mistral-stock-model

Model Card for Model ID

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

• Developed by: Dehaze
• Funded by: Dehaze
• Model type: Text-generation
• Language(s) (NLP): English
• License: [More Information Needed]
• Finetuned from model: Mistral-7B-v0.1

Model Sources

• Repository: DeHazeLabs/llm-case-study/stock-analysis

Uses

Direct Use

The model can be directly used to analyze stock option data and provide actionable trading insights based on the input provided. It can assist users in understanding key metrics such as implied volatility, option prices, technical indicators, and more, to make informed trading decisions.

Downstream Use

Users can fine-tune the model for specific tasks related to stock market analysis or integrate it into larger systems for automated trading strategies, financial advisory services, or sentiment analysis of financial markets.

Bias, Risks, and Limitations

The model's predictions may be influenced by biases present in the training data, such as historical market trends or prevailing market sentiment. Additionally, the model's effectiveness may vary depending on the quality and relevance of the input data provided by users.

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Users should exercise caution and validate the model's predictions with additional research and analysis before making any trading decisions. It's also recommended to consider multiple sources of information and consult with financial experts when interpreting the model's output.

How to Get Started with the Model

Installation

Ensure that you have the transformers library installed. If not, you can install it via pip:

pip install transformers

You can load the model using the provided pipeline or directly with the AutoTokenizer and AutoModelForCausalLM classes from the transformers library. Once the model is loaded, you can use it for text generation tasks. If you prefer a high-level interface, you can use the pipeline approach as well. Alternatively, you can directly interact with the model using the tokenizer and model objects as well.

Training Details

Training Data

The model was trained on a dataset containing examples of stock option data paired with corresponding trading insights. The dataset includes information such as implied volatility, option prices, technical indicators, and trading recommendations for various stocks.

Training Procedure

Preprocessing

The input data was preprocessed to tokenize and encode the text input before training.

Training Hyperparameters

• Training regime: Training regime: Mixed precision training with bf16 precision. Warmup steps: 1 Per-device train batch size: 2 Gradient accumulation steps: 1 Max steps: 500 Learning rate: 2.5e-5 Optimizer: paged_adamw_8bit Logging and saving strategy: Logging and saving checkpoints every 25 steps with wandb integration.

Testing Data, Factors & Metrics

Testing Data

The testing data consisted of examples similar to the training data, with stock option data and expected trading insights provided.

Factors

Factors considered during evaluation include the quality of the model's predictions, alignment with expected trading recommendations, and consistency across different test cases.

Metrics

Evaluation metrics include accuracy of trading recommendations, relevance of generated insights, and overall coherence of the model's output.

Results

The model demonstrated the ability to provide relevant and actionable trading insights based on the input stock option data.

Technical Specifications

Compute Infrastructure

1 x A100 GPU - 80GB VRAM 117 GB RAM 12 vCPU