|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- trading |
|
|
- finance |
|
|
- deepseek |
|
|
- fine-tuning |
|
|
--- |
|
|
|
|
|
# DeepSeek Trading Assistant |
|
|
|
|
|
This is a fine-tuned version of `DeepSeek-R1-Distill-Qwen-32B` specialized for generating trading strategies and market analysis. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Model Description |
|
|
|
|
|
- **Developed by:** latchkeyChild |
|
|
- **Model type:** Decoder-only language model |
|
|
- **Language(s):** English |
|
|
- **License:** MIT |
|
|
- **Finetuned from model:** [deepseek-ai/DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) |
|
|
|
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
|
|
|
This model is designed to: |
|
|
|
|
|
1. Analyze market conditions using technical indicators |
|
|
2. Generate trading strategies based on market analysis |
|
|
3. Implement risk management rules |
|
|
4. Create Python code for strategy implementation |
|
|
|
|
|
### Training Data |
|
|
|
|
|
The model is trained on a custom dataset containing: |
|
|
|
|
|
- Market analysis using technical indicators (RSI, MACD, Moving Averages) |
|
|
- Trading strategy implementations |
|
|
- Risk management rules |
|
|
- Python code examples using QuantConnect framework |
|
|
|
|
|
### Training Procedure |
|
|
|
|
|
#### Training Hyperparameters |
|
|
|
|
|
- **Number of epochs:** 3 |
|
|
- **Batch size:** 2 |
|
|
- **Learning rate:** 1e-5 |
|
|
- **Gradient accumulation steps:** 8 |
|
|
- **Warmup steps:** 100 |
|
|
- **Training regime:** fp16 mixed precision with gradient checkpointing |
|
|
- **Temperature:** 0.6 (recommended for DeepSeek-R1 series) |
|
|
|
|
|
## Technical Specifications |
|
|
|
|
|
### Compute Infrastructure |
|
|
|
|
|
- **Required Hardware:** 2x NVIDIA A10G GPUs or 1x A100 GPU |
|
|
- **Training Time (estimated):** 2-4 hours |
|
|
|
|
|
## Model Card Contact |
|
|
|
|
|
For questions or issues, please open an issue in the repository. |
|
|
|
