v0.1.37-beta

Hey guys, we revamped the entire Unsloth Studio UI and UX experience to put an emphasis on chat and training:
* Added a collapsible sidebar based on community feedback <br><img width="500" alt="new ui" src="https://github.com/user-attachments/assets/1f7326c3-347a-4ebf-b2a7-43e26ebbf1ab" />
* You can now delete chats and search past conversations<br><img width="400"  alt="Delete messages" src="https://github.com/user-attachments/assets/a9268174-ab80-477e-870d-142d7cb9305b" /> <img width="300" alt="search chats" src="https://github.com/user-attachments/assets/a1e795ea-a645-4c9f-bd9f-69ea54229514" />
* New Preserve Thinking toggle for models that support it like Qwen3.6
* Cleaner, more consistent design with easier navigation
* Expanded Settings page with options to change your profile picture, name, and more<br><img width="350" alt="change profile settings" src="https://github.com/user-attachments/assets/4c784adf-fa90-4e5c-a02f-755cefcabe27" />
* No more entering your Hugging Face token twice
* gpt-oss now has low, medium and high thinking toggles.
* Now uses latest llama.cpp prebuilt, even on Linux CUDA 
* Lots of bug, consistency and stability fixes
* Kimi-K2.6 can now be run!
* We also added experimental API support. Guides, announcement etc will come next week.

Qwen3.6 was also also previously already supported in Unsloth Studio for running and training. You can train and run Qwen3.6-27B right now!

## What's Changed
* Only run ldconfig CUDA-linking recovery when we have permission by @danielhanchen in https://github.com/unslothai/unsloth/pull/4930
* Fix Mistral DPO/preference training crash on non-xformers platforms (e.g. Intel XPU) by @cheehook in https://github.com/unslothai/unsloth/pull/4889
* Fix raw text paragraph break normalization by @kiankyars in https://github.com/unslothai/unsloth/pull/4884
* Studio: keep chat input visible and fix compare pane clipping by @Imagineer99 in https://github.com/unslothai/unsloth/pull/4924
* fix: check find() return value before adding offset in try_fix_tokenizer by @Ricardo-M-L in https://github.com/unslothai/unsloth/pull/4923
* updated models template mappers. added lfm2.5vl450m to transformers 5… by @rolandtannous in https://github.com/unslothai/unsloth/pull/4939
* Revert "updated models template mappers. added lfm2.5vl450m to transformers 5…" by @rolandtannous in https://github.com/unslothai/unsloth/pull/4945
* Add AMD ROCm/HIP support across installer and hardware detection by @danielhanchen in https://github.com/unslothai/unsloth/pull/4720
* Pin bitsandbytes to continuous-release_main on ROCm (4-bit decode fix) by @danielhanchen in https://github.com/unslothai/unsloth/pull/4954
* Fix Gemma-4 GRPO catastrophic KL divergence with TRL 1.0.0+ by @danielhanchen in https://github.com/unslothai/unsloth/pull/4934
* Add ROCm test suite (companion to #4720) by @danielhanchen in https://github.com/unslothai/unsloth/pull/4824
* updating gemma4 script by @Manan17 in https://github.com/unslothai/unsloth/pull/4992
* Move gemma4 script by @Manan17 in https://github.com/unslothai/unsloth/pull/4994
* studio: fix route transition DOM duplication via AnimatePresence mode="wait" by @AdamPlatin123 in https://github.com/unslothai/unsloth/pull/4987
* Studio: Prompt manager, message deletion, and chat UI improvements by @Imagineer99 in https://github.com/unslothai/unsloth/pull/4938
* Pin kernels==0.12.1 to fix training import failure by @rolandtannous in https://github.com/unslothai/unsloth/pull/5000
* Studio: Expose openai and anthropic compatible external API end points  by @danielhanchen in https://github.com/unslothai/unsloth/pull/4956
* studio: skip training status/metrics polling when idle by @AdamPlatin123 in https://github.com/unslothai/unsloth/pull/4988
* studio: fix api-keys access + refresh by @wasimysaid in https://github.com/unslothai/unsloth/pull/5005
* Studio: Polish API key copy button and harden async clipboard fallback by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5006
* fix(studio): default chart view to full training history by @Barath19 in https://github.com/unslothai/unsloth/pull/5007
* [Studio] Show non exported models in chat UI by @Datta0 in https://github.com/unslothai/unsloth/pull/4892
* [Studio] Install flash attn at setup time for linux by @Datta0 in https://github.com/unslothai/unsloth/pull/4979
* fix(studio): remove 300s cap on load_checkpoint (inherits 3600s default) by @TF-MTGE in https://github.com/unslothai/unsloth/pull/4922
* Studio: honor explicit GGUF ctx and default to 4096 when weights exceed VRAM by @danielhanchen in https://github.com/unslothai/unsloth/pull/5011
* Studio: make GGUF disk-space preflight cache-aware by @danielhanchen in https://github.com/unslothai/unsloth/pull/5012
* Studio: anchor ctx-slider warning threshold at 4096 when weights exceed VRAM by @danielhanchen in https://github.com/unslothai/unsloth/pull/5014
* studio: show HF model download progress in training start overlay by @danielhanchen in https://github.com/unslothai/unsloth/pull/4894
* studio: stream export worker output into the export dialog by @danielhanchen in https://github.com/unslothai/unsloth/pull/4897
* Fix num_items_in_batch GA for Gemma4 by @Datta0 in https://github.com/unslothai/unsloth/pull/4998
* studio: pin peft to 0.18.1 to fix export subprocess issues by @rolandtannous in https://github.com/unslothai/unsloth/pull/5015
* Studio: live model-load progress + rate/ETA on download and load by @danielhanchen in https://github.com/unslothai/unsloth/pull/5017
* [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in https://github.com/unslothai/unsloth/pull/5004
* Fix bitsandbytes ROCm install by using pip instead of uv by @edamamez in https://github.com/unslothai/unsloth/pull/4966
* Studio: split model-load progress label across two rows by @danielhanchen in https://github.com/unslothai/unsloth/pull/5020
* Studio: hard-stop at n_ctx with a 'Context limit reached' toast by @danielhanchen in https://github.com/unslothai/unsloth/pull/5021
* [moe][gemma4] Target MoE for gemma4 by @Datta0 in https://github.com/unslothai/unsloth/pull/4913
* Add configurable PyTorch mirror via UNSLOTH_PYTORCH_MIRROR env var by @rolandtannous in https://github.com/unslothai/unsloth/pull/5024
* Studio: support GGUF variant selection for non-suffixed repos by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5023
* fix: prevent offline freeze by fixing stats retry and forwarding local_files_only by @DavidSolanas in https://github.com/unslothai/unsloth/pull/5016
* Respect classification head skip list on pre-quantized 4-bit checkpoints (#5027) by @danielhanchen in https://github.com/unslothai/unsloth/pull/5034
* fix(rocm): tighten gfx regex to ignore generic ISA lines by @danielhanchen in https://github.com/unslothai/unsloth/pull/5033
* Fix grad-accum accepts_loss_kwargs detection for vision wrappers by @danielhanchen in https://github.com/unslothai/unsloth/pull/5036
* grpo_compute_loss_slow called with wrong positional args by @jonahsamost in https://github.com/unslothai/unsloth/pull/4887
* Gate trl disable_gradient_checkpointing warning on UNSLOTH_ENABLE_LOGGING by @danielhanchen in https://github.com/unslothai/unsloth/pull/5038
* Studio: refresh Downloaded GGUF list and recurse into variant subdirs by @danielhanchen in https://github.com/unslothai/unsloth/pull/5032
* feat: Add support for OLMo-3 model by @OnePunchMonk in https://github.com/unslothai/unsloth/pull/4678
* feat: Add cactus QAT scheme support by @OnePunchMonk in https://github.com/unslothai/unsloth/pull/4679
* Re-apply #4939: updated models template mappers by @rolandtannous in https://github.com/unslothai/unsloth/pull/4950
* Studio: add folder browser modal for Custom Folders by @danielhanchen in https://github.com/unslothai/unsloth/pull/5035
* Bump Studio installer minimum to 2026.4.5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/5041
* fix Gemma4 flash attn disable by @mmathew23 in https://github.com/unslothai/unsloth/pull/5045
* BUG: fix _fix_chat_template for ChatML templates missing add_generation_prompt (#4150) by @kimimgo in https://github.com/unslothai/unsloth/pull/4426
* fix: use direct registry API for PATH writes instead of SetEnvironmentVariable by @Etherll in https://github.com/unslothai/unsloth/pull/4961
* Chat-template repair: warn-by-default, AST classification, dict support by @danielhanchen in https://github.com/unslothai/unsloth/pull/5049
* Restrict flash attn to <=256 head dim. Consolidate attn impl checks by @Datta0 in https://github.com/unslothai/unsloth/pull/5051
* Remove legacy venv Scripts entry from User PATH on upgrade by @danielhanchen in https://github.com/unslothai/unsloth/pull/5060
* Fix review findings for chat-template repair (#5049) by @danielhanchen in https://github.com/unslothai/unsloth/pull/5056
* Studio: Ollama support, recommended folders, Custom Folders UX polish by @danielhanchen in https://github.com/unslothai/unsloth/pull/5050
* feat(studio): replace navbar with collapsible sidebar by @wasimysaid in https://github.com/unslothai/unsloth/pull/4936
* fix audio dataset preview and finetuning by @CodeMan62 in https://github.com/unslothai/unsloth/pull/5043
* Chat first onboarding by @wasimysaid in https://github.com/unslothai/unsloth/pull/5063
* Fix onboarding followups by @wasimysaid in https://github.com/unslothai/unsloth/pull/5064
* Studio: Default Gemma fallback for chat + AI assist by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5066
* fix: multi-GPU inference crash for bnb 4-bit/8-bit models by @danielhanchen in https://github.com/unslothai/unsloth/pull/5068
* Add Qwen3.6 inference defaults for Studio by @danielhanchen in https://github.com/unslothai/unsloth/pull/5065
* Add qwen3.6 script by @Manan17 in https://github.com/unslothai/unsloth/pull/5084
* Studio: forward standard OpenAI tools / tool_choice to llama-server by @rolandtannous in https://github.com/unslothai/unsloth/pull/5099
* fix(studio/chat): stop stream when trashing a thread from sidebar by @rolandtannous in https://github.com/unslothai/unsloth/pull/5067
* Studio: Local profile customization in settings and sync sidebar identity by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5088
* Studio: Show LoRA live logs and update GGUF quant options by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5058
* Studio: prefer mainstream clipboard copy over deprecated one by @G07cha in https://github.com/unslothai/unsloth/pull/5109
* Studio: Improve chat composition, fix scroll behaviour, and refine sidebar UX by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5089
* Studio: forward standard OpenAI tools / tool_choice on /v1/responses (Codex compat) by @rolandtannous in https://github.com/unslothai/unsloth/pull/5122
* Studio: support images on /v1/messages (Anthropic-compat) by @rolandtannous in https://github.com/unslothai/unsloth/pull/5128
* Coerce TRL's tuple-cached _*_available flags to bool by @danielhanchen in https://github.com/unslothai/unsloth/pull/5129
* Studio: Smoother thread switching in chat by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5126
* Studio: Replace assistant UI shared autoscroll with per-panel scrolling by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5127
* [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in https://github.com/unslothai/unsloth/pull/5117
* Fix tokenizer save gemma by @Datta0 in https://github.com/unslothai/unsloth/pull/5115
* update gema4 chat templates by @Datta0 in https://github.com/unslothai/unsloth/pull/5116
* Bump installer floor to 2026.4.7 by @danielhanchen in https://github.com/unslothai/unsloth/pull/5134
* fix/llamacpp_prebuilt_install by @mmathew23 in https://github.com/unslothai/unsloth/pull/5135
* Studio: fix stale test_exception_result_cached test for vision cache by @danielhanchen in https://github.com/unslothai/unsloth/pull/5145
* fix: patch CONTROL type for special tokens in sentencepiece GGUF export by @octo-patch in https://github.com/unslothai/unsloth/pull/5080
* fix(install): clear STUDIO_LOCAL_* env on POSIX normal install by @danielhanchen in https://github.com/unslothai/unsloth/pull/5146
* Add tauri by @wasimysaid in https://github.com/unslothai/unsloth/pull/5144
* Studio: detect reasoning_effort and preserve_thinking in chat templates by @danielhanchen in https://github.com/unslothai/unsloth/pull/5149

## New Contributors
* @cheehook made their first contribution in https://github.com/unslothai/unsloth/pull/4889
* @Ricardo-M-L made their first contribution in https://github.com/unslothai/unsloth/pull/4923
* @Barath19 made their first contribution in https://github.com/unslothai/unsloth/pull/5007
* @TF-MTGE made their first contribution in https://github.com/unslothai/unsloth/pull/4922
* @edamamez made their first contribution in https://github.com/unslothai/unsloth/pull/4966
* @DavidSolanas made their first contribution in https://github.com/unslothai/unsloth/pull/5016
* @jonahsamost made their first contribution in https://github.com/unslothai/unsloth/pull/4887
* @kimimgo made their first contribution in https://github.com/unslothai/unsloth/pull/4426
* @CodeMan62 made their first contribution in https://github.com/unslothai/unsloth/pull/5043
* @G07cha made their first contribution in https://github.com/unslothai/unsloth/pull/5109
* @octo-patch made their first contribution in https://github.com/unslothai/unsloth/pull/5080

**Full Changelog**: https://github.com/unslothai/unsloth/compare/v0.1.36-beta...v0.1.37-beta