Crypto Trader L M

Introduction
CryptoTrader-LM is a machine learning model designed to predict trading decisions—buy, sell, or hold—for Bitcoin (BTC) and Ethereum (ETH) based on cryptocurrency news and historical price data. The model is fine-tuned using LoRA on the Ministral-8B-Instruct-2410 base model, tailored for the FinNLP @ COLING-2025 Cryptocurrency Trading Challenge.

Architecture
CryptoTrader-LM uses a LoRA fine-tuned version of the Mistral-8B model, which is a transformer-based architecture optimized for instruction-following tasks. It is fine-tuned using the PEFT framework with a focus on parameter efficiency, making it suitable for financial decision-making tasks.

Training
The model was trained on data from 2022-01-01 to 2024-10-15, incorporating cryptocurrency to USD exchange rates and news articles. Training involved preprocessing of text and price data, with hyperparameters including a batch size of 1, learning rate of 5e-5, and 3 epochs. The process took approximately 3 hours using a 4x A100 GPU setup. Evaluation metrics included Sharpe Ratio, Profit and Loss, and accuracy, with a Sharpe Ratio of 0.94 achieved on the validation set.

Guide: Running Locally

1. Set Up Environment: Ensure you have Python 3.10 and necessary libraries like PyTorch and Hugging Face Transformers installed.
2. Clone Repository: Clone the model from Hugging Face.
3. Install PEFT: Install the PEFT library for fine-tuning.
4. Load Model: Use the provided code snippet to load and run the model locally.
5. Inference: Input your prompt and generate predictions.

For optimal performance, consider using cloud GPUs such as NVIDIA A100 with at least 24GB of VRAM.

License
The model and its components are made available under the terms specified by the creators. For detailed licensing information, refer to the model card on Hugging Face.