Kh

Very cool thanks! I think OpenAI already hate Open Source :)))))
Products that are trying so hard to monetize are created in one day.

Yes, it's been tested, and it's false. It's even worse than the regular LLAMA 3.1 70b. It's even funny to compare it to Claude.
https://www.reddit.com/r/LocalLLaMA/s/BH5A2ngyui

Yeah, I was thinking the same thing. A large vocabulary does improve the performance of smaller LLMs and judging by the GPT-4o the same is true for larger LLM. Give it a try. I'm just doing this for small size models up to 3B parameters.