Stock Market Prediction Model

A fine-tuned Llama 3B model for stock market sentiment analysis and prediction.

Overview

This project fine-tunes a Llama 3B model using LoRA to predict stock market movements (Bullish/Bearish) based on trend data and news sentiment.

Model

• Base Model: Llama 3B (4-bit quantized)
• Fine-tuning: LoRA with MLX
• Training Data: Stock news with trend and sentiment labels

Project Structure

.
├── data/                    # Training data
│   ├── train.jsonl          # Training examples
│   └── valid.jsonl         # Validation examples
├── clean_adapters/         # Fine-tuned LoRA adapters
├── fused_clean_f16/        # Merged model (FP16)
├── Modelfile              # Ollama configuration
└── README.md

Setup

# Install dependencies
pip install mlx-lm transformers huggingface_hf

# Fine-tune the model
mlx_lm lora \
  --model ./models/llama-3b-4bit \
  --train \
  --data ./data \
  --iters 600 \
  --batch-size 4 \
  --learning-rate 4e-5 \
  --adapter-path ./clean_adapters

Usage with Ollama

# Create Ollama model
ollama create stock-expert -f Modelfile

# Run inference
ollama run stock-expert "NVDA is up 2% with strong AI chip demand"

Inference with Hugging Face

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("your-username/stock-expert-model")
tokenizer = AutoTokenizer.from_pretrained("your-username/stock-expert-model")

# Generate prediction
prompt = "Stock: NVDA. Trend: +2%. Sentiment: Strong AI chip demand."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0]))

License

MIT