FINE TUNE

To fine-tune a model on Replicate using a dataset uploaded on Hugging Face:

Step 1: Upload Dataset on Hugging Face

Ensure your dataset is publicly accessible on Hugging Face Datasets.

Step 2: Install Dependencies

pip install replicate transformers datasets

Step 3: Load Dataset from Hugging Face

from datasets import load_dataset
dataset = load_dataset('huggingface/ghost')

Step 4: Fine-tune on Replicate

Use Replicate’s API to fine-tune:

import replicate

training = replicate.trainings.create(
    version="your-model-version",
    input={
        "train_data": "URL_to_your_Hugging_Face_dataset",
        "num_train_epochs": 3
    },
    destination="your-replicate-username/your-model"
)

Step 5: Monitor and Deploy

Track on the Replicate dashboard and use the trained model via the API.

git clone https://huggingface.co/spaces/livethelifetv/ghost
git commit -am 'Update space' && git push
npm i -D @gradio/client

import { Client } from "@gradio/client";
const client = await Client.connect("livethelifetv/ghost");
const result = await client.predict("/chat", { 
message: "Hello!!", 
system_message: "Hello!!", 
max_tokens: 1, 
temperature: 0.1, 
top_p: 0.1, 
});

console.log(result.data);