unslothai/unsloth v0.1.52-beta on GitHub

Hey everyone! For folks who missed the news - Kimi K3 & DeepSeek v4 Flash can now run locally with Unsloth Dynamic GGUFs! We now added more efficient and faster downloading for Colab, low memory systems and also high memory and CPU systems - we auto fallback to HTTP as well if XET is stuck

DeepSeek V4 Flash 0731

You can run our DeepSeek V4 Dynamic GGUFs through Unsloth. Unsloth automatically detects multi-GPU setups and can offload model layers to system memory.

UD-IQ1_S is 83GB in disk space.
UD-IQ3_XXS is 104GB in disk space and fits in 128GB machines.
UD-Q4_K_XL is 155GB in disk space.
For lossless inference, use UD-Q8_K_XL, which is 162TB in disk space.

Read our full DeepSeek V4 Flash guide

Kimi K3

Moonshot AI’s Kimi K3 is a 2.8T-parameter MoE model with 104B active parameters, native vision support and a 1M context window. Kimi K3 is thinking-only and Unsloth supports low, high and max reasoning efforts.

You can run our Kimi K3 Dynamic GGUFs through Unsloth. Unsloth automatically detects multi-GPU setups and offloads layers automatically.

Kimi K3 is a very large model, so plan your hardware accordingly:

UD-IQ1_S is 595GB in disk space.
UD-Q4_K_XL is 1.51TB in disk space.
For lossless inference, use UD-Q8_K_XL, which is 1.56TB in disk space.

Read our full Kimi K3 guide and learn more about Unsloth Dynamic 2.0 GGUFs.

Updating / installing Unsloth

macOS, Linux, WSL:

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

Windows:

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

Unsloth is NOT affected by the Keyv security incident

As per https://www.aikido.dev/blog/keyv-and-friends-compromised-in-npm-supply-chain-attack, Unsloth uses ESLint which uses keyv==4.5.4, flat-cache==4.0.1, file-entry-cache==8.0.0. The compromised versions are keyv==6.0.0, flat-cache==6.1.24, file-entry-cache==11.1.6

Unsloth DOES NOT use the compromised versions, so you do NOT need to take any action.

However please check your own systems and other non Unsloth packages for any compromise - see https://github.com/wiz-sec-public/wiz-research-iocs/blob/main/reports/keyv-packages.csv for compromised packages

What's Changed

Bump install.sh / install.ps1 pin to unsloth>=2026.7.6 by @danielhanchen in #7621
Fix two pre-existing Backend CI failures blocking every open PR by @danielhanchen in #7639
fix(trainer): warn loudly when requested packing is silently disabled by @eliazonta in #7234
Preserve reasoning across turns for Qwen3.x by @mahille in #7289
Make tests/utils runnable again and cover test_packing.py in CI by @danielhanchen in #7642
Uninstaller: reject unrecognized arguments instead of uninstalling by @oobabooga in #7631
Make the uninstaller test suites run on macOS and Windows PowerShell 5.1 by @danielhanchen in #7643
Studio: Don't disarm Ctrl+C in agents launched by unsloth start by @NilayYadav in #7630
Studio: keep a slow model load alive through a proxy timeout by @danielhanchen in #7635
Tighten the comments around the uninstaller argument guard by @danielhanchen in #7647
Route the explicit Vulkan setup failures through the setup failure helpers by @danielhanchen in #7645
Honor an explicit map_eos_token=False in get_chat_template by @vineethsaivs in #7475
Fix desktop repair version skew and output decoding by @wasimysaid in #7641
Desktop: Improve the installation message by @oobabooga in #7657
Studio: make the API keys list scrollable by @NilayYadav in #7622
Desktop: notarize the final macOS disk image on release by @oobabooga in #7615
Desktop: preserve first app window size after setup by @oobabooga in #7659
data-recipe: recognize current sampling progress log line by @devangpratap in #6760
Desktop: clean up the backend on Unix termination signals by @oobabooga in #7614
Desktop: show the app version in Settings > About by @oobabooga in #7612
Studio: Fix hub search input overlapping the format filter at narrow widths by @NilayYadav in #7658
Add 2026.7.6 release notes to CHANGELOG.md by @shimmyshimmer in #7671
Desktop: exit the backend when the app that spawned it dies by @NilayYadav in #7655
Desktop: open uploaded files via the OS instead of window.open by @NilayYadav in #7660
Run OpenClaw in the launched project workspace by @Imagineer99 in #7568
Desktop: drop documents into a chat instead of rejecting them as non-GGUF (#7661) by @LeoBorcherding in #7668
feat(install): use pre-built Tauri icon for macOS .app bundle by @ArjunDivecha in #7628
Studio: report active training status by @oobabooga in #7613
studio: apply saved per-model settings on API loads, add API monitor by @shimmyshimmer in #7473
fix: replace deprecated transformers attention mask imports by @Taranum01 in #6880
Offline: detect an unreachable hub, not just dead DNS by @shimmyshimmer in #7591
Assert the Vulkan installer bailouts fail closed via the setup failure helper by @danielhanchen in #7644
Studio: charge the split compute-buffer rate on a multi-GPU auto fit by @danielhanchen in #7637
Add 2026.7.7 release notes to CHANGELOG.md by @danielhanchen in #7685
Fix multiline prompts through Windows npm shims by @Imagineer99 in #7567
Revert "Add 2026.7.7 release notes to CHANGELOG.md" by @danielhanchen in #7687
docs: fix --quantization flag in unsloth-cli.py usage example by @latent-9 in #7688
Studio: Fix GGUF classification for custom MTP scan roots by @Lyxot in #7648
Studio: complete translation coverage for all supported locales by @oobabooga in #7666
Guard VLM detection against config.architectures being None in the save paths by @vineethsaivs in #7372
Centralize device backend helpers by @danielhanchen in #4252
Make the auto-appended EOS token deterministic by @vineethsaivs in #7702
Unsloth Studio: fix sidebar rail bottom fade rendering as a lighter block in dark mode by @LeoBorcherding in #7674
Fix the three CI failures blocking every open PR by @danielhanchen in #7722
Close two false passes in the uninstaller argument tests, and document the piped help form by @danielhanchen in #7690
Uninstall the whisper.cpp prebuilt and the node/whisper install locks by @danielhanchen in #7686
studio update --local: refuse site-packages as the repo root by @danielhanchen in #7693
studio update: fail instead of reporting success on a damaged install by @danielhanchen in #7691
studio setup: let Windows PowerShell load its own Security module by @danielhanchen in #7692
Fix unsloth run on Windows: the venv entry point is unsloth.exe by @danielhanchen in #7718
Studio: restore cached models hidden by a dangling HF ref, and stop auto-load downloading after a failed load (#7374) by @Souravrajvi0 in #7375
Desktop: stop losing the backend's exit, its output and its startup by @danielhanchen in #7719
Studio setup: report a missing llama.cpp in Tauri mode instead of aborting the install by @danielhanchen in #7720
Studio: honour the GPU-layer split for DiffusionGemma loads by @LeoBorcherding in #7575
Desktop: reopen hidden macOS window from Dock by @wasimysaid in #7723
Fix the six backend tests blocking every PR: companion-only GGUF folders, and vision fixtures on empty projectors by @danielhanchen in #7728
Fix Windows llama.cpp prebuilt setup with inaccessible PATH entries by @Etherll in #7696
Desktop: bypass proxies for loopback requests by @oobabooga in #7709
Desktop: keep Windows updater installer alive by @oobabooga in #7705
Stop and send buttons on the chat dictation bar by @shimmyshimmer in #7683
Repair .venv_t5_* sidecars whose files are damaged by @danielhanchen in #7725
Desktop: fix remaining file downloads by @oobabooga in #7710
Desktop: restore the confirmations that only worked in the browser by @NilayYadav in #7707
Studio: bring the login screen up before the ML stack loads by @danielhanchen in #7607
Lint CI: scope the import-hoist check to the PR's own commits by @danielhanchen in #7738
fix(studio): detect truncated sidecar files via RECORD in _venv_dir_is_valid (#7715) by @Anai-Guo in #7721
Security audit: re-baseline the benign findings that reopened on main by @danielhanchen in #7741
Improve toast layout by @mahiatlinux in #7734
Wait for the floating monitor to stop moving, not just stop growing by @danielhanchen in #7739
Desktop: add Check for Updates and Update Now to Settings by @oobabooga in #7694
Escape caller template text spliced into Jinja string literals by @vineethsaivs in #7731
Fix validate_dataset KeyError on default (tokenized) RawTextDataLoader output by @adinfarel in #7729
Model picker: default Show all quantizations off, collapse single-quant rows, say Reveal in Folder outside macOS by @shimmyshimmer in #7736
Model picker: show row actions on collapsed quant rows and align the GGUF tag by @shimmyshimmer in #7745
Escape the system message spliced into predefined chat templates by @danielhanchen in #7746
Follow the third load site in the pinned-row contract test by @danielhanchen in #7748
Windows: do not abort setup on an unreadable llama.cpp install by @shimmyshimmer in #7735
CI: harden the Windows Pester bootstrap against the nuget.exe PSGallery failure by @danielhanchen in #7743
Windows: move the installer out of System32 instead of failing in it by @danielhanchen in #7744
Align the sidebar nav and Recents spinners on one column by @danielhanchen in #7749
CI: compare the workflow path as posix so the Pester guard passes on Windows by @danielhanchen in #7750
Model picker: remember the Run Settings advanced toggle across models by @shimmyshimmer in #7747
Bump install.sh / install.ps1 pin to unsloth>=2026.8.1 by @danielhanchen in #7754
Model picker: context length hint, and the preset note as a tooltip by @shimmyshimmer in #7755
Studio: simplify the GGUF loading backend by @oobabooga in #7663
Studio: add Italian (it) locale by @LRimes90 in #7650
Forward revision to the config, weight and tokenizer loads by @danielhanchen in #4222
CI: run the Pester guard on Windows without the smoke suite by @danielhanchen in #7753
Support audio input in Studio chat on Apple Silicon by @Lyxot in #7699
Fix formatting_func race in UnslothVisionDataCollator without a lock by @adinfarel in #7698
Windows: guard the source-build and whisper.cpp probes on an unreadable install tree by @danielhanchen in #7757
Stop the atexit teardown printing tracebacks after the run has ended by @danielhanchen in #7756
Linux/macOS: tell an unreadable llama.cpp install apart from someone else's by @danielhanchen in #7771
Model picker: lay the rows out in columns by @shimmyshimmer in #7770
Fix the model-config Playwright driver for the collapsed single-quant row by @danielhanchen in #7772
Anchor the host-defaults assertions and take the file off the skip list by @danielhanchen in #7774
Model picker: drop the green box around the downloaded icon by @shimmyshimmer in #7775
Studio: fix the hub download rate estimator and bound long ETAs by @ousamabenyounes in #7679
Studio: pin GGUF to CPU on virtualised Metal, clamp parallel slots under MTP by @danielhanchen in #7717
Give the auto-load harness the adapter's new import, and catch the next one by @danielhanchen in #7768
Fix and harden the model-config Playwright job, and pin the hooks it drives by @danielhanchen in #7760
Windows: validate managed Python before package installation by @Etherll in #7763
Studio: Auto download transport, Xet memory caps, and stall detection on the hub path by @danielhanchen in #7742
fix(studio): refresh token count after model load (#7450) by @Souravrajvi0 in #7453
Keep a generating chat's spinner clear of the kebab on hover by @danielhanchen in #7751
fix(studio): neutralize chat-template control markup in client text (#7066) by @Souravrajvi0 in #7334
Rebrand Tauri desktop app as Unsloth by @wasimysaid in #7608
Studio: bump anyhow and event-listener for RustSec advisories by @danielhanchen in #7780
Studio frontend: override brace-expansion 5.0.8 and fast-uri 3.1.4 by @danielhanchen in #7781
Dependabot: pin glib and gdk-pixbuf to the GTK3 0.18 line by @danielhanchen in #7779
Fix model picker stuck on "Loading variants" by @shimmyshimmer in #7767
Fix model settings overflow with many GPUs by @Imagineer99 in #7777
Bump install.sh / install.ps1 pin to unsloth>=2026.8.2 by @danielhanchen in #7786
fix python tool images not rendering in the desktop app by @mahiatlinux in #7808
fix studio toast action alignment by @mahiatlinux in #7809
Model picker: follow the UI font size preference by @shimmyshimmer in #7806
fix(studio): sanitize the healer catalog when no render target is available by @danielhanchen in #7812
Retry macOS disk image stapling by @wasimysaid in #7787
Studio: scale the Model hub to the window instead of a fixed 1100px by @shimmyshimmer in #7816
studio: pair slim whisper bundles on the ggml tree id, not the -mix- tag suffix by @danielhanchen in #7817
Scan Windows desktop bundles with Defender before publishing by @danielhanchen in #7813
Assert chat tracking in em, not in pixels at a 16px base by @danielhanchen in #7826
Add Intel Arc GPU detection and XPU PyTorch install to Windows installer by @legobele in #7706
test(studio): bound the chat font size and accept a normal tracking by @danielhanchen in #7828
Studio: let unsloth_zoo size the download worker instead of overriding it by @danielhanchen in #7815
Desktop: open Data Recipes without triggering an update by @Etherll in #7762
Name the missing shared library when llama-server cannot start by @danielhanchen in #7782
Sign the NSIS plugin DLLs, and fail the release on any unsigned file by @danielhanchen in #7819
Embed the WebView2 bootstrapper instead of downloading it at install time by @danielhanchen in #7820
Repoint three guard tests at where the code moved, and derive the font-size expectations by @danielhanchen in #7833
Studio: filter DSpark/DFlash drafters out of the GGUF quant picker by @danielhanchen in #7811
Studio: handle a deleted dataset file instead of failing with a 500 by @NilayYadav in #7800
Installer: select CUDA wheels that cover the host's GPUs by @oobabooga in #7814
Studio: return 400 for a rejected dataset path in check-format by @danielhanchen in #7842
Desktop: fix dictation failing with a blocked speech service by @NilayYadav in #7799
Fix GGUF fallback cache context source by @Imagineer99 in #7788
Skip an unreadable notebooks candidate instead of failing collection by @danielhanchen in #7844
Keep the package importable on the Python floor pyproject declares by @danielhanchen in #7840
Make the Colab oracle tripwire able to fire, and stop blaming start.py for a hung agent CLI by @danielhanchen in #7838
Keep the signed NSIS plugin directory across a bundler upgrade, and retry the vc_redist fetch by @danielhanchen in #7841
Studio: improve translations across all supported locales by @oobabooga in #7797
Studio: add image generation, editing workflows and LoRA training with Unsloth GGUFs by @oobabooga in #6763

New Contributors

@eliazonta made their first contribution in #7234
@mahille made their first contribution in #7289
@devangpratap made their first contribution in #6760
@ArjunDivecha made their first contribution in #7628
@Taranum01 made their first contribution in #6880
@latent-9 made their first contribution in #7688
@adinfarel made their first contribution in #7729
@LRimes90 made their first contribution in #7650
@ousamabenyounes made their first contribution in #7679
@legobele made their first contribution in #7706

Full Changelog: v0.1.51-beta...v0.1.52-beta

unslothai/unsloth v0.1.52-beta Faster downloading + DeepSeek-V4 Flash 0731 on GitHub