🤖 Gopilot - AI-Assitant for Golang

GoPilot is a 290M parameters language model trained exclusively on Go code using a small research budget (~100$).

Demo of the Gopilot VSCode Code Extension

⭐️ Diego Rojas ⭐️ Wenbin Leow ⭐️ Jayden Macdonald ⭐️ Qinkai Zheng

Overview

Gopilot is a GPT-style Transformer model trained on 20B tokens on a single RTX4090 for less than a week using the Go split of The Stack Dedup v1.2 dataset. It comes in two flavours: a HuggingFace tokenizer based model and a model based on a custom Go tokenizer that we developped.

The pre-training and fine-tuning weights are made available here.

aws s3 ls s3://gopilot/checkpoints/ --region ap-east-1

Installation

You need to have conda and go installed on your machine. You can install the necessary dependencies using conda and the provided environment_cpu.yml (choose environment_cuda.yml when running CUDA). Dependencies may not be up to date, hence, using the official Docker image is preferred.

Build the Go tokenizer binary:

# Linux, MacOS
go build -o tokenizer/libgotok.so -buildmode=c-shared ./tokenizer/libgotok.go
# Windows
go build -o tokenizer/libgotok.dll -buildmode=c-shared ./tokenizer/libgotok.go

Usage

A CUDA Docker image is made available here.

Pre-Training

The pre-training script trains the model for the specified token budget. Expects a pre-tokenized dataset.

python train.py \
        --model-cf model/config/gopilot-290M.yml \
        --tokenizer hugging-face \
        --tokenizer-cf tokenizer/config/hugging-face.json \
        --s3-dataset-prefix <prefix-of-your-s3-dataset> \
        --s3-bucket <your-s3-bucket> \
        --gradient-accumulation-steps 64 \
        --optimizer sophiag \
        --batch-size 8 \
        --lr 0.0005 \
        --token-budget 10000000000 \
        --device cuda \
        --precision fp16 \
        --s3-checkpoints \
        --warmup 1000 \
        --neptune \
        --compile

Fine-tuning

You can fine-tune Gopilot on any JSONL dataset composed of samples of the following form: {"sample": "package main\nconst Value = 1..."}. We use a mix of pre-training samples, AI-generated samples to perform finetuning.

python finetune.py \
    --model-cf model/config/gopilot-290M.yml \
    --tokenizer-cf tokenizer/config/hugging-face.json \
    --tokenizer hugging-face \
    --in-model-weights checkpoints/hugging-face.pt \
    --out-model-weights checkpoints/hugging-face-ft.pt \
    --dataset-filepath all \
    --gradient-accumulation-steps 16 \
    --batch-size 8 \
    --dropout 0.1 \
    --weight-decay 0.1 \
    --lr 0.000025 \
    --num-epochs 10 \
    --precision fp16 \
    --neptune

Evaluation

The evaluation script runs evaluation on the HumanEvalX benchmark. Our best model obtains 7.4% pass@10 and 77.1% compile@10. Check out the results folder for more information.

python evaluate.py \
    --model-cf model/config/gopilot-290M.yml \
    --tokenizer-cf tokenizer/config/gopilot.json \
    --tokenizer gopilot \
    --model-weights /checkpoints/gopilot-ft.pt \
    --device cuda \
    --k 10 \
    --max-new-tokens 128 \
    --verbose

Inference Server

The inference server is a simple HTTP server that hosts the model and exposes a /complete endpoint to submit samples to auto-complete. It's used by the VSCode extension to provide completions.

python inference_server.py \
    --model-cf model/config/gopilot-290M.yml \
    --tokenizer-cf tokenizer/config/gopilot.json \
    --tokenizer gopilot \
    --device mps \
    --checkpoint-path .cache/checkpoints/gopilot-ft.pt

VSCode Extension

Check out the Gopilot VSCode extension here. Works with the inference server.

Acknowledgements & Notes

Thank you to Qinkai Zheng for providing guidance and the hardware resources.
We did not check for leakage when performing HumanEvalX evaluation. Do not include these results in research.
This project was made during the course of Deep Learning (80240743-0) at Tsinghua University.
While fun to play around with, we do not recommend using a model of this size for code completion in your editor. It's a school project!
Feel free to use the code, tweak the checkpoints, and all!

Future Work

Release the model weights on HuggingFace
Quantize the model weights for fast inference
Interactive online demo
Try on other languages such as Rust or C++
Experiment with different tokenization strategies
Train for longer on more data

shawnharmsen/gopilot