unsloth-studio
| Version | 2026.168.1108 |
| Repo | superproject |
Unsloth Studio web UI for GPU LLM fine-tuning, backed by a pixi PyTorch/transformers environment and composed llama.cpp GGUF export tools unsloth-studio owns the pixi fine-tuning environment (Python 3.13 + PyTorch/CUDA + transformers + PEFT/TRL) and composes the llama-cpp and unsloth candies, then runs the Unsloth Studio web UI as a supervisord service on port 8888. Every owned artifact lands at a fixed path – the pixi interpreter at ~/.pixi/envs/default and the composed llama.cpp CLI at ~/llama.cpp – so the environment and the composition’s effect are directly checkable.
Services
Section titled “Services”unsloth-studio
Acceptance plan
Section titled “Acceptance plan”This candy’s plan: — the runnable spec charly check executes against a live deployment. check: steps are idempotent probes; run: steps change state.
| Intent | Step |
|---|---|
run |
install the Unsloth Studio CLI and bootstrap its runtime venv via the official installer |
check |
the Unsloth Studio CLI launcher is installed by the official installer (the unsloth console-script is NOT in the pixi env — it comes from install.sh) |
check |
the pixi fine-tuning environment provides a Python interpreter at ~/.pixi/envs/default |
check |
PyTorch imports successfully inside that environment |
check |
the composed llama.cpp CLI binary is present so fine-tuned models can be exported to GGUF |
agent-check |
the Unsloth Studio web UI is served on port 8888 and is usable for interactive LLM fine-tuning |
check |
command=${HOME}/.pixi/envs/default/bin/python -c “import transformers, peft, trl; print(‘finetune-stack-ok’)” |
check |
service=unsloth-studio |
check |
addr=127.0.0.1:${HOST_PORT:8888} |