Tiny Crypto Sentiment Analysis

🚀 Tiny Crypto Sentiment Analysis

A fine - tuned (with LoRA) version of TinyLlama for predicting the sentiment and subject of cryptocurrency news articles.

This model is a fine - tuned (using LoRA) variant of TinyLlama. It is trained on cryptocurrency news articles to predict the sentiment and subject of an article. The training dataset is Crypto News+.

🚀 Quick Start

✨ Features

• Fine - tuned on cryptocurrency news articles to predict sentiment and subject.
• Based on the TinyLlama model.

📦 Installation

No specific installation steps are provided here, but you need to have torch and transformers libraries installed to use the model.

💻 Usage Examples

Basic Usage

Load the model:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

MODEL_NAME = "curiousily/tiny-crypto-sentiment-analysis"

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=True)

model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    device_map="auto",
    torch_dtype=torch.float16
)

pipe = pipeline(
    task="text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=16,
    return_full_text=False,
)

Prompt format:

prompt = """
### Title:
<YOUR ARTICLE TITLE>
### Text:
<YOUR ARTICLE PARAGRAPH>
### Prediction:
""".strip()

Here's an example:

prompt = """
### Title:
Bitcoin Price Prediction as BTC Breaks Through $27,000 Barrier Here are Price Levels to Watch
### Text:
Bitcoin, the world's largest cryptocurrency by market capitalization, has been making headlines recently as it broke through the $27,000 barrier for the first time. This surge in price has reignited speculation about where Bitcoin is headed next, with many analysts and investors offering their predictions.
### Prediction:
""".strip()

Get a prediction:

outputs = pipe(prompt)
print(outputs[0]["generated_text"].strip())

The output will be like:

subject: bitcoin
sentiment: positive

📚 Documentation

How to Train Your Own Tiny LLM?

Follow the complete tutorial on how this model was trained: https://www.mlexpert.io/bootcamp/fine-tuning-tiny-llm-on-custom-dataset

📄 License

This project is licensed under the Apache-2.0 license.