---
language: en
license: apache-2.0
tags:
  - text-classification
  - finance
  - trading
  - distilbert
datasets:
  - custom
metrics:
  - accuracy
  - f1
---

# 📊 Ohanvi Fix-Trade Signal Classifier

A tiny DistilBERT model fine-tuned to classify market news headlines and analysis snippets
into three trade signals: **BUY**, **SELL**, or **HOLD**.

## Model Description

- **Base model**: `distilbert-base-uncased` (67M parameters)
- **Task**: 3-class text classification (trade signal detection)
- **Labels**: `BUY (0)` | `SELL (1)` | `HOLD (2)`
- **Training data**: 150 hand-crafted trade signal examples (50 per class)
- **Training time**: ~20-30 seconds on Apple Silicon (MPS)

## Usage

```python
from transformers import pipeline

pipe = pipeline("text-classification", model="ohanvi/ohanvi-fix-trade-model", top_k=None)
result = pipe("Strong earnings beat, revenue up 25%, guidance raised.")
print(result)
# [{'label': 'BUY', 'score': 0.92}, {'label': 'HOLD', 'score': 0.05}, {'label': 'SELL', 'score': 0.03}]
```

## Live Demo

Try it in your browser: [Gradio Space](https://huggingface.co/spaces/ohanvi/ohanvi-fix-trade-model-demo)

## Training Details

| Hyperparameter | Value |
|---|---|
| Base model | distilbert-base-uncased |
| Epochs | 5 |
| Batch size | 8 |
| Learning rate | 3e-4 |
| Max sequence length | 64 |
| Training samples | 120 |
| Eval samples | 30 |

## Limitations

⚠️ This is a **demonstration model** trained on a very small dataset.
Do **not** use it for real trading or financial decisions.

## Project Structure

```
ohanvi-fix-trade-model/
├── train.py          # Fine-tune the model (runs in ~30 sec)
├── inference.py      # Run predictions locally
├── app.py            # Gradio demo (deployed to HF Spaces)
├── push_to_hub.py    # Upload model + Space to Hugging Face
├── config.py         # Central config (loads from .env)
├── requirements.txt  # Python dependencies
├── .env.example      # Template for secrets
└── README.md         # This file
```

## Quick Start

```bash
# 1. Install dependencies
pip install -r requirements.txt

# 2. Copy and fill in your credentials
cp .env.example .env

# 3. Train the model (~30 sec)
python train.py

# 4. Test locally
python inference.py --demo

# 5. Push to Hugging Face Hub
python push_to_hub.py
```