- Python 71.5%
- TypeScript 22.7%
- Shell 1.9%
- PowerShell 1.6%
- Rust 1.5%
- Other 0.7%
* Studio: wire Anthropic server-side context compaction
Anthropic ships server-side context compaction as a beta
(`compact-2026-01-12`). When the rendered prompt crosses the
configured input-token threshold, Anthropic runs an extra LLM pass
that summarises older turns and the request continues against the
compacted prefix. The response carries the original top-level fields
plus a new `context_management` block (with `applied_edits`) and
`usage.iterations[]` accounting per pass.
Per the docs the feature is currently supported on Opus 4.6, Opus 4.7,
Sonnet 4.6, and Mythos preview. The minimum threshold is 50k tokens;
under-50k requests 400.
Changes:
- Add prefix gate + helper `_anthropic_supports_compaction` plus
constants `_ANTHROPIC_COMPACTION_PREFIXES`, `_ANTHROPIC_COMPACTION_BETA`,
`_ANTHROPIC_COMPACTION_TYPE`, `_ANTHROPIC_COMPACTION_MIN`.
- Add `compaction_threshold: Optional[int]` to ChatCompletionRequest
(50k ge bound, 2M le bound). Thread through `routes/inference.py`
-> `stream_chat_completion` -> `_stream_anthropic`.
- In `_stream_anthropic`, when threshold is set AND the model
accepts compaction, attach `context_management.edits[{type:
"compact_20260112", trigger:{type:"input_tokens", value:N}}]` to
the outbound body. Sub-50k values are clamped up to 50k to keep
the request well-formed.
- Refactor the anthropic-beta header builder to merge any combination
of `code-execution-2025-08-25` + `compact-2026-01-12` flags into
one header value. Unrelated betas added at the registry level still
pass through.
- Add `test_anthropic_compaction.py` with 16 cases: gate matrix
(every doc-listed model), correct body shape, threshold clamping,
beta header merge with code execution, silent no-op on unsupported
models, omitted-threshold pass-through.
Live verified end-to-end against the real Anthropic API:
`compact_20260112` accepted on Opus 4.7, response carries
`context_management.applied_edits` + `usage.iterations[]` as
documented. (The first WebFetch-summarised version of these docs
suggested `compact_20260120`; the actual API only accepts
`compact_20260112`, matching the beta-header date. Worth pinning
behind a test so a future doc update can't drift back.)
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address review: drop ge=50_000 clamp + parse usage.iterations[]
Two reviewer follow-ups on the compaction PR:
1. Pydantic ge=50_000 on compaction_threshold was dead code.
FastAPI rejected sub-50k threshold values with a 422 before the
`max(int(...), _ANTHROPIC_COMPACTION_MIN)` clamp in
_stream_anthropic could ever fire. Relaxed the floor to ge=1 so
the in-helper clamp actually does its job; the schema comment
now explains why this is intentional. Added a regression test
that posts a value of 1 and 49_999 through the real request
schema.
2. Anthropic publishes per-iteration token counts in
`usage.iterations[]` whenever a fresh compaction has run, and
the top-level input_tokens / output_tokens cover only the
`message` iteration -- billing must add the compaction
iterations on top. Aggregate compaction iteration tokens into
`last_usage["compaction_input_tokens" / "compaction_output_tokens"]`
so the cost surface (PR 5690) can read them without re-walking
the array, and surface both figures in the closing stream
summary log. Added two tests: one that pins the aggregation on a
compacted turn and one that pins `None` when no fresh
iterations land (so re-applied compaction blocks don't double-bill).
Sourcing: https://platform.claude.com/docs/en/build-with-claude/compaction
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address review: round-trip Anthropic compaction blocks across turns
Codex P1: once context_management is enabled and Anthropic runs
server-side compaction mid-stream, the response carries a
`{type:"compaction", content:"<summary>"}` content block on the
assistant message. The translator only handled text_delta and
input_json_delta on content_block_delta, so the compaction block
was silently dropped. Worse, the request schema's ContentPart
discriminated Union didn't accept `type:"compaction"`, and
_build_external_messages didn't pass it through, so even a
hand-crafted assistant message carrying the block would 422 at
parse time. Net result: Anthropic re-compacted from scratch on
every subsequent turn, wasting input tokens and reasoning budget.
End-to-end backend wiring of the round-trip:
1. SSE translator. _stream_anthropic now tracks a `current_compaction`
state slot. content_block_start with type=="compaction" seeds it
(Anthropic may include the summary on the start event AND/OR
stream it via text_delta events on the same block index --
handle both). text_delta inside a compaction block routes into
the compaction buffer instead of the user-visible content
stream, since the summary is opaque internal state, not
assistant prose. content_block_stop emits a `compaction_block`
tool_event carrying the full summary so the chat-adapter can
persist it. compaction_blocks_seen is surfaced in the closing
summary log.
2. Pydantic schema. Added CompactionContentPart with Tag("compaction")
on the ContentPart Union so requests carrying the block parse
cleanly. Required `content` field with a docstring pointing at
the Anthropic docs.
3. Message builder. _build_external_messages forwards compaction
parts on both vision and non-vision paths; the per-provider
stream helper decides whether to forward to the wire (Anthropic
does; other providers ignore the part). When a non-vision route
ends up with a single text part, collapse back to a string
so providers that don't accept content arrays still get the
expected shape.
4. _stream_anthropic outbound translator. {type:"compaction"} parts
on an assistant message land on the wire verbatim. Empty/missing
`content` is skipped so a malformed stored block can't 400
Anthropic.
Tests added (5): stream emits compaction_block tool event with the
summary intact; user-visible content stream does NOT carry the
summary text; outbound body forwards compaction parts verbatim on
the next turn; Pydantic schema accepts the part; builder passes
it through on both vision and non-vision provider routes.
Frontend follow-up: the chat-adapter needs to persist the
compaction_block tool_event onto the stored assistant message so
turn N+1 includes it in payload.messages. Pinned in the PR
description.
Sourcing: https://platform.claude.com/docs/en/build-with-claude/compaction
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Address review: gate compaction-part passthrough to Anthropic only
Codex P1: my previous round-trip change preserved {type:"compaction"}
parts on every provider route in _build_external_messages. That
meant a chat history with prior compaction state silently leaked
the Anthropic-specific block to OpenAI/DeepSeek/Mistral/Gemini/
Kimi/OpenRouter on a provider switch, where generic
/chat/completions passthrough hands the unknown content type to
the upstream API and 400s the whole turn.
Added a `provider_type` kwarg to _build_external_messages and
gated the compaction forwarder on `provider_type == "anthropic"`.
Every other value (including the legacy None for callers that
don't pass it yet) strips the part. The Anthropic stream helper
still maps it to a native `compaction` block on the wire.
Threaded provider_type through from _proxy_to_external_provider's
call site.
Tests updated: vision + provider="anthropic" still forwards; six
non-anthropic providers strip the part; missing provider_type
strips defensively; non-vision + anthropic still forwards; non-vision
+ non-anthropic collapses back to a text string.
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
|
||
|---|---|---|
| .github | ||
| images | ||
| scripts | ||
| studio | ||
| tests | ||
| unsloth | ||
| unsloth_cli | ||
| .gitattributes | ||
| .gitignore | ||
| .pre-commit-ci.yaml | ||
| .pre-commit-config.yaml | ||
| build.sh | ||
| cli.py | ||
| CODE_OF_CONDUCT.md | ||
| CONTRIBUTING.md | ||
| COPYING | ||
| install.ps1 | ||
| install.sh | ||
| LICENSE | ||
| pyproject.toml | ||
| README.md | ||
| unsloth-cli.py | ||
Unsloth Studio lets you run and train models locally.
Features • Quickstart • Notebooks • Documentation
⚡ Get started
macOS, Linux, WSL:
curl -fsSL https://unsloth.ai/install.sh | sh
Windows:
irm https://unsloth.ai/install.ps1 | iex
Community:
⭐ Features
Unsloth Studio (Beta) lets you run and train text, audio, embedding, vision models on Windows, Linux and macOS.
Inference
- Search + download + run models including GGUF, LoRA adapters, safetensors
- Export models: Save or export models to GGUF, 16-bit safetensors and other formats.
- Tool calling: Support for self-healing tool calling and web search
- Code execution: lets LLMs test code in Claude artifacts and sandbox environments
- API inference endpoint: Deploy and run local LLMs in Claude Code, Codex tools with Unsloth
- Auto set inference settings and customize chat templates.
- We work directly with teams behind gpt-oss, Qwen3, Llama 4, Mistral, Gemma 1-3, and Phi-4, where we’ve fixed bugs that improve model accuracy.
- Chat with images, audio, PDFs, code, DOCX and more. Connect API providers (OpenAI, Anthropic) or servers (vLLM, Ollama).
Training
- Train and RL 500+ models up to 2x faster with up to 70% less VRAM, with no accuracy loss.
- Custom Triton and mathematical kernels. See some collabs we did with PyTorch and Hugging Face.
- Data Recipes: Auto-create datasets from PDF, CSV, DOCX etc. Edit data in a visual-node workflow.
- Reinforcement Learning (RL): The most efficient RL library, using 80% less VRAM for GRPO, FP8 etc.
- Supports full fine-tuning, RL, pretraining, 4-bit, 16-bit and, FP8 training.
- Observability: Monitor training live, track loss and GPU usage and customize graphs.
- Multi-GPU training is supported, with major improvements coming soon.
📥 Install
Unsloth can be used in two ways: through Unsloth Studio, the web UI, or through Unsloth Core, the code-based version. Each has different requirements.
Unsloth Studio (web UI)
Unsloth Studio (Beta) works on Windows, Linux, WSL and macOS.
- CPU: Supported for Chat and Data Recipes currently
- NVIDIA: Training works on RTX 30/40/50, Blackwell, DGX Spark, Station and more
- macOS: Training, MLX and GGUF inference are ALL supported.
- AMD: Chat + Data works. Train with Unsloth Core. Studio support is out soon.
- Multi-GPU: Available now, with a major upgrade on the way
macOS, Linux, WSL:
curl -fsSL https://unsloth.ai/install.sh | sh
Windows:
irm https://unsloth.ai/install.ps1 | iex
Launch
unsloth studio -p 8888
For cloud or global access, add -H 0.0.0.0. By default, Unsloth is accessible only locally.
Update
To update, use the same install commands above or use unsloth studio update.
Docker
Use our Docker image unsloth/unsloth container. Run:
docker run -d -e JUPYTER_PASSWORD="mypassword" \
-p 8888:8888 -p 8000:8000 -p 2222:22 \
-v $(pwd)/work:/workspace/work \
--gpus all \
unsloth/unsloth
Developer, Nightly, Uninstall
To see developer, nightly and uninstallation etc. instructions, see advanced installation.
Unsloth Core (code-based)
Linux, WSL:
curl -LsSf https://astral.sh/uv/install.sh | sh
uv venv unsloth_env --python 3.13
source unsloth_env/bin/activate
uv pip install unsloth --torch-backend=auto
Windows:
winget install -e --id Python.Python.3.13
winget install --id=astral-sh.uv -e
uv venv unsloth_env --python 3.13
.\unsloth_env\Scripts\activate
uv pip install unsloth --torch-backend=auto
For Windows, pip install unsloth works only if you have PyTorch installed. Read our Windows Guide.
You can use the same Docker image as Unsloth Studio.
AMD, Intel:
For RTX 50x, B200, 6000 GPUs: uv pip install unsloth --torch-backend=auto. Read our guides for: Blackwell and DGX Spark.
To install Unsloth on AMD and Intel GPUs, follow our AMD Guide and Intel Guide.
📒 Free Notebooks
Train for free with our notebooks. You can use our new free Unsloth Studio notebook to run and train models for free in a web UI. Read our guide. Add dataset, run, then deploy your trained model.
| Model | Free Notebooks | Performance | Memory use |
|---|---|---|---|
| Gemma 4 (E2B) | ▶️ Start for free | 1.5x faster | 50% less |
| Qwen3.5 (4B) | ▶️ Start for free | 1.5x faster | 60% less |
| gpt-oss (20B) | ▶️ Start for free | 2x faster | 70% less |
| Qwen3.5 GSPO | ▶️ Start for free | 2x faster | 70% less |
| gpt-oss (20B): GRPO | ▶️ Start for free | 2x faster | 80% less |
| Qwen3: Advanced GRPO | ▶️ Start for free | 2x faster | 70% less |
| embeddinggemma (300M) | ▶️ Start for free | 2x faster | 20% less |
| Mistral Ministral 3 (3B) | ▶️ Start for free | 1.5x faster | 60% less |
| Llama 3.1 (8B) Alpaca | ▶️ Start for free | 2x faster | 70% less |
| Llama 3.2 Conversational | ▶️ Start for free | 2x faster | 70% less |
| Orpheus-TTS (3B) | ▶️ Start for free | 1.5x faster | 50% less |
- See all our notebooks for: Kaggle, GRPO, TTS, embedding & Vision
- See all our models and all our notebooks
- See detailed documentation for Unsloth here
🦥 Unsloth News
- Connections: Connect any API provider (OpenAI, Anthropic) or server (vLLM, Ollama). Guide
- MTP: Run Qwen3.6 MTP in Unsloth. MTP settings are autoset specific to your hardware. Guide
- API inference endpoint: Deploy and run local LLMs in Claude Code, Codex tools. Guide
- Qwen3.6: Qwen3.6-35B-A3B can now be trained and run in Unsloth Studio. Blog
- Gemma 4: Run and train Google’s new models directly in Unsloth. Blog
- Introducing Unsloth Studio: our new web UI for running and training LLMs. Blog
- Qwen3.5 - 0.8B, 2B, 4B, 9B, 27B, 35-A3B, 112B-A10B are now supported. Guide + notebooks
- Train MoE LLMs 12x faster with 35% less VRAM - DeepSeek, GLM, Qwen and gpt-oss. Blog
- Embedding models: Unsloth now supports ~1.8-3.3x faster embedding fine-tuning. Blog • Notebooks
- New 7x longer context RL vs. all other setups, via our new batching algorithms. Blog
- New RoPE & MLP Triton Kernels & Padding Free + Packing: 3x faster training & 30% less VRAM. Blog
- 500K Context: Training a 20B model with >500K context is now possible on an 80GB GPU. Blog
- FP8 & Vision RL: You can now do FP8 & VLM GRPO on consumer GPUs. FP8 Blog • Vision RL
📥 Advanced Installation
The below advanced instructions are for Unsloth Studio. For Unsloth Core advanced installation, view our docs.
Developer installs: macOS, Linux, WSL:
git clone https://github.com/unslothai/unsloth
cd unsloth
./install.sh --local
unsloth studio -p 8888
Then to update :
unsloth studio update
Developer installs: Windows PowerShell:
git clone https://github.com/unslothai/unsloth.git
cd unsloth
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\install.ps1 --local
unsloth studio -p 8888
Then to update :
unsloth studio update
Nightly: MacOS, Linux, WSL:
git clone https://github.com/unslothai/unsloth
cd unsloth
git checkout nightly
./install.sh --local
unsloth studio -p 8888
Then to launch every time:
unsloth studio -p 8888
Nightly: Windows:
Run in Windows Powershell:
git clone https://github.com/unslothai/unsloth.git
cd unsloth
git checkout nightly
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\install.ps1 --local
unsloth studio -p 8888
Then to launch every time:
unsloth studio -p 8888
Uninstall
The recommended way to fully remove Unsloth Studio is the matching uninstall script for your OS. It stops any running servers, removes the install dir, the launcher data dir, the desktop shortcut, and any platform-specific entries (macOS .app bundle + Launch Services on Mac; Start Menu, HKCU\Software\Unsloth registry key and user PATH entries on Windows):
- MacOS, WSL, Linux:
curl -fsSL https://raw.githubusercontent.com/unslothai/unsloth/main/scripts/uninstall.sh | sh - Windows (PowerShell):
irm https://raw.githubusercontent.com/unslothai/unsloth/main/scripts/uninstall.ps1 | iex
If you only want to drop the install dir and keep the launcher/shortcut for a later reinstall, you can instead run rm -rf ~/.unsloth/studio (Mac/Linux/WSL) or Remove-Item -Recurse -Force "$HOME\.unsloth\studio" (Windows). The model cache at ~/.cache/huggingface is not touched by any of these.
For more info, see our docs.
Deleting model files
You can delete old model files either from the bin icon in model search or by removing the relevant cached model folder from the default Hugging Face cache directory. By default, HF uses:
- MacOS, Linux, WSL:
~/.cache/huggingface/hub/ - Windows:
%USERPROFILE%\.cache\huggingface\hub\
💚 Community and Links
| Type | Links |
|---|---|
| Join Discord server | |
| Join Reddit community | |
| 📚 Documentation & Wiki | Read Our Docs |
| Follow us on X | |
| 🔮 Our Models | Unsloth Catalog |
| ✍️ Blog | Read our Blogs |
Citation
You can cite the Unsloth repo as follows:
@software{unsloth,
author = {Daniel Han, Michael Han and Unsloth team},
title = {Unsloth},
url = {https://github.com/unslothai/unsloth},
year = {2023}
}
If you trained a model with 🦥Unsloth, you can use this cool sticker!
License
Unsloth uses a dual-licensing model of Apache 2.0 and AGPL-3.0. The core Unsloth package remains licensed under Apache 2.0, while certain optional components, such as the Unsloth Studio UI are licensed under the open-source license AGPL-3.0.
This structure helps support ongoing Unsloth development while keeping the project open source and enabling the broader ecosystem to continue growing.
Thank You to
- The llama.cpp library that lets users run and save models with Unsloth
- The Hugging Face team and their libraries: transformers and TRL
- The Pytorch and Torch AO team for their contributions
- NVIDIA for their NeMo DataDesigner library and their contributions
- And of course for every single person who has contributed or has used Unsloth!