX inference by @Lyxot in https://github.com/unslothai/unsloth/pull/11170
* Fix rowwise FP8 scale axes in fused LoRA backward by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/11799
* Unsloth Studio Installer: ask Python for a path identity before giving up on an exact one by @danielhanchen in https://github.com/unslothai/unsloth/pull/11104
* Ask Python for the process image table when the native helper is unavailable by @danielhanchen in https://github.com/unslothai/unsloth/pull/11115
* Refresh a rewritten shortcut's icon where the shell cannot define the type by @danielhanchen in https://github.com/unslothai/unsloth/pull/11116
* Recover CUDA compute capabilities without emitting a P/Invoke type by @danielhanchen in https://github.com/unslothai/unsloth/pull/11173
* Remove the reflection-emit apparatus and all four emitted types by @danielhanchen in https://github.com/unslothai/unsloth/pull/11193
* Add a Windows probe for code integrity blocks, and audit bundle signatures in CI by @danielhanchen in https://github.com/unslothai/unsloth/pull/10408
* Windows installers find an NVIDIA GPU on the PCI bus, and say which CUDA it can use by @danielhanchen in https://github.com/unslothai/unsloth/pull/11166
* Fix Gemma2 padding masks during batched cached decoding by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/12008
* Gemma2: use flash_attn_with_kvcache for cached decoding by @danielhanchen in https://github.com/unslothai/unsloth/pull/12112
* Fix Gemma and Gemma2 embeddings scaled twice on transformers 5.4+ by @danielhanchen in https://github.com/unslothai/unsloth/pull/12116
* Keep scan_packages.py from tripping Bitdefender's Python stealer signature by @danielhanchen in https://github.com/unslothai/unsloth/pull/12169
* Studio: show when a prompt was sent while hovering it by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12170
* Studio: add a Scroll while generating setting (Auto-scroll or Manual) by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12110
* Studio: make compile knobs reach the render thread on torch 2.12+ by @danielhanchen in https://github.com/unslothai/unsloth/pull/12075
* Studio: stop a hung system node or npm from stalling setup by @oobabooga in https://github.com/unslothai/unsloth/pull/12165
* Studio: price MLX loads in the memory panel and fit an unpinned context to available memory by @Lyxot in https://github.com/unslothai/unsloth/pull/10287
* Gemma2: keep softcapping attention under int32 indexing and fall back to eager if compile fails by @danielhanchen in https://github.com/unslothai/unsloth/pull/12154
* Fix training with accelerate 1.15 on torch without a distributed backend (AMD Windows ROCm) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12162
* Studio: stop runaway tool output from using up memory by @NilayYadav in https://github.com/unslothai/unsloth/pull/11723
* Studio: reset the download progress bar when a retry restarts the file by @NilayYadav in https://github.com/unslothai/unsloth/pull/11593
* Studio: Select all only picks the models the search shows by @NilayYadav in https://github.com/unslothai/unsloth/pull/11485
* Studio: ignore an SSLKEYLOGFILE the process cannot write instead of failing every HTTPS client by @danielhanchen in https://github.com/unslothai/unsloth/pull/12166
* Tests: stop Windows tests tripping Bitdefender and the 16-bit application dialog on real machines by @danielhanchen in https://github.com/unslothai/unsloth/pull/12167
* Studio: Chats library in the Library by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12122
* Studio: round hover for the Settings close button by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12172
* unsloth start opencode: size the output limit to the context and add --max-tokens by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12111
* SAC probe: verify Studio identity before sending a password by @danielhanchen in https://github.com/unslothai/unsloth/pull/12176
* Attention resolver: respect a declared _supports_sdpa = False, and skip flash_attention_2 when a class's compatible flash kernels exclude it (MiMo-V2-Flash) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12147
* Keep Nemotron-H mixer.out_proj unquantized under a caller's BitsAndBytesConfig by @danielhanchen in https://github.com/unslothai/unsloth/pull/12131
* Phi-4-reasoning-vision: give remote multimodal prep an indexable cache view on transformers 5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/12123
* Studio: LTX-2.3 about 4.5x faster per clip (unguided distilled sampling, compile fixes, hosted FP8) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12067
* Let a loaded Kimi K2.5 / K2.7 processor take processor(text=..., images=...) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12126
* Build the native image processor at defaults when a VLM repo has no preprocessor_config.json by @danielhanchen in https://github.com/unslothai/unsloth/pull/12125
* Refuse K-EXAONE 2.0 on a transformers that ignores its config by @danielhanchen in https://github.com/unslothai/unsloth/pull/12128
* Studio: list at most the 12 most recently active projects and sections in Move to by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/12155
* Turn off the MoE aux loss for dense models that carry a router config by @danielhanchen in https://github.com/unslothai/unsloth/pull/12076
* Studio: keep conversation recall working across repeated compactions by @danielhanchen in https://github.com/unslothai/unsloth/pull/12174
* Studio: never ask for approval to run search_conversation by @danielhanchen in https://github.com/unslothai/unsloth/pull/12175
* Add a CI gate for code shapes that heuristic antivirus scanners quarantine by @danielhanchen in https://github.com/unslothai/unsloth/pull/12178
* Dequantize block-FP8 weights with a ragged last block on 16-bit loads (GLM-5.3) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12133
* Rebuild CohereTokenizer from tokenizer.json when transformers v5 changes its ids by @danielhanchen in https://github.com/unslothai/unsloth/pull/12130
* Keep flash attention off towers the Auto classes do not register by @danielhanchen in https://github.com/unslothai/unsloth/pull/12129
* Keep Linear layers an FP8 checkpoint stores in bf16 unconverted by @danielhanchen in https://github.com/unslothai/unsloth/pull/12124
* Load speech-to-text models (Voxtral, Qwen2-Audio) through FastModel by @danielhanchen in https://github.com/unslothai/unsloth/pull/12127
* Narrow Kimi-K3's zero-padded KDA A_log to num_heads so the fla backward runs by @danielhanchen in https://github.com/unslothai/unsloth/pull/12132
* Repair chat templates that always append the generation prompt, also on FastModel loads by @danielhanchen in https://github.com/unslothai/unsloth/pull/12139
* Route GKD distillation through the chunked generalized JSD by @danielhanchen in https://github.com/unslothai/unsloth/pull/12136
* Studio: list HF cache models written without symlinks in the local inventory by @danielhanchen in https://github.com/unslothai/unsloth/pull/12156
* Studio: read public Hub repos without a saved token the Hub rejects by @danielhanchen in https://github.com/unslothai/unsloth/pull/12158
* Studio: list a crashed or cancelled run's saved checkpoints on the Export page by @NilayYadav in https://github.com/unslothai/unsloth/pull/11600
* Studio: pick the right model size when the file names are lowercase by @NilayYadav in https://github.com/unslothai/unsloth/pull/11722
* Refresh WSL shortcut icons through Python instead of an emitted native stub by @danielhanchen in https://github.com/unslothai/unsloth/pull/12184
* Read the gradient checkpointing precondition off transformers' own method by @danielhanchen in https://github.com/unslothai/unsloth/pull/12187
* Studio: Triton-fused VAE norms, caches and attention for every image and video VAE (1.7x to 6.3x decode) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12078
* Studio: fix Decision API validation and TypeSafe SDK metadata by @wasimysaid in https://github.com/unslothai/unsloth/pull/12186
* Studio: grant MXC Tier 3 read access to runtime folders once, not per launch by @danielhanchen in https://github.com/unslothai/unsloth/pull/12121
* Studio: tell a rejected Hugging Face token apart from an unreachable Hub by @danielhanchen in https://github.com/unslothai/unsloth/pull/12159
* Studio: run the isolated Windows Terminal on cmd.exe with stock git when Git Bash cannot start in MXC by @danielhanchen in https://github.com/unslothai/unsloth/pull/12164
* Studio: load a downloaded GGUF from disk when the Hub cannot be read or refuses it by @danielhanchen in https://github.com/unslothai/unsloth/pull/12161
* Join the npm scanner's credential path markers from pieces by @danielhanchen in https://github.com/unslothai/unsloth/pull/12181
* Pass the MXC host-prep script as plain text instead of an encoded command by @danielhanchen in https://github.com/unslothai/unsloth/pull/12182
* Count only the retry loop's own sleeps in the JSON fallback backoff test by @danielhanchen in https://github.com/unslothai/unsloth/pull/12189
* Store the Studio sandbox credential path names in pieces by @danielhanchen in https://github.com/unslothai/unsloth/pull/12188
* Name utf-8 on the SSLKEYLOGFILE writability probe by @danielhanchen in https://github.com/unslothai/unsloth/pull/12190
* Studio: fused int8 MLP kernels and real-arithmetic RoPE for DiT denoisers (FLUX.1 11% less GPU time per step) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12083
* Studio: capability tags in the model picker match the vision pill and get their own colours by @danielhanchen in https://github.com/unslothai/unsloth/pull/12192
* Studio: fix inductor CantSplit on torch 2.12/2.13 and speed up the Qwen-Image-2.1 VAE by @danielhanchen in https://github.com/unslothai/unsloth/pull/12059
* Load block-FP8 checkpoints in 4-bit (NF4) when load_in_4bit=True is passed by @danielhanchen in https://github.com/unslothai/unsloth/pull/12146
* Studio: compile FLUX.1 dynamic, CUDA-graph the SDXL U-Net, decode one-frame Qwen-Image latents in 2D by @danielhanchen in https://github.com/unslothai/unsloth/pull/12060
* Studio: compile VAEs by repeated block, one H3 graph per first render, vectorise the HunyuanVideo-1.5 VAE mask by @danielhanchen in https://github.com/unslothai/unsloth/pull/12061
* Unsloth Studio: stop a Mac browser hiding GPU-only models from a remote CUDA / ROCm / Intel server by @danielhanchen in https://github.com/unslothai/unsloth/pull/8833
* Studio: return no-op instead of 500 for an out-of-range scan folder id by @danielhanchen in https://github.com/unslothai/unsloth/pull/8397
* Do not reinstall llm-compressor when it is already installed by @danielhanchen in https://github.com/unslothai/unsloth/pull/6806
* Fix Studio launcher repair for missing install id by @danielhanchen in https://github.com/unslothai/unsloth/pull/6933
* Load Mistral-format checkpoints (params.json only) through transformers' Mistral4 (Mistral-Large-3) by @danielhanchen in https://github.com/unslothai/unsloth/pull/12144
* Studio: name the MCP server while the tool call is still streaming by @danielhanchen in https://github.com/unslothai/unsloth/pull/9212
* Stop gpt-oss generation at the harmony tool call token by @danielhanchen in https://github.com/unslothai/unsloth/pull/11449
* Studio: vendor laya 0.3.5 and hold its weights in float16 by @danielhanchen in https://github.com/unslothai/unsloth/pull/12202
* Unpack the mirrored uv wheel with python3 -m zipfile instead of an inline one-liner by @danielhanchen in https://github.com/unslothai/unsloth/pull/12194
* Keep xFormers attention causal when the decoder is called without a mask by @danielhanchen in https://github.com/unslothai/unsloth/pull/12199
* GKD: right-align left-padded rows before the student forward by @danielhanchen in https://github.com/unslothai/unsloth/pull/12200

## New Contributors
* @L4XB made their first contribution in https://github.com/unslothai/unsloth/pull/11803
* @Abhishek-B-R made their first contribution in https://github.com/unslothai/unsloth/pull/11779
* @tweekli made their first contribution in https://github.com/unslothai/unsloth/pull/11781
* @goodmai made their first contribution in https://github.com/unslothai/unsloth/pull/11907
* @breken-ai made their first contribution in https://github.com/unslothai/unsloth/pull/11893
* @MohammadHijjawi97 made their first contribution in https://github.com/unslothai/unsloth/pull/11923

**Full Changelog**: https://github.com/unslothai/unsloth/compare/v0.1.815-beta...v0.1.900-beta