avenia

Train your model on a GPU you don't own. The terminal face of Avenia — a decentralized LoRA fine-tuning network. Submit a dataset + a base model, the network fine-tunes a LoRA on someone else's GPU, you get the weights back, and you chat with the model you just trained — all from your terminal. Settled in $AVEN on Robinhood Chain.

Run it

npx avenia            # boot + the animated mascot + the command menu

Commands

avenia                boot + mascot + menu
avenia train          fine-tune a model and watch the loss fall
avenia chat           talk to the model you trained
avenia status         status of your most-recent run
avenia runs           your past training runs
avenia models         base models you can fine-tune
avenia network        live GPU workers on the mesh
avenia mascot         just the animated portrait
avenia help           all commands

Typical flow

npx avenia            # meet her, see the menu
avenia train          # give a dataset, watch it train, she comes back
avenia chat           # talk to your freshly-trained model

avenia train accepts a base model and a dataset: avenia train aven-qwen-0.5b --dataset ./data.jsonl (JSONL of {"prompt":"…","completion":"…"}).

Config

• COORDINATOR_HTTP_URL — coordinator base (default https://avenia.fly.dev).
• NO_COLOR — disable color.

Node ≥ 20. MIT.