tt-finetune

Parameter-efficient fine-tuning — LoRA, rsLoRA, LoRA+, DoRA, and IA3 — on a single Blackhole P150a, behind a Hugging Face/PEFT-style trainer API. Ships as a self-contained Linux wheel bundling the TT-XLA/PJRT plugin, TTNN, and TT-Metal user-space libraries, so no source checkout, Docker, or PYTHONPATH setup is required. Includes a static planner that reports memory admission before you compile, deterministic checkpoint/resume, and standard PEFT adapter export.