v0.1.906-beta

Models can now use the browser in Unsloth. Turn on Browser use and your model can open pages, read them, click, type and scroll while you watch. Qwen-Image-2.1-Turbo is here too, making images in just 8 steps. Also included: text-diffusion LLM training, faster MoE GGUFs and speculative decoding on Mac.

## Highlights

- **Browser use (in + menu in chatbox)** Local models can take control of the Unsloth Browser
- **Qwen-Image-2.1-Turbo**: images in 8 steps (2.5x faster). [See Guide](https://unsloth.ai/docs/models/qwen-image-2.1)
- **Train text-diffusion LLMs** like DiffusionGemma, LLaDA and Dream
- **Up to 3x faster MoE GGUFs** that are bigger than your VRAM
- **Speculative decoding on Mac** for up to 2.8x faster MLX replies

## Browser use

- Turn on Browser use from the + menu in chat, then ask your model to look something up or use a website.
- Models can open pages, read them, click, type, pick options and scroll while you watch.
- The model browses in its own tab beside the chat, with no access to your other tabs.
- Browser use works with local models and API providers, including small text-only models.
- Unsloth asks you before the model signs in, posts a form or presses a button that buys, sends or deletes.
- The model never fills in passwords, card numbers or one-time codes. Those are yours to enter.
- Choose which sites models can open, and their access to your browsing history and downloads, in Settings > Browser.

## Image generation

- Run Qwen-Image-2.1-Turbo, the official 8-step version of Qwen-Image-2.1, from the Images page, with a 4.2 GB Q4 GGUF. [GGUF](https://huggingface.co/unsloth/Qwen-Image-2.1-Turbo-GGUF)
- Export the image LoRAs you train in Unsloth as files you can reuse or share, and load image LoRAs from your own folders.
- ComfyUI models now use the text encoders already in your ComfyUI folder instead of downloading them again.

## Training

- Train text-diffusion LLMs (DiffusionGemma, LLaDA, LLaDA2, Dream, Nemotron and SDAR) with the new `DiffusionTrainer`.
- Train with LoRA+ by setting `loraplus_lr_ratio`, and fine-tune GPTQ models with LoRA.
- GRPO now trains on video prompts and tool calls with TRL 1.15, and `AsyncGRPOTrainer` can train and generate on one GPU.
- Fine-tuning on tool-calling datasets now includes each row's tool definitions, and works on multi-turn agent traces.
- On the Train page, set Offload layers with a VRAM budget (NVIDIA and AMD), and continue LoRA training from a saved adapter.
- Train Laya and Clef decision models on Mac. Clef can also train on images.

## Performance

- MoE GGUFs bigger than your VRAM now run up to 3x faster on NVIDIA and AMD GPUs.
- Training with sample packing is now up to 8x faster when flash-attn and xformers are not installed (NVIDIA, PyTorch 2.12+).
- Speculative decoding makes MLX replies on Mac up to 2.8x faster, with the same output.
- MLX speculative decoding sets its draft depth automatically, and supports MTP, DFlash, DSpark and EAGLE-3 drafters.
- Typing and scrolling stay smooth in long chats, and with hundreds of chats in the sidebar.

## Chat + Desktop

- React code blocks (JSX and TSX) in a reply now run as live previews in the Browser.
- Deep Research can use your MCP tools as sources and cite them, and retrying a failed run resumes from the steps it already finished.
- Unsloth now reads `AGENTS.md` from your Unsloth home and project folders into chat instructions.
- In Desktop, a project can now run its tools in an existing folder on your computer.
- Turn on Read replies automatically in Settings > Voice to hear each reply aloud when it finishes.

## Models + settings

- Use the optional vLLM engine on AMD GPUs, on Linux and through WSL on Windows.
- Run DeepSeek-OCR and DeepSeek-OCR-2.
- Filter Hub models by parameter count, context length and organization.
- Turn on the detailed Live monitor to see GPU load, temperature and power.
- On multi-user installs, limit which models each managed account can use.
- Install the Unsloth web UI as an app from a supported browser.

## Download Unsloth Desktop

Unsloth Desktop is **free and open source**. Download it for:
<table>
  <tr>
    <td><b>Platform</b></td>
    <td><b>Link</b></td>
  </tr>
  <tr>
    <td><b>Windows</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Windows.exe'>Download</a></td>
  </tr>
  <tr>
    <td><b>macOS</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-MacOS.dmg'>Download</a></td>
  </tr>
  <tr>
    <td><b>Linux x64 / Ubuntu (deb)</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Ubuntu.deb'>Download</a></td>
  </tr>
  <tr>
    <td><b>Linux ARM64 / Ubuntu 24.04+ (deb)</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Ubuntu-ARM64.deb'>Download</a></td>
  </tr>
  <tr>
    <td><b>Linux x64 (AppImage)</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/download/Unsloth-Desktop-Linux.AppImage'>Download</a></td>
  </tr>
  <tr>
    <td><b>Windows ARM64</b></td>
    <td><a href='https://github.com/unslothai/unsloth/releases/latest/Unsloth-Desktop-Windows-ARM64.exe'>Download</a></td>
  </tr>
</table>

## What's Changed
* Bump install.sh / install.ps1 pins to unsloth>=2026.10.3 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13083
* Keep loss-only evaluation on the fused CE path instead of materializing logits by @danielhanchen in https://github.com/unslothai/unsloth/pull/13065
* Studio: int8 ConvRot Qwen-Image-2.1 text encoder on Apple Silicon, and release text encoders during denoise on unified memory by @danielhanchen in https://github.com/unslothai/unsloth/pull/13026
* Save models loaded from pre-quantized bitsandbytes checkpoints on transformers 5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13054
* Studio: re-measure annotate marks after a tab zoom by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13082
* Studio: install the released Diffusers 0.41.0 instead of the pinned main commit by @danielhanchen in https://github.com/unslothai/unsloth/pull/13045
* Studio: restart llama-server after a lost GPU device instead of locking the chat by @danielhanchen in https://github.com/unslothai/unsloth/pull/13044
* Studio: keep CUDA 12 llama.cpp prebuilts off drivers older than CUDA 12.4 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13031
* Studio: don't take a port another process holds on 0.0.0.0 by @SulimanAbdulrazzaq in https://github.com/unslothai/unsloth/pull/12011
* Studio: load community SDXL single-file GGUFs on the Images page by @danielhanchen in https://github.com/unslothai/unsloth/pull/13035
* Studio: fix Qwen-Image-2.1 input-image encode on Apple Silicon (MPS frame-axis pad returns wrong data) by @Lyxot in https://github.com/unslothai/unsloth/pull/13028
* Dequantize pre-quantized bitsandbytes checkpoints for full finetuning by @danielhanchen in https://github.com/unslothai/unsloth/pull/13079
* Add loraplus_lr_ratio for LoRA+ training by @danielhanchen in https://github.com/unslothai/unsloth/pull/13062
* MLX: train decision models on Apple Silicon by @Lyxot in https://github.com/unslothai/unsloth/pull/13016
* Unsloth Studio / Desktop: Block Swap settings, a VRAM budget, and a live layer panel on the training page by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/12996
* Unsloth Studio / Desktop: per-GPU Block Swap budget and layer panel for multi-GPU training by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/12998
* Studio: keep typing and streaming fast in long chats by @danielhanchen in https://github.com/unslothai/unsloth/pull/13059
* Reject Activated LoRA adapters instead of training them as plain LoRA by @danielhanchen in https://github.com/unslothai/unsloth/pull/13089
* Studio: provide Git lazily for FP8/FP4 export instead of at install time by @danielhanchen in https://github.com/unslothai/unsloth/pull/13084
* Studio: let Gemma 4 GGUF models think after a tool call by @NilayYadav in https://github.com/unslothai/unsloth/pull/13097
* GGUF adapter export and llama.cpp installer: use unslothai/llama.cpp releases only, no git by @danielhanchen in https://github.com/unslothai/unsloth/pull/13056
* Windows installer: stop installing Git and the VS toolchain without asking by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/12971
* Studio: keep table columns in attached Word and HTML files by @NilayYadav in https://github.com/unslothai/unsloth/pull/13105
* Studio: stop link URLs from crowding out long page text by @NilayYadav in https://github.com/unslothai/unsloth/pull/13102
* Studio: keep Think off working on Claude Opus 5.5 and Sonnet 5.5 by @NilayYadav in https://github.com/unslothai/unsloth/pull/13104
* Studio: raise the llama-server micro-batch to 2048 when MoE experts spill to RAM by @danielhanchen in https://github.com/unslothai/unsloth/pull/12950
* Studio: let imported chats with image links answer follow-up questions by @NilayYadav in https://github.com/unslothai/unsloth/pull/13099
* Studio: hold message hover still while a wheel scroll moves the thread by @danielhanchen in https://github.com/unslothai/unsloth/pull/13077
* Studio: fill the context bar for Ollama connections by @NilayYadav in https://github.com/unslothai/unsloth/pull/13106
* Studio: read news articles whose pages carry megabytes of inline code by @NilayYadav in https://github.com/unslothai/unsloth/pull/13100
* Studio: open Images workflows from More when Images is unpinned by @NilayYadav in https://github.com/unslothai/unsloth/pull/13101
* Studio: let Deep Research search MCP tools as extra sources by @NilayYadav in https://github.com/unslothai/unsloth/pull/13103
* Studio: let Gemma 4 read web search and tool results on safetensors by @NilayYadav in https://github.com/unslothai/unsloth/pull/13096
* Studio: reopen a chat on the branch it was left on by @oobabooga in https://github.com/unslothai/unsloth/pull/13109
* fix(chat): import branch threads parents-before-children on refresh by @CelesteHeartsong in https://github.com/unslothai/unsloth/pull/13113
* Studio: show Gemini's thinking in chat by @NilayYadav in https://github.com/unslothai/unsloth/pull/13107
* Studio: show the context meter for active project chats by @lorenzozanee in https://github.com/unslothai/unsloth/pull/12605
* Fix returned logit scaling and softcapping when labels are provided by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/13069
* Studio: head a decision score with its most likely level by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/13111
* Keep slow tokenizers transformers cannot convert instead of crashing the load by @danielhanchen in https://github.com/unslothai/unsloth/pull/13073
* Turn torch.compile off on GPUs older than Volta by @danielhanchen in https://github.com/unslothai/unsloth/pull/13085
* Keep paged bitsandbytes optimizer state paged on checkpoint resume by @danielhanchen in https://github.com/unslothai/unsloth/pull/13092
* Fix LLaVA image token count when processor_config lacks num_additional_image_tokens by @danielhanchen in https://github.com/unslothai/unsloth/pull/13091
* Let GPTQ checkpoints train and generate with LoRA by @danielhanchen in https://github.com/unslothai/unsloth/pull/13068
* Fix PPOTrainer with an Unsloth policy (gradient checkpointing, generate, RoPE on value and reward models) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13067
* Fix beam search (num_beams > 1) on the fast decode path by @danielhanchen in https://github.com/unslothai/unsloth/pull/13053
* Fix TRL PPOTrainer under Unsloth: rollout crash, sampling filters and left padding by @danielhanchen in https://github.com/unslothai/unsloth/pull/13108
* Load Whisper in FastModel without auto_model by @danielhanchen in https://github.com/unslothai/unsloth/pull/13080
* Repair RoPE for plain transformers loads after an Unsloth load (reward models in Online DPO / GRPO) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13055
* Point a 4-bit load that does not fit the GPU at offload_layers = "auto" by @danielhanchen in https://github.com/unslothai/unsloth/pull/13052
* Make precomputed inputs_embeds train under gradient checkpointing by @danielhanchen in https://github.com/unslothai/unsloth/pull/13088
* Right-pad encoder-decoder tokenizers and load SeamlessM4T without auto_model by @danielhanchen in https://github.com/unslothai/unsloth/pull/13066
* SyntheticDataKit: take a port argument and move off 8000 when it is taken by @danielhanchen in https://github.com/unslothai/unsloth/pull/13072
* Load Qwen2.5-Omni checkpoints that name Qwen2_5OmniModel by @danielhanchen in https://github.com/unslothai/unsloth/pull/13087
* Fix granite-vision 4-bit generate crash on transformers 4.x (nn.MultiheadAttention out_proj) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13090
* Studio: avoid repeated Qwen-Image-2.1 VAE tile OOMs on ROCm by @oobabooga in https://github.com/unslothai/unsloth/pull/13116
* fix(studio): repair orphan tool results during chat rendering by @lorenzozanee in https://github.com/unslothai/unsloth/pull/11733
* Studio: fall back to the legacy stream when an older backend refuses a new chat field by @danielhanchen in https://github.com/unslothai/unsloth/pull/13130
* Studio: use the Vulkan llama.cpp bundle, not CPU, on Windows NVIDIA drivers below CUDA 12.4 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13129
* Studio: load GGUF models on Windows when the user profile path is not ASCII by @danielhanchen in https://github.com/unslothai/unsloth/pull/13133
* README: install PyTorch before Unsloth on Windows so uv does not resolve an old Unsloth by @danielhanchen in https://github.com/unslothai/unsloth/pull/13134
* fix(install): silence the missing AppArmor userns sysctl read by @jayzhou2309 in https://github.com/unslothai/unsloth/pull/13125
* Studio: send Decision API images to the MLX engine for Clef by @Lyxot in https://github.com/unslothai/unsloth/pull/13126
* MLX decision training: Clef rows and predict take images by @Lyxot in https://github.com/unslothai/unsloth/pull/13127
* Studio: keep merged table cells in their columns when reading web pages by @NilayYadav in https://github.com/unslothai/unsloth/pull/13098
* Studio: install the llm-compressor-main shadow from a source archive instead of git by @danielhanchen in https://github.com/unslothai/unsloth/pull/13132
* Decide accepts_loss_kwargs from the loss head so gradient accumulation gets num_items_in_batch for every model by @danielhanchen in https://github.com/unslothai/unsloth/pull/12946
* Keep Dia's audio pad token in fast generate by @danielhanchen in https://github.com/unslothai/unsloth/pull/13128
* Studio: speculative decoding for the MLX inference backend by @Lyxot in https://github.com/unslothai/unsloth/pull/13014
* Studio: repair ROCm attention packages and bound Qwen-Image-2.1 math prefill by @oobabooga in https://github.com/unslothai/unsloth/pull/13115
* Studio: offer vLLM or SGLang when the default engine cannot run a quantized checkpoint by @danielhanchen in https://github.com/unslothai/unsloth/pull/13135
* Fix evaluate() under auto padding-free and keep compute_metrics rows per example by @danielhanchen in https://github.com/unslothai/unsloth/pull/13136
* GRPO: train on video prompts instead of scoring them as text by @danielhanchen in https://github.com/unslothai/unsloth/pull/13144
* Studio: stop with a clear error when a pre-cast text encoder fails and its dense shards were skipped by @KafKafrnZ in https://github.com/unslothai/unsloth/pull/13117
* Studio: stop RAG PDF parsing from starving the server on scanned pages by @danielhanchen in https://github.com/unslothai/unsloth/pull/13137
* Studio: keep queued prompts when the model is reloaded or switched by @danielhanchen in https://github.com/unslothai/unsloth/pull/13147
* Baseline the unsloth-zoo 2026.10.3 and dnspython 2.9.0 findings after review by @danielhanchen in https://github.com/unslothai/unsloth/pull/13143
* Studio: keep a failed folder link from spending its grant by @danielhanchen in https://github.com/unslothai/unsloth/pull/13131
* Clean machine install: assert the no-git path by what setup.ps1 prints after #12971 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13170
* Installer differential: ignore winget's partly filled download bars by @danielhanchen in https://github.com/unslothai/unsloth/pull/13172
* Studio: keep annotate marks on PDF pages and slides that remount by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13119
* Fix 4-bit VLM DDP training crash on ranks >= 1 (CheckpointError) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13149
* Installers: hash-check the uv fallback script; fail closed on undigested Lemonade assets by @danielhanchen in https://github.com/unslothai/unsloth/pull/13189
* Studio: cap every /api write body before auth; keep chat text out of INFO logs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13192
* Studio: open http(s): source links without slashes as web links; flag Windows theme downloads by @danielhanchen in https://github.com/unslothai/unsloth/pull/13195
* Studio: add a sandbox memory limit setting to Settings > Sandbox by @danielhanchen in https://github.com/unslothai/unsloth/pull/13194
* Studio: translate the browser Downloads and video player strings in every locale by @danielhanchen in https://github.com/unslothai/unsloth/pull/13169
* Fix the Backend CI test failures after the branch-restore and Omni routing merges by @danielhanchen in https://github.com/unslothai/unsloth/pull/13173
* Apply embedding_learning_rate to embeddings under full finetuning by @danielhanchen in https://github.com/unslothai/unsloth/pull/13171
* Use int64 row offsets in RoPE and norm kernels by @arcusbuilds in https://github.com/unslothai/unsloth/pull/13121
* Hub token identity tests: let the llama-backend stubs take the caller's token after #13005 by @lonexreb in https://github.com/unslothai/unsloth/pull/13022
* Studio: stop a linked folder re-indexing failed files forever and let Unlink stop a running sync by @indrajeetapache in https://github.com/unslothai/unsloth/pull/13145
* Fix DecisionTrainer crashing on step 1 without a prior inference pass by @danielhanchen in https://github.com/unslothai/unsloth/pull/13176
* Studio: add a spell check switch to Settings > General by @KafKafrnZ in https://github.com/unslothai/unsloth/pull/13023
* Studio: support Qwen-Image-2.1-Turbo with its 8-step sampling schedule by @oobabooga in https://github.com/unslothai/unsloth/pull/13159
* version_compat tests: follow TRL 1.15's chunked_nll alias and GPU-only DPO scorer by @danielhanchen in https://github.com/unslothai/unsloth/pull/13207
* Studio/feat: add a run-setting toggle to disable tool-call deduplication by @frostyautumnleaf in https://github.com/unslothai/unsloth/pull/11685
* Studio: make OpenBLAS thread limits work on Windows ROCm, and default numpy to up to 8 threads instead of 1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13048
* Unsloth Studio installer (AMD/Linux): use ROCm 6.4 PyTorch on ROCm 6.0 to 6.3 so bitsandbytes training doesn't crash by @TaskPuppyNatani in https://github.com/unslothai/unsloth/pull/10277
* Studio: read AGENTS.md into chat instructions by @danielhanchen in https://github.com/unslothai/unsloth/pull/13199
* Studio: stop cutting a fetched page down to one article element by @NilayYadav in https://github.com/unslothai/unsloth/pull/13178
* Fix case-sensitive Hugging Face projector pairing by @Imagineer99 in https://github.com/unslothai/unsloth/pull/13161
* Studio: send each line of a file attached in Chat with files once by @NilayYadav in https://github.com/unslothai/unsloth/pull/13179
* Studio: mount sidebar Recents a page at a time so long chat histories stay fast by @oobabooga in https://github.com/unslothai/unsloth/pull/13165
* Studio: train tool-calling datasets with their tool list by @NilayYadav in https://github.com/unslothai/unsloth/pull/13180
* Studio: turn on agent access (MCP) from Settings > API by @Etherll in https://github.com/unslothai/unsloth/pull/13167
* Studio: name the /v1/models VRAM-fit context estimate by @AzizMuminov in https://github.com/unslothai/unsloth/pull/12599
* Fix illegal memory access in block-FP8 LoRA kernels on long batches by @danielhanchen in https://github.com/unslothai/unsloth/pull/13141
* Studio: show the whole conversation for imported Claude chats by @NilayYadav in https://github.com/unslothai/unsloth/pull/13182
* Studio: keep HTML table rows together in knowledge bases and sources by @NilayYadav in https://github.com/unslothai/unsloth/pull/13186
* Studio: let small models send an attached image to a mapped MCP tool by @Etherll in https://github.com/unslothai/unsloth/pull/13166
* Studio: let @skill mentions work without Code and add bulk skill toggles by @Etherll in https://github.com/unslothai/unsloth/pull/13163
* Studio: fill the context bar for Qwen connections by @NilayYadav in https://github.com/unslothai/unsloth/pull/13187
* Studio: regenerate a chat's title from its latest turns by @danielhanchen in https://github.com/unslothai/unsloth/pull/13197
* Fall back to Unsloth inference for gpt-oss with fast_inference=True by @danielhanchen in https://github.com/unslothai/unsloth/pull/13155
* Record the repo id for ModelScope loads so merged_16bit and GGUF export work by @danielhanchen in https://github.com/unslothai/unsloth/pull/13150
* Studio: give compare chat the same composer as normal chat by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13212
* Studio: keep the composer expand button in the top corner when a file is attached by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13198
* Studio: cap sd.cpp threads at the cgroup CPU quota and honour UNSLOTH_CPU_THREADS by @danielhanchen in https://github.com/unslothai/unsloth/pull/13213
* Studio: treat a Smart App Control block of unsloth.exe like the AppLocker one by @oobabooga in https://github.com/unslothai/unsloth/pull/13168
* Studio: list unsloth/Qwen-Image-2.1-Turbo-GGUF in the Images picker by @danielhanchen in https://github.com/unslothai/unsloth/pull/13206
* Studio: shorten the vLLM / SGLang install prompt by @danielhanchen in https://github.com/unslothai/unsloth/pull/13216
* Studio: fit MCP image mappings in the dialog and keep one field per tool by @Etherll in https://github.com/unslothai/unsloth/pull/13162
* Studio: show the installed Lemonade and FastFlowLM versions on the NPU by @oobabooga in https://github.com/unslothai/unsloth/pull/13151
* Studio: choose how often update notifications appear, including Unsloth updates by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13200
* Studio: bound Qwen-Image-2.1 attention on Apple Silicon and fail loudly on NaN images by @danielhanchen in https://github.com/unslothai/unsloth/pull/13188
* Unsloth Studio (AMD): pick a discrete GPU that fits over a larger APU in automatic GPU selection by @danielhanchen in https://github.com/unslothai/unsloth/pull/13196
* install.sh: install XPU PyTorch on Linux when an Intel Arc / Data Center GPU is the only GPU by @danielhanchen in https://github.com/unslothai/unsloth/pull/13193
* Studio: name the app Unsloth in the MCP settings copy by @danielhanchen in https://github.com/unslothai/unsloth/pull/13230
* Studio (Jetson): keep JetPack's CUDA ahead of pip CUDA in llama-server and backend loader paths by @danielhanchen in https://github.com/unslothai/unsloth/pull/13191
* Studio: keep the chat composer typing after a send on WebKitGTK Wayland by @danielhanchen in https://github.com/unslothai/unsloth/pull/13222
* Studio: keep tool calls and results when training on ShareGPT tool data by @NilayYadav in https://github.com/unslothai/unsloth/pull/13181
* Train text-diffusion LMs: DiffusionTrainer with recipes for DiffusionGemma, LLaDA, LLaDA-MoE, LLaDA2, Dream, Nemotron and SDAR by @danielhanchen in https://github.com/unslothai/unsloth/pull/13157
* Fall back to a dynamic KV cache when the static one would not fit by @danielhanchen in https://github.com/unslothai/unsloth/pull/13075
* Tests: drop duplicate tests and fix three that checked nothing new by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/13211
* Unsloth Studio / Desktop (AMD, Windows): warn at startup when the AMD driver has the RDNA4 idle-eviction bug by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/13208
* Studio: keep heading text that sits inside a button on fetched pages by @NilayYadav in https://github.com/unslothai/unsloth/pull/13184
* Studio: browser Downloads button matches the toolbar and closes on a second click by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13226
* Studio: keep list numbers in attached Word and HTML files by @NilayYadav in https://github.com/unslothai/unsloth/pull/13185
* Studio: pin Qwen-Image 2.1 Turbo directly under Qwen-Image 2.1 in the Images picker by @danielhanchen in https://github.com/unslothai/unsloth/pull/13239
* Support TRL 1.14 and 1.15: ORPO/CPO row cap, masked GRPO sampling metrics, cap to 1.15.0 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13203
* GRPO tool calling: accept Unsloth chat templates in TRL's response schema and training template lookup by @danielhanchen in https://github.com/unslothai/unsloth/pull/13202
* assistant_only_loss with Unsloth chat templates: fall back to response markers by @danielhanchen in https://github.com/unslothai/unsloth/pull/13204
* Let TRL AsyncGRPOTrainer train Unsloth models; fix GRPO vision rewards for tool images and TRL 1.15 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13205
* Make Unsloth-patched tokenizers and processors picklable by @danielhanchen in https://github.com/unslothai/unsloth/pull/13209
* Studio: show context usage for a revisited chat under a cloud model or none by @danielhanchen in https://github.com/unslothai/unsloth/pull/13225
* Add opt-in tokenizers 1.0 release candidate encode path by @danielhanchen in https://github.com/unslothai/unsloth/pull/13214
* Studio: size a GPU cache for spilled MoE experts with --moe-cache-mib auto by @danielhanchen in https://github.com/unslothai/unsloth/pull/12951
* Unsloth Studio / Desktop: export image LoRAs and pick them up again from a custom models folder by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/13146
* Chunked mask test: unwrap every layer, including unsloth_zoo's mask wrapper by @danielhanchen in https://github.com/unslothai/unsloth/pull/13248
* Studio: keep exponents when the model reads a web page by @NilayYadav in https://github.com/unslothai/unsloth/pull/13183
* Chunked mask test: drop the now-unused _swap_function_references import by @danielhanchen in https://github.com/unslothai/unsloth/pull/13257
* README: point notebook-only Docker users at the core tag again by @danielhanchen in https://github.com/unslothai/unsloth/pull/13258
* Studio: keep the sidebar More flyout shut when a modal opens during its hover delay by @danielhanchen in https://github.com/unslothai/unsloth/pull/13237
* Unsloth Studio (AMD, Windows): don't treat a GPU whose multi-arch kernel pack is missing as covered by @danielhanchen in https://github.com/unslothai/unsloth/pull/13231
* Tests: fix three checks that fail for reasons other than a regression by @danielhanchen in https://github.com/unslothai/unsloth/pull/13265
* Studio: add a tool output length setting to Settings > Chat by @danielhanchen in https://github.com/unslothai/unsloth/pull/13228
* Studio installer: drop a flash-attn built for a torch that is gone, and pair torch 2.9 with torchao 0.15.0 by @alkinun in https://github.com/unslothai/unsloth/pull/13247
* Disconnect tests: give the hanging client seconds, not 0.2 s, to be reached by @danielhanchen in https://github.com/unslothai/unsloth/pull/13272
* OpenClaw 2026.10.1: define the CI agent under keyed agents.entries by @danielhanchen in https://github.com/unslothai/unsloth/pull/13266
* chore(deps): apply six validated Studio dependency bumps by @wasimysaid in https://github.com/unslothai/unsloth/pull/13148
* Studio: select all and shift-click ranges in the Library by @Padi142 in https://github.com/unslothai/unsloth/pull/13250
* Docs: secure-mode Docker comment also binds JupyterLab to 127.0.0.1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13275
* Studio: confine native dataset drops to the account workspace; cap child API key expiry by @danielhanchen in https://github.com/unslothai/unsloth/pull/13273
* Studio: title refresh follows the connection that answered the chat by @danielhanchen in https://github.com/unslothai/unsloth/pull/13274
* Fix steering follow-ups disappearing when a response is cancelled by @Imagineer99 in https://github.com/unslothai/unsloth/pull/13249
* Studio: pick up a GPU whose driver loads after Studio starts by @danielhanchen in https://github.com/unslothai/unsloth/pull/13232
* Fix fp16 LoRA training with plain transformers.Trainer (fused CE gradients underflowed) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13152
* Packed SDPA rows: one torch varlen_attn call instead of one SDPA call per segment by @danielhanchen in https://github.com/unslothai/unsloth/pull/13221
* Studio vLLM: allow Triton non-constexpr globals so DeepSeek-OCR loads by @danielhanchen in https://github.com/unslothai/unsloth/pull/13264
* Keep the padding mask for left-padded label-less training forwards by @danielhanchen in https://github.com/unslothai/unsloth/pull/13154
* Skip the fused LoRA kernels when a DoRA adapter is active by @vineethsaivs in https://github.com/unslothai/unsloth/pull/13254
* Pass GradScaler to GKD chunked JSD with an older-Zoo fallback by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/13190
* Treat Docker on WSL2 as WSL when deciding embedding offload by @danielhanchen in https://github.com/unslothai/unsloth/pull/13259
* Run the RoPE and NF4 Triton kernels on GPUs older than Volta by @danielhanchen in https://github.com/unslothai/unsloth/pull/13153
* Studio: stop charging SWA context checkpoints against VRAM on discrete GPUs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13242
* feat(api-monitor): add Load button to API page toolbar by @michupeel in https://github.com/unslothai/unsloth/pull/10378
* Studio: check loopback for an existing listener before taking the port by @Nitish-1303 in https://github.com/unslothai/unsloth/pull/11634
* [Bug] Images tab only show Reference/Edit, etc sections if its pinned to the sidebar by @aBiR1D in https://github.com/unslothai/unsloth/pull/12767
* Studio: name a missing GGUF metadata key instead of blaming the file or memory by @danielhanchen in https://github.com/unslothai/unsloth/pull/13285
* Installer: name the missing tools, not Debian packages, on non-apt Linux by @danielhanchen in https://github.com/unslothai/unsloth/pull/13284
* Studio: start vLLM on hybrid models whose Mamba state outruns max_num_seqs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13235
* Studio: keep auto-load and compare panes off the default engine for compressed-tensors, AWQ and GPTQ checkpoints by @danielhanchen in https://github.com/unslothai/unsloth/pull/13229
* Studio (Jetson): size llama-server context from system RAM when nvidia-smi reports [N/A] by @danielhanchen in https://github.com/unslothai/unsloth/pull/13286
* Studio: keep a media chat's context usage after a local model switch by @danielhanchen in https://github.com/unslothai/unsloth/pull/13281
* Studio: let an MCP call run again after other work changed state by @danielhanchen in https://github.com/unslothai/unsloth/pull/13287
* Studio: resolve the FLUX.2-klein 9B base for a local ComfyUI file by @danielhanchen in https://github.com/unslothai/unsloth/pull/13290
* Run packed rows on the transformers modeling path through varlen flash attention by @danielhanchen in https://github.com/unslothai/unsloth/pull/13234
* Studio: name an MLX or managed-engine token count by the id status publishes by @danielhanchen in https://github.com/unslothai/unsloth/pull/13282
* Studio: install and run vLLM on AMD GPUs, on Linux and through WSL on Windows by @oobabooga in https://github.com/unslothai/unsloth/pull/12785
* Studio: bound the ChatGPT subscription and NPU disconnect watcher teardown by @danielhanchen in https://github.com/unslothai/unsloth/pull/13293
* Studio: move chat with files into a panel above the composer by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13078
* Studio: stop dropdowns jumping on open under Reduce Motion by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13238
* Studio: pair Qwen3.5 / 3.6 vision projectors from another quantizer by @danielhanchen in https://github.com/unslothai/unsloth/pull/13288
* Studio: keep the export Save folder input as typed by @jayzhou2309 in https://github.com/unslothai/unsloth/pull/13142
* Studio: shared media bar, Open with apps, steady image lightbox by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13241
* Studio: keep export status polling single-flight by @ousamabenyounes in https://github.com/unslothai/unsloth/pull/10679
* Studio: draw browser menus, tooltips and toasts over a live native page by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13122
* Run GRPO under FSDP2: no inference mode around FSDP, and an optimizer that steps DTensors by @danielhanchen in https://github.com/unslothai/unsloth/pull/13156
* Studio: fix image and video loads on torchao 0.14 (FqnToConfig import error with diffusers 0.41) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13308
* Studio installer (macOS): start the Unsloth Studio.app shortcut natively on Apple Silicon by @danielhanchen in https://github.com/unslothai/unsloth/pull/13305
* Build the prebuilt CUDA 13 kernel wheels for Windows and ARM (win_amd64, linux_aarch64, win_arm64) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13261
* Studio: install prebuilt flash-attn, causal-conv1d and mamba-ssm wheels on Windows and ARM by @danielhanchen in https://github.com/unslothai/unsloth/pull/13262
* Studio: torch 2.13 for new Windows x64 cu130 Python 3.13 installs, existing installs keep their torch by @danielhanchen in https://github.com/unslothai/unsloth/pull/13263
* Studio (Windows): read a runtime lock held by an elevated Unsloth process as busy, not a broken install by @danielhanchen in https://github.com/unslothai/unsloth/pull/13301
* Unsloth Studio: export a trained image LoRA from the Train tab, list exported copies once by @danielhanchen in https://github.com/unslothai/unsloth/pull/13289
* Studio: price the int8 ConvRot text encoder on Apple Silicon when the dense one is still cached by @danielhanchen in https://github.com/unslothai/unsloth/pull/13306
* Studio: fix torch.compile for image generation on AMD (training config leak, Windows without Visual Studio, max tier) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13227
* Studio: reload without MTP when llama-server dies mid-generation on any MTP launch by @danielhanchen in https://github.com/unslothai/unsloth/pull/13316
* Studio: keep llama.cpp's Metal fit inside the GPU wired headroom by @danielhanchen in https://github.com/unslothai/unsloth/pull/13319
* Intel XPU: 4bit loads without a C compiler by falling back to bitsandbytes' PyTorch quantize by @danielhanchen in https://github.com/unslothai/unsloth/pull/13309
* Apply an explicit rope_scaling to the model and to vLLM by @danielhanchen in https://github.com/unslothai/unsloth/pull/13315
* Studio: remember Max Tokens on Max as Max, not as the old context length by @danielhanchen in https://github.com/unslothai/unsloth/pull/13223
* torchao alias tests: stop leaving a fake diffusers in sys.modules by @danielhanchen in https://github.com/unslothai/unsloth/pull/13336
* Studio: pair torch <= 2.8 with torchao 0.15 and gate torchao export on 0.15 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13329
* Studio: gate auth.db and rag.db opens and closes against the SQLite 3.51.0-3.51.1 deadlock by @danielhanchen in https://github.com/unslothai/unsloth/pull/13307
* Studio: keep loop notes ahead of the MCP image envelope by @ItsRoy69 in https://github.com/unslothai/unsloth/pull/11367
* Studio: stop listing Bonsai DSpark drafter files as models by @kevin9327 in https://github.com/unslothai/unsloth/pull/11265
* Studio (AMD/ROCm): use MIOpen immediate mode for image and video convs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13323
* Studio: explain GGUFs the bundled llama.cpp cannot read, and default Ternary Bonsai to its loadable pack by @danielhanchen in https://github.com/unslothai/unsloth/pull/13304
* Keep offloaded embeddings in host RAM when Trainer and accelerate place the model by @danielhanchen in https://github.com/unslothai/unsloth/pull/13320
* Studio: list Kimi K3 for Kimi API keys by @danielhanchen in https://github.com/unslothai/unsloth/pull/13300
* Studio: let a text Expression column in Data Recipes render empty by @danielhanchen in https://github.com/unslothai/unsloth/pull/13296
* Studio: install the Vulkan audio.cpp build on AMD GPUs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13298
* Studio: skip the SNAC detokenize probe when the ids are past the vocab by @danielhanchen in https://github.com/unslothai/unsloth/pull/13297
* Studio: launch q4_0 instead of iq4_nl KV cache on CUDA/HIP GPUs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13330
* Studio: fit an oversized DiffusionGemma GGUF to the GPU in Auto mode by @danielhanchen in https://github.com/unslothai/unsloth/pull/13312
* Studio: let YuE2 finish a song instead of cutting it at the asked length by @danielhanchen in https://github.com/unslothai/unsloth/pull/13346
* Studio: make Search and Code toggle states clear in dark palettes by @umran666 in https://github.com/unslothai/unsloth/pull/12108
* Studio: use a ComfyUI text_encoders file for a local ComfyUI DiT instead of downloading the base repo's encoder by @danielhanchen in https://github.com/unslothai/unsloth/pull/13314
* Studio: classify qwen_image21 GGUFs as text-to-image by @abhishekacharya200 in https://github.com/unslothai/unsloth/pull/13333
* Studio: Recommended Hub filter and chat picker list FP8 / NVFP4 on capable GPUs, GGUF lists every GGUF by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13240
* AsyncGRPOTrainer with fast_inference: train and generate on one GPU with shared weights by @danielhanchen in https://github.com/unslothai/unsloth/pull/13292
* Save LoRA-GA adapters relative to the original base so they reload correctly by @danielhanchen in https://github.com/unslothai/unsloth/pull/13322
* Studio: keep prompt cache on seeded tool-loop continuations by @indrajeetapache in https://github.com/unslothai/unsloth/pull/10707
* Studio: load and run DeepSeek-OCR and DeepSeek-OCR-2 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13310
* fix(studio): expose tool settings for external models by @Hal9999 in https://github.com/unslothai/unsloth/pull/13251
* Studio: stop offscreen reasoning transcripts listening to thread scroll by @danielhanchen in https://github.com/unslothai/unsloth/pull/13046
* Studio: continue LoRA training from a saved adapter checkpoint by @indrajeetapache in https://github.com/unslothai/unsloth/pull/13158
* Keep token fields when DiffusionTrainer converts a config by @vineethsaivs in https://github.com/unslothai/unsloth/pull/13335
* Studio: stop a chat switch mid-reply crashing the app on a stale message index by @danielhanchen in https://github.com/unslothai/unsloth/pull/13352
* Mask packed document boundary labels for TRL older than 0.23.1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13324
* Fall back instead of failing import when torchcodec cannot load its native library by @danielhanchen in https://github.com/unslothai/unsloth/pull/13348
* Keep modules_to_save trainable when reloading a LoRA checkpoint by @danielhanchen in https://github.com/unslothai/unsloth/pull/13303
* Studio: lighter chat with files border in light mode by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13362
* Unsloth Desktop: redact dashed sk- keys, JWTs and JSON credentials in support diagnostics by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/13260
* Studio: bound a live Thinking block to its newest lines by @danielhanchen in https://github.com/unslothai/unsloth/pull/13086
* Studio: show the compaction count in the context meter by @danielhanchen in https://github.com/unslothai/unsloth/pull/13355
* Unsloth Studio / Desktop: colour a companion download's bar by model, text encoder and VAE by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/13112
* Studio: make the web UI installable as a PWA by @danielhanchen in https://github.com/unslothai/unsloth/pull/13360
* Fix Q-GaLore checkpoint weights and stacked MoE grouping by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/13276
* Never train with a KV cache, so padding-free rows keep their document boundaries by @danielhanchen in https://github.com/unslothai/unsloth/pull/13177
* Studio: keep a wrapper-less Gemma value a string unless JSON reads it as a number by @MohammadHijjawi97 in https://github.com/unslothai/unsloth/pull/13279
* Train embed_tokens and lm_head in FastModel after add_new_tokens by @danielhanchen in https://github.com/unslothai/unsloth/pull/13057
* Studio: run React and TypeScript code blocks as live previews by @Etherll in https://github.com/unslothai/unsloth/pull/13039
* Studio: stop refusing C:\path\app.exe as the blocked command '.' by @danielhanchen in https://github.com/unslothai/unsloth/pull/13347
* AsyncGRPO vLLM server: raise the listen backlog past socketserver's 5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13381
* Studio: upload files straight into a chat's folder by @danielhanchen in https://github.com/unslothai/unsloth/pull/13366
* studiobench: carry on with image_upload when the click opened the menu but timed out by @danielhanchen in https://github.com/unslothai/unsloth/pull/13377
* Studio: Deep Research Retry continues with the current time budget by @danielhanchen in https://github.com/unslothai/unsloth/pull/13358
* Studio Hub: filter models by parameter count, context length and publisher by @danielhanchen in https://github.com/unslothai/unsloth/pull/13363
* Keep a user-set UNSLOTH_RETURN_LOGITS=1 across trainer.train() by @danielhanchen in https://github.com/unslothai/unsloth/pull/13311
* Studio: let LAN access bind only the addresses you choose by @KafKafrnZ in https://github.com/unslothai/unsloth/pull/12839
* Studio: per-account model allowlist for multi-user installs by @danielhanchen in https://github.com/unslothai/unsloth/pull/13359
* Studio: optional detailed Live monitor with GPU load, temperature and power by @danielhanchen in https://github.com/unslothai/unsloth/pull/13351
* Studio: add an embedding device setting for RAG by @danielhanchen in https://github.com/unslothai/unsloth/pull/13365
* Studio: torch 2.13 for new Linux aarch64 cu130 Python 3.13 installs, existing installs keep their torch by @danielhanchen in https://github.com/unslothai/unsloth/pull/13380
* Studio: forward llama-server /metrics, /slots, /tokenize and /detokenize by @danielhanchen in https://github.com/unslothai/unsloth/pull/13354
* Studio: add an opt-in setting to read replies aloud automatically by @danielhanchen in https://github.com/unslothai/unsloth/pull/13357
* Forward output head bias in GRPO policy scoring by @taking-lying-flat in https://github.com/unslothai/unsloth/pull/13302
* Revert "Studio: keep the composer expand button in the top corner when a file is attached" by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13394
* Studio: classify the 10 new audio.cpp-gguf folders as needing a newer runtime by @danielhanchen in https://github.com/unslothai/unsloth/pull/13393
* Studio: screen a quoted program path that cmd /c or start launches by @danielhanchen in https://github.com/unslothai/unsloth/pull/13386
* Studio: Working... placeholder with a soft glint by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13382
* Tests: keep five tests off what the host has installed or downloaded by @danielhanchen in https://github.com/unslothai/unsloth/pull/13395
* Honor padding masks in Llama and Qwen3 cached attention by @Arthur031221 in https://github.com/unslothai/unsloth/pull/13256
* Studio: hide the response-details badge without a fade by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/13397
* Switch between training and inference mode in one tree walk by @danielhanchen in https://github.com/unslothai/unsloth/pull/13374
* Studio: let a project run its tools in an existing local folder by @danielhanchen in https://github.com/unslothai/unsloth/pull/13353
* Perf/scroll performance by @Sneakr in https://github.com/unslothai/unsloth/pull/13255
* Studio: let a python/terminal call ask for a lower output cap by @csarushan1729 in https://github.com/unslothai/unsloth/pull/10398
* feat(dataprep): add audit_supervision, a pre-flight report of what an SFT dataset actually supervises by @RubenHaisma in https://github.com/unslothai/unsloth/pull/10852
* Keep packed documents apart on Unsloth's own models (flash path, eval, cached eval) and match transformers' Mistral window by @danielhanchen in https://github.com/unslothai/unsloth/pull/13313
* Sandbox delete test: only the chat's detached tree counts as the delete by @danielhanchen in https://github.com/unslothai/unsloth/pull/13403
* Studio: sd.cpp on Macs, CPU backend on a paravirtual GPU and no direct conv on Metal by @danielhanchen in https://github.com/unslothai/unsloth/pull/13369
* Studio: refuse LTX-2.5 instead of loading it as LTX-2 by @danielhanchen in https://github.com/unslothai/unsloth/pull/13370
* Studio MPS: opt-in fp16 compute with fixed headroom for Z-Image by @danielhanchen in https://github.com/unslothai/unsloth/pull/13375
* Studio: judge hosted pre-quant checkpoints before downloading them, and make the int8 fingerprint torchao-version-neutral by @danielhanchen in https://github.com/unslothai/unsloth/pull/13339
* Studio: refresh the context meter when a tool loop hands off to a tool by @danielhanchen in https://github.com/unslothai/unsloth/pull/13356
* Studio: opt-in MLX image engine for Apple Silicon (Z-Image) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13379
* Allow offload_layers = "auto" under a DDP launch by @danielhanchen in https://github.com/unslothai/unsloth/pull/13364
* Opt-in DDP batched_grad_copy (UNSLOTH_DDP_BATCHED_GRAD_COPY=1) by @danielhanchen in https://github.com/unslothai/unsloth/pull/13350
* Studio: document the intentional torch and torchcodec floor gap by @RudraBhaskar9439 in https://github.com/unslothai/unsloth/pull/10635
* Studio: Wan2.1 T2V 1.3B and 14B video families by @danielhanchen in https://github.com/unslothai/unsloth/pull/13376
* Packed rows on sliding-window layers take varlen on the transformers path, plus opt-in FA4 / FA3 for packed training by @danielhanchen in https://github.com/unslothai/unsloth/pull/13326
* FastLanguageModel.for_training: undo the UNSLOTH_RETURN_LOGITS=1 that generate() forces by @danielhanchen in https://github.com/unslothai/unsloth/pull/13392

## New Contributors
* @SulimanAbdulrazzaq made their first contribution in https://github.com/unslothai/unsloth/pull/12011
* @frostyautumnleaf made their first contribution in https://github.com/unslothai/unsloth/pull/11685
* @michupeel made their first contribution in https://github.com/unslothai/unsloth/pull/10378
* @Nitish-1303 made their first contribution in https://github.com/unslothai/unsloth/pull/11634
* @aBiR1D made their first contribution in https://github.com/unslothai/unsloth/pull/12767
* @kevin9327 made their first contribution in https://github.com/unslothai/unsloth/pull/11265
* @abhishekacharya200 made their first contribution in https://github.com/unslothai/unsloth/pull/13333
* @Hal9999 made their first contribution in https://github.com/unslothai/unsloth/pull/13251
* @Arthur031221 made their first contribution in https://github.com/unslothai/unsloth/pull/13256
* @csarushan1729 made their first contribution in https://github.com/unslothai/unsloth/pull/10398
* @RubenHaisma made their first contribution in https://github.com/unslothai/unsloth/pull/10852

**Full Changelog**: https://github.com/unslothai/unsloth/compare/v0.1.905-beta...v0.1.906-beta