How to use from
SGLang
Install from pip and serve model
# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "GoCodeo/TestCodeo" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GoCodeo/TestCodeo",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'
Use Docker images
docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "GoCodeo/TestCodeo" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GoCodeo/TestCodeo",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'
Quick Links

TestCodeo - GoCodeo's fine-tuned Language Model dedicated to Python unit test generation.

www.gocodeo.com

Approach

Our team curated a unique dataset of 200,000 prompt-completion pairs in alpaca format, specifically designed for Python unit test generation.

Two-Stage Finetuning Process

Stage 1: We fine-tuned the base Codellama 7B Python model with 25k easy and 75k medium instructions.

Stage 2: The resulting Test-Codeo-Base was further refined with the remaining medium-hard questions to develop TestCodeo.

Evaluation Methodology

Utilizing OpenAI's human eval dataset, we generated test cases for 164 coding instructions and measured code coverage.

Results

TestCodeo achieved an impressive 89% code coverage, surpassing Codellama's 17% and approaching GPT-3.5-turbo's 93%.

Downloads last month
16
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support