gliched_robot

If anyone wants to try this out, here is the hugging-face spaces demo: https://huggingface.co/spaces/google/paligemma

This is far more superior than SORA, there is no comparison.

Lmsys devs have all the answers, I am not sure how this has not leaked yet. They must a strong NDAs.

I do not understand the taught process here. They are regulating it so fast. It's almost like regulating car before even engine is invented.

GPU server locations, maybe?

Inference speed is not a great metric given the horizontal scalability of LLMs.

Disagree on Nvidia, most folks fine-tune model. Proof: there are about 20k models in huggingface derived from llama 2, all of them trained on Nvidia GPUs.

Maybe a typo?

This llama model some made it run on an iphone. https://x.com/1littlecoder/status/1781076849335861637?s=46

I see what you did here <q> carrying the "torch" <q>. LOL

The code it writes is getting worse eg. lazy and not updating the function, not following prompts etc. So we can objectively say its getting worse.

Wild considering, GPT-4 is 1.8T.

If any one is interesting in seeing how 400B model compares with other opensource models, here is a useful chart: https://x.com/natolambert/status/1780993655274414123

Is this real? Seems suspicious given that its just one model not a family of models like llama and llama-2.

This is very cool and will change the way we do lora now.