README.md

---
license: apache-2.0
language:
- en
library_name: transformers
pipeline_tag: text-generation
tags:
- pinescript
- tradingview
- code-generation
- finance
- trading
- codegemma
- fine-tuned
base_model: google/codegemma-7b-it
datasets:
- anthonym21/pinescript-v5-instructions
model-index:
- name: pinescript-v5-instructions-merged
  results: []
---

# PineScript v5 Code Generator

A fine-tuned CodeGemma 7B model specialized in generating TradingView PineScript v5 code for trading indicators, strategies, and libraries.

## Model Description

This model was fine-tuned on 5,000+ PineScript v5 examples from the [PineScripts-Permissive](https://huggingface.co/datasets/mrmegatelo/PineScripts-Permissive) dataset, which contains high-quality, permissively-licensed scripts from TradingView.

- **Base Model:** [google/codegemma-7b-it](https://huggingface.co/google/codegemma-7b-it)
- **Fine-tuning Method:** QLoRA (4-bit quantization + LoRA adapters)
- **Training Data:** 4,774 instruction/response pairs
- **Context Length:** 4096 tokens

## Intended Use

Generate PineScript v5 code for:
- Technical indicators (RSI, MACD, Bollinger Bands, custom indicators)
- Trading strategies with backtesting
- Reusable libraries
- Alert conditions
- Custom visualizations

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "anthonym21/pinescript-v5-instructions-merged"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

def generate_pinescript(prompt, max_tokens=1024):
    formatted = f"### Human: {prompt}\n### Assistant:"
    inputs = tokenizer(formatted, return_tensors="pt").to(model.device)

    outputs = model.generate(
        **inputs,
        max_new_tokens=max_tokens,
        temperature=0.7,
        top_p=0.9,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id,
    )

    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("### Assistant:")[-1].strip()

# Example
code = generate_pinescript("Write a PineScript v5 RSI indicator with overbought/oversold zones")
print(code)
```

## Example Prompts

| Prompt | Description |
|--------|-------------|
| "Write a PineScript v5 indicator that shows RSI with dynamic overbought/oversold levels" | RSI with adaptive levels |
| "Create a MACD crossover strategy with stop loss and take profit" | Complete trading strategy |
| "Write a Bollinger Bands indicator with squeeze detection" | Volatility indicator |
| "Create a multi-timeframe moving average indicator" | MTF analysis tool |

## Training Details

| Parameter | Value |
|-----------|-------|
| Epochs | 3 |
| Batch Size | 2 |
| Gradient Accumulation | 8 |
| Learning Rate | 2e-4 |
| LoRA r | 64 |
| LoRA alpha | 128 |
| Max Seq Length | 4096 |
| Quantization | 4-bit (nf4) |

## Limitations

- Generates PineScript v5 syntax; may not be compatible with older versions
- Code should be reviewed and tested before live trading
- Complex multi-indicator strategies may require refinement
- Does not provide financial advice

## Dataset

Training data sourced from [mrmegatelo/PineScripts-Permissive](https://huggingface.co/datasets/mrmegatelo/PineScripts-Permissive):
- 5,848 PineScript v5 scripts
- Filtered to 4,774 high-quality examples
- Includes indicators, strategies, and libraries
- All scripts under permissive licenses (MPL-2.0, Apache-2.0)

## Citation

```bibtex
@misc{pinescript-v5-instructions-merged,
  author = {Anthony Maio},
  title = {PineScript v5 Code Generator},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/anthonym21/pinescript-v5-instructions-merged}
}
```

## License

Apache 2.0 (same as base model)