AI FixCode Model 🛠️

A Transformer-based code fixing model trained on diverse buggy → fixed code pairs. Built using CodeT5, this model identifies and corrects syntactic and semantic errors in source code.

📌 Model Details

Base Model: Salesforce/codet5p-220m
Type: Seq2Seq (Encoder-Decoder)
Trained On: Custom dataset with real-world buggy → fixed examples.
Languages: Python (initially), can be expanded to JS, Go, etc.

🔧 Intended Use

Input a buggy function or script and receive a syntactically and semantically corrected version.

Example:

# Input:
def add(x, y)
 return x + y

# Output:
def add(x, y):
    return x + y

🧠 How it Works

The model learns from training examples that map erroneous code to corrected code. It uses token-level sequence generation to predict patches.

🚀 Inference

Use transformers pipeline or run via CLI:

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
model = AutoModelForSeq2SeqLM.from_pretrained("YOUR_USERNAME/aifixcode-model")
tokenizer = AutoTokenizer.from_pretrained("YOUR_USERNAME/aifixcode-model")
input_code = "def foo(x):\n print(x"
inputs = tokenizer(input_code, return_tensors="pt")
out = model.generate(**inputs, max_length=512)
print(tokenizer.decode(out[0], skip_special_tokens=True))

📂 Dataset Format

[
  {
    "input": "def add(x, y)\n return x + y",
    "output": "def add(x, y):\n    return x + y"
  }
]

🛡️ License

MIT License

🙏 Acknowledgements

Built using 🤗 HuggingFace Transformers + Salesforce CodeT5.