My blog on Medium:
(How I Landed a Google DeepMind Project in Google Summer of Code 2025: A Step-by-Step Guide)
Update on May 7: Got selected by Google DeepMind for the Gemma project!
Update on May 8: Got rejected by two other orgs lol.
I’m planning to make all my proposals public in case anyone’s curious about the GSoC application process.
Here’s what I submitted for DeepMind:
- A proposal
- A demo repo
- A blog post (shared under the [demo] tag in the issue section of the Gemma repo): google-deepmind/gemma#244
Feel free to reach out. And good luck to anyone applying for the 2026 batch!
This benchmark suite enables researchers and practitioners to systematically evaluate Google's Gemma language models across a variety of tasks. The project has been implemented and is available as an open-source repository.
- Systematic evaluation of Gemma models across standard academic benchmarks
- Comparison between different Gemma model sizes and variants
- Benchmarking against other open models like Llama 2 and Mistral
- Generation of informative visualizations for easy interpretation
- Automation of the benchmarking process with reproducible scripts
- Statistical validation of observed performance differences
The benchmark suite is structured with a modular architecture:
- Core framework for loading models and datasets
- Task-specific implementations (MMLU, coding, math reasoning, etc.)
- Efficiency evaluation modules for measuring latency and memory usage
- Visualization components for generating charts and reports
The implementation is available at: github.com/heilcheng/gemma-benchmark