Unsloth

GLM 5.2 GGUFs are now supported in Unsloth Studio! All reasoning levels supported. 3x longer context lengths are now achievable with our new auto fit algorithm with MTP, allowing longer chats. Bypass permissions mode, forkable chats, queue-able chats, a new hub for model discovery, parallel modules + HTTPS Cloudflare support and more! Use unsloth studio --secure for secure HTTPS global access!

To update Unsloth or install a new Unsloth Studio, you must use the below.
Ensure your version is 2026.6.8 or v0.1.47-beta for the latest.

MacOS, Linux, WSL:

curl -fsSL https://unsloth.ai/install.sh | sh

Windows:

irm https://unsloth.ai/install.ps1 | iex

Better context length algorithm

As per https://github.com/unslothai/unsloth/pull/6312 and https://github.com/unslothai/unsloth/pull/6447, we made Unsloth Studio's determination of memory usage and context length much better, achieving 3x longer context overall:
scenario | KV | before | after
-- | -- | -- | --
1x 32GB pipeline (~31 GB free) | f16 | 23,040 | 64,000
  | q8_0 | 43,520 | 114,944
  | q4_0 | 82,432 | 199,680
2x 32GB pipeline | any | 262,144 | 262,144
2x 24GB tensor (~23 GB free) | f16 | 134,049 | 262,144
  | q8_0 | 252,329 | 262,144

Chat Canvas, Forking & Queueing

• Edit assistant messages in place and re-run from any point in the thread.
• Fork a thread to branch a conversation without losing the original.
• Temporary (incognito) chats that leave nothing behind.
• Queue new prompts while a generation is still running instead of waiting.
• Chat "artifacts" are now canvas, with inline HTML canvas cards that auto-render, a Code view, and DiffusionGemma keeps its raw code visible inline instead of collapsing.
• Chat search now covers every message and surfaces your own messages first.

Hub (Redesigned)

• Full-page Hub with a trending feed, search, and custom model paths support.
• README preview in a split-view feed so you can read before you download.
• Downloads default to the faster Xet transport, with automatic HTTP fallback if a transfer stalls.
• New "Load on selection" toggle to set load options before a model loads.
• Google logo shown for DiffusionGemma and future Gemma derivatives.

Models & Inference

• DeepSeek-OCR and more vision models now load and run without errors.
• Fixed fast inference on the latest vLLM (0.22+) so speed-ups work again.
• Tensor parallelism is more reliable: if the faster MTP path fails, it now recovers on its own instead of crashing.
• DiffusionGemma now shows the image forming live as it denoises, with accurate speed stats.

Security & Cloudflare Encrypted Studios

• New --secure Cloudflare-only mode for end-to-end encrypted studios, with server-side tools staying enabled under --secure. Use unsloth studio --secure!
• Bypass Permissions mode to skip confirmations and disable the tool sandbox when you want it.
• Auto detect Hugging Face Virus scanning + dangerous files in repos.

Logging and API

• New API server monitor in Studio.
• Faster API calling and less latency
• Much better streamlined logs - now with throughput and latency and removed a lot of bloated logs.

Hardware & Backend

• Better support for Blackwell RTX 50X and 60X GPUs
• Fix silent downgrading to CPU and not GPU
• torchao version is now selected from the installed torch.
• Installer now auto-repairs a broken or CPU-only PyTorch install and warns on silent CPU fallback, across NVIDIA + AMD on Win/Linux/Mac/WSL.
• Frees the chat model's VRAM when training starts, but only when the GPU is actually tight (no needless reloads otherwise).
• If llama-server hard-crashes at startup, Studio now steps through a recovery ladder instead of just failing.

Training & General Fixes & Parallel Modules

• MLX training updates.
• Improved GRPO training reliability with vLLM.
• Training startup made more reliable, with clearer errors for invalid VLM batches.
• Studio now cleans up leftover backend processes more reliably after crashes, restarts, or interrupted shutdowns.
• Export, Chat, Training, Recipes are all individualized / compartmentalized! This means you can do all 4 in parallel now! You can chat / do inference while you wait for a training run or an export!

What's Changed

• Bump install.sh / install.ps1 pin to unsloth>=2026.6.4 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6257
• Studio: account for mmproj VRAM in GGUF fit budget (#5825) by @hoobnn in https://github.com/unslothai/unsloth/pull/5849
• fix(studio): keep local GGUF vision on llama-server by @alkinun in https://github.com/unslothai/unsloth/pull/5770
• install.sh: keep the studio launch from draining the curl | sh script (WSL/dash) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6258
• DiffusionGemma: set UNSLOTH_IS_PRESENT so the shim runs on a clean install by @danielhanchen in https://github.com/unslothai/unsloth/pull/6259
• studio: add keyboard navigation to model picker by @alkinun in https://github.com/unslothai/unsloth/pull/5628
• Bump install.sh / install.ps1 pin to unsloth>=2026.6.5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6260
• Installer: drop the lemonade ROCm fallback now the fork ships identical per-gfx prebuilts by @oobabooga in https://github.com/unslothai/unsloth/pull/6225
• Studio: keep distinct bpw flavors of the same GGUF quant by @bouclem in https://github.com/unslothai/unsloth/pull/5729
• studio: declare UNSLOTH_IS_PRESENT at backend startup (clean-install + Windows) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6262
• Studio: extend llama.cpp first-token timeout by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5841
• Studio: polish update banner layout and sidebar settings icon by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6266
• Studio: only advertise a Cloudflare tunnel once it actually serves by @oobabooga in https://github.com/unslothai/unsloth/pull/6264
• Studio: backfill the DiffusionGemma visual-server on a tag-matching update by @danielhanchen in https://github.com/unslothai/unsloth/pull/6267
• Studio: keep llama-server discovery from crashing on an access-denied candidate by @danielhanchen in https://github.com/unslothai/unsloth/pull/6268
• Bump install.sh / install.ps1 pin to unsloth>=2026.6.6 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6270
• Tidy update banner and auth button spacing by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6279
• Fix llama.cpp prebuilt: skip already-installed same-release fallback by @danielhanchen in https://github.com/unslothai/unsloth/pull/6285
• Installer: drop redundant -WindowStyle Hidden from the Windows launcher VBS by @danielhanchen in https://github.com/unslothai/unsloth/pull/6284
• Fix Responses tool output content arrays by @alkinun in https://github.com/unslothai/unsloth/pull/6287
• Studio: UI polish for sidebar, menus, hub and toasts by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6288
• Upgrade setuptools and wheel in the auto-install command by @danielhanchen in https://github.com/unslothai/unsloth/pull/6282
• Studio: fix training output dir escaping outputs root for models on another drive by @danielhanchen in https://github.com/unslothai/unsloth/pull/6293
• Studio: offer llama.cpp update for same-base mix builds on source installs by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6280
• Studio: resolve studio home before the llama-only setup split by @danielhanchen in https://github.com/unslothai/unsloth/pull/6289
• Studio: force-refresh the llama.cpp update check so new builds are not masked by the 24h cache by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6278
• Studio: decide diffusion routing before the SWA resolver by @danielhanchen in https://github.com/unslothai/unsloth/pull/6299
• Bump install.sh / install.ps1 pin to unsloth>=2026.6.7 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6301
• Studio: don't silently fall back to a CPU prebuilt on NVIDIA Linux GPU hosts by @oobabooga in https://github.com/unslothai/unsloth/pull/6310
• Studio: clarify llama.cpp update banner copy by @danielhanchen in https://github.com/unslothai/unsloth/pull/6313
• MLX Training updates by @mmathew23 in https://github.com/unslothai/unsloth/pull/5656
• Studio: add temporary (incognito) chat by @oobabooga in https://github.com/unslothai/unsloth/pull/5956
• Studio: enable stdio MCP servers on a loopback bind by @oobabooga in https://github.com/unslothai/unsloth/pull/6295
• fix(studio): Windows GGUF cancel hang + CPU spinlock overhead (#5692) by @anmolxlight in https://github.com/unslothai/unsloth/pull/5749
• feat: add Anthropic-compatible thinking parameter by @maattm in https://github.com/unslothai/unsloth/pull/5856
• Studio: Bypass Permissions (skip confirmation, disable tool sandbox) by @danielhanchen in https://github.com/unslothai/unsloth/pull/5895
• Studio: add --secure Cloudflare-only mode and revamp API usage examples by @danielhanchen in https://github.com/unslothai/unsloth/pull/6300
• Studio: arm the VRAM-settle wait after the startup orphan reaper by @danielhanchen in https://github.com/unslothai/unsloth/pull/6315
• Studio: fix Mac IME input-method switch leaving composer Send disabled by @narakai in https://github.com/unslothai/unsloth/pull/5762
• fix: use partial hipinfo output on crash to avoid CPU fallback (RDNA 4 / gfx1200) by @mvanhorn in https://github.com/unslothai/unsloth/pull/6292
• Rename chat artifacts copy to canvas by @wasimysaid in https://github.com/unslothai/unsloth/pull/6298
• studio: surface GGUF chat-template alternation errors clearly by @danielhanchen in https://github.com/unslothai/unsloth/pull/5980
• Fix version parsing NameError and older Python syntax compatibility by @umran666 in https://github.com/unslothai/unsloth/pull/6318
• Studio: loading spinner consolidation and sidebar hover pill polish by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6334
• Studio: show llama.cpp version and GPU specs in the About panel by @oobabooga in https://github.com/unslothai/unsloth/pull/6261
• Studio: keep training from failing when a namespace-package shadows unsloth by @danielhanchen in https://github.com/unslothai/unsloth/pull/6269
• feat(studio): expose provider_type selector in model provider dialog by @octo-patch in https://github.com/unslothai/unsloth/pull/4277
• Fix the libaray path for probe_server_capabilities() by @huaj1ng in https://github.com/unslothai/unsloth/pull/5797
• feat: implement thread forking functionality with associated database… by @Erildo in https://github.com/unslothai/unsloth/pull/5810
• Studio: rewrite logging middleware as pure ASGI by @wasimysaid in https://github.com/unslothai/unsloth/pull/6337
• Studio: queue chat prompts while generation is running by @Imagineer99 in https://github.com/unslothai/unsloth/pull/6244
• Studio: clearer error when adding a stdio (local command) MCP server by @NilayYadav in https://github.com/unslothai/unsloth/pull/6341
• Studio: add Studio version info to General settings by @Imagineer99 in https://github.com/unslothai/unsloth/pull/5675
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in https://github.com/unslothai/unsloth/pull/6343
• Studio: make lifespan shutdown resilient to a dead default executor by @danielhanchen in https://github.com/unslothai/unsloth/pull/6307
• Studio: warn when a GPU model silently loaded on CPU by @danielhanchen in https://github.com/unslothai/unsloth/pull/6339
• Studio: improve logging for dynamic transformers version switching by @danielhanchen in https://github.com/unslothai/unsloth/pull/6108
• Chat search: search all messages, user messages first by @danielhanchen in https://github.com/unslothai/unsloth/pull/6350
• Studio: omit --threads when unset so llama.cpp picks physical cores by @danielhanchen in https://github.com/unslothai/unsloth/pull/5894
• Package scanners: cut false positives and make the CI gate blocking by @danielhanchen in https://github.com/unslothai/unsloth/pull/6355
• CLI: fix --local-dataset being parsed as a string instead of a list by @NilayYadav in https://github.com/unslothai/unsloth/pull/6357
• Fix GGUF variant file selection by @wasimysaid in https://github.com/unslothai/unsloth/pull/6342
• ci(studio): fix Linux tool-calling flake (Q4_K_XL) and capture server logs on 500 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6360
• Make the Studio installers (sh + ps1) resilient to transient uv download failures by @danielhanchen in https://github.com/unslothai/unsloth/pull/6281
• Studio: remove the Windows VBS launcher to clear the Kaspersky false positive by @danielhanchen in https://github.com/unslothai/unsloth/pull/6326
• Harden Trainer._load_rng_state against malicious checkpoints (CVE-2026-1839) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6351
• Studio: hide the llama-server install validation probe from model pickers by @danielhanchen in https://github.com/unslothai/unsloth/pull/6366
• Studio: Xet-primary model downloads with automatic HTTP fallback on stall by @danielhanchen in https://github.com/unslothai/unsloth/pull/6372
• Add API server monitor in Studio by @alkinun in https://github.com/unslothai/unsloth/pull/5558
• Studio: HTML canvas cards in chat with auto-render, a Code view, and visible diffusion code by @danielhanchen in https://github.com/unslothai/unsloth/pull/6374
• fix(unsloth-cli): route hub_path/hub_token correctly in --push_model save block by @Anai-Guo in https://github.com/unslothai/unsloth/pull/6346
• studio: deterministic VRAM auto-fit for GGUF (MTP reserve, compute buffer, total-based budget) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6312
• Studio: surface the real reason a model fails validation by @danielhanchen in https://github.com/unslothai/unsloth/pull/6398
• studio: select torchao version from the installed torch by @danielhanchen in https://github.com/unslothai/unsloth/pull/6400
• feat(hub): full-page redesign with trending feed, search, and persisted state by @Sneakr in https://github.com/unslothai/unsloth/pull/6349
• Studio: trim serving-log noise and surface llama-server engine stats by @danielhanchen in https://github.com/unslothai/unsloth/pull/6377
• Studio Hub: preview first README in split view and round status chips into pills by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6404
• Studio: add 'Load on selection' toggle to configure load options before loading by @oobabooga in https://github.com/unslothai/unsloth/pull/6348
• Feature: implement editable assistant messages by @CelesteHeartsong in https://github.com/unslothai/unsloth/pull/6397
• Studio: follow OS theme by default and apply it on mount by @Etherll in https://github.com/unslothai/unsloth/pull/6405
• CLI: add unsloth connect to point coding agents at a local Studio server by @NilayYadav in https://github.com/unslothai/unsloth/pull/6407
• Shim removed vllm.transformers_utils.tokenizer so fast_inference works on vLLM >= 0.22 by @GodlyDonuts in https://github.com/unslothai/unsloth/pull/6390
• Studio: stop the llama.cpp update banner flickering and show the download size by @danielhanchen in https://github.com/unslothai/unsloth/pull/6338
• studio: fix tests turning main CI red/flaky (kill-process, install overrides, UI re-login) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6419
• Studio: prefer native cuda13 over torch's cuda12 line on Blackwell Linux hosts by @danielhanchen in https://github.com/unslothai/unsloth/pull/6379
• Studio: convert SecurityHeadersMiddleware to pure ASGI by @danielhanchen in https://github.com/unslothai/unsloth/pull/6394
• Studio: serialize non-streaming responses once and pool the proxy client by @danielhanchen in https://github.com/unslothai/unsloth/pull/6393
• Studio: pin CUDA_DEVICE_ORDER=PCI_BUS_ID and list GPUs at startup by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/6353
• test: regression guards for SecurityHeadersMiddleware pure-ASGI by @danielhanchen in https://github.com/unslothai/unsloth/pull/6424
• Studio: make toast text selectable by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6423
• Fix _kill_process AttributeError when _stats_logger is unset by @danielhanchen in https://github.com/unslothai/unsloth/pull/6417
• Reduce and tighten comments and docstrings across the test suite by @danielhanchen in https://github.com/unslothai/unsloth/pull/6429
• Studio: harden GPU startup detection and clarify multi-GPU listing by @danielhanchen in https://github.com/unslothai/unsloth/pull/6427
• Studio: make sidebar nav hover a fully rounded pill by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6435
• Studio Hub: show Google logo for diffusiongemma and future gemma derivatives by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6432
• Studio Hub: default downloads to Xet transport by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6433
• Harden model fetching by @danielhanchen in https://github.com/unslothai/unsloth/pull/6391
• Runtime MTP fallback for tensor parallelism (try MTP, recover if it crashes) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6324
• Studio: scale export GGUF size estimates from the real model size by @danielhanchen in https://github.com/unslothai/unsloth/pull/6418
• Reap Studio child processes when the parent dies abnormally by @danielhanchen in https://github.com/unslothai/unsloth/pull/6425
• Studio: cross-session backstop to reap a leftover llama-server on startup by @danielhanchen in https://github.com/unslothai/unsloth/pull/6431
• Keep server-side tools enabled under --secure by @danielhanchen in https://github.com/unslothai/unsloth/pull/6403
• Studio: reach the published source asset when a mix build's commit 404s by @danielhanchen in https://github.com/unslothai/unsloth/pull/6314
• Studio: show an actionable message when the GGUF runtime is missing by @danielhanchen in https://github.com/unslothai/unsloth/pull/6327
• Fix: scan_packages.py --fix crash on download_packages() tuple return by @parveshsaini in https://github.com/unslothai/unsloth/pull/6413
• Align GRPO vllm_enable_sleep_mode with the engine's actual sleep state by @danielhanchen in https://github.com/unslothai/unsloth/pull/6420
• Studio: fail fast on an invalid first training batch (base VLM empty chat template) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6358
• Package scanners: close fail-open gaps in the sdist fallback and hidden-payload paths by @danielhanchen in https://github.com/unslothai/unsloth/pull/6359
• Load DeepSeek-OCR and other VLMs that register AutoModel in auto_map by @danielhanchen in https://github.com/unslothai/unsloth/pull/6421
• Pin unsloth-zoo>=2026.6.5 in install.sh / install.ps1 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6440
• Studio: square off the sidebar profile button hover so it isn't a pill by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6443
• Studio: polish Bypass permissions toggle and add it to chat menu settings by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6442
• Installer: repair stale/CPU-only PyTorch and warn on silent CPU fallback (NVIDIA + AMD, Win/Linux/Mac/WSL) by @danielhanchen in https://github.com/unslothai/unsloth/pull/5942
• Studio: free chat model VRAM at training start only when the GPU is tight by @danielhanchen in https://github.com/unslothai/unsloth/pull/6243
• Studio: graceful recovery ladder when llama-server hard-crashes at startup by @danielhanchen in https://github.com/unslothai/unsloth/pull/6291
• Studio: restore sidebar roundness to its pre-#6349 state by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6446
• Studio: Bypass Permissions menu fix, decimal GB sizes, and GLM-5.2 high/max/disabled thinking by @danielhanchen in https://github.com/unslothai/unsloth/pull/6444
• Studio: reserve the duplicated MTP target KV context for MLA models (GLM-5.2 OOM) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6447
• diffusion_studio: split shim thought channels into reasoning_content by @raydeStar in https://github.com/unslothai/unsloth-zoo/pull/769
• Use packaging.version to compare numpy versions by @ccoulombe in https://github.com/unslothai/unsloth-zoo/pull/753
• MLX Update Training by @mmathew23 in https://github.com/unslothai/unsloth-zoo/pull/684
• Guard mlx import in test_qwen35_vjp_metal so Linux collection skips cleanly by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/775
• fix(mlx): wrong gated-delta grads for batch rows past the first by @Lyxot in https://github.com/unslothai/unsloth-zoo/pull/776
• test(mlx): batched gradient-parity regression for gated-delta VJP by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/778
• fix(full-FT): upcast LayerNorm weights to fp32 for bf16 training by @mmathew23 in https://github.com/unslothai/unsloth-zoo/pull/680
• Tighten comments in bf16 full-FT norm upcast code by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/780
• fix(diffusion_studio): add torch/lib to PATH on Windows so visual-server loads CUDA backend by @Anai-Guo in https://github.com/unslothai/unsloth-zoo/pull/770
• fix(diffusion): load bundled CUDA runtime + auto-size canvas for DiffusionGemma on Linux/WSL2 by @ThrownLemon in https://github.com/unslothai/unsloth-zoo/pull/771
• fix(tests): import unsloth_zoo (not unsloth) in fresh-interpreter pickle tests by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/781
• Fix harmony analysis channel using thinking instead of content by @mvanhorn in https://github.com/unslothai/unsloth-zoo/pull/772
• Fix Gemma-4 26B-A4B MoE LoRA merge dropping expert deltas by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/779
• Fix fast_inference crash on vLLM >= 0.22 (removed transformers_utils.tokenizer) by @GodlyDonuts in https://github.com/unslothai/unsloth-zoo/pull/783
• Optimize LoRA merge to 16bit save and add end-to-end merge correctness suite by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/777
• fix(mlx): enforce save_total_limit checkpoint rotation by @Lyxot in https://github.com/unslothai/unsloth-zoo/pull/782
• fix(mlx): cast Gemma3 language MLP activation to fp32 to prevent backward NaN by @BardiaKoopah in https://github.com/unslothai/unsloth-zoo/pull/785
• fix(mlx): keep Qwen3-VL vision MLP fp32 when activation dtype is fp16 by @BardiaKoopah in https://github.com/unslothai/unsloth-zoo/pull/787
• Fix: handle key format mismatch for VLM models (Ministral-3-14B) by @anmolxlight in https://github.com/unslothai/unsloth-zoo/pull/773
• Fix LoRA fast_inference on new vLLM: get_dummy_lora_warmup_rank + compile downgrade by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/788
• test(mlx): verify VLM resume_from_checkpoint matches fresh run step-for-step by @BardiaKoopah in https://github.com/unslothai/unsloth-zoo/pull/786
• fix(mlx): align response-mask batching with CUDA by @Lyxot in https://github.com/unslothai/unsloth-zoo/pull/784
• Fix vision GRPO under vLLM standby, bundle converter resolution, DeepseekV2MoE rename by @danielhanchen in https://github.com/unslothai/unsloth-zoo/pull/768

New Contributors

• @hoobnn made their first contribution in https://github.com/unslothai/unsloth/pull/5849
• @bouclem made their first contribution in https://github.com/unslothai/unsloth/pull/5729
• @maattm made their first contribution in https://github.com/unslothai/unsloth/pull/5856
• @narakai made their first contribution in https://github.com/unslothai/unsloth/pull/5762
• @umran666 made their first contribution in https://github.com/unslothai/unsloth/pull/6318
• @huaj1ng made their first contribution in https://github.com/unslothai/unsloth/pull/5797
• @Erildo made their first contribution in https://github.com/unslothai/unsloth/pull/5810
• @CelesteHeartsong made their first contribution in https://github.com/unslothai/unsloth/pull/6397
• @GodlyDonuts made their first contribution in https://github.com/unslothai/unsloth/pull/6390
• @parveshsaini made their first contribution in https://github.com/unslothai/unsloth/pull/6413

Full Changelog: https://github.com/unslothai/unsloth/compare/v0.1.46-beta...v0.1.47-beta