python_code_assistant

• finetuned Salesforce/codegen-350M-mono model into a python code assistant using peft.

• Dataset used : iamtarun/python_code_instructions_18k_alpaca

• Base model size : 350 Million

• Total tokens Trained for 6 epochs : 23 Million

• Trainable parameters of the peft model : 2,621,440

Metrics

• RogueL : 0.43 on 150 test samples
• Training loss - 0.51

Training

• 17k samples for Training.
• 3.5 Hours on Tesla P100 -16 GB (Free kaggle GPU support)
• Dataset : https://huggingface.co/datasets/iamtarun/python_code_instructions_18k_alpaca

Test the model Yourself

Notebook link : https://colab.research.google.com/drive/1vs94XnCnolhfwvYmgEhZrR_1UqcHmBzg