Skip to content
中文Sign inRaise GPU yieldStart mining
Menu

Fine-tune the top open models.

LoRA on Qwen and Llama bases, through a job API in the OpenAI format. When the job is done, download your adapter.

Get an API key API & docs

The top open bases.

  • Qwen Alibaba’s open models, in several sizes.
  • Llama Meta’s open models, in several sizes.

Fine-tunes of Llama bases: Built with Llama.

Three steps.

  1. Upload your data.

    Chat examples in JSONL, the format OpenAI fine-tuning uses.

  2. Create a job.

    One call: the base model, the training file, the method. The OpenAI SDK works as it is.

  3. Download your adapter.

    The LoRA weights are yours, to run wherever you like.

Every training job also mines.

Training mines $PEARL inside the work and in the gaps between its steps. The GPUs earn twice on the same compute.

How Peryn works

A job API in the OpenAI format.

The OpenAI SDK and the calls you know: set our base URL and your key.

import osfrom openai import OpenAI client = OpenAI(base_url="https://api.peryn.ai/v1",api_key=os.environ["PERYN_API_KEY"],) data = client.files.create(file=open("train.jsonl", "rb"), purpose="fine-tune")job = client.fine_tuning.jobs.create(model="llama-3.1-8b-instruct",training_file=data.id,method={"type": "supervised"},)print(job.id, job.status)