Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- Created 3 models (from 315M to 672M parameters) from pre-training to GRPO. Sadly GRPO did hurt the models when it comes to general knowledge. Ended with building a KV cache and serving them on HF Space so people can try them out.
https://github.com/JohnEnev/modern-llm/blob/main/README.md with the code and links to the articles.
by Johnene