polymorph1sm

usually it’s just “small”. In current state, any model that range between 100B to 600B, but it used to be 27B a 2 years back ( Gemini-1 .5 I think )

Some interesting findings from the chat template designs:

1. The template name is Onyx ATEM as found in the tool call exception message

2. It appears to be following a harmony-style chat template. But the tool use seems to be a xml like :<atem:function_calls> / <atem:invoke> / <atem:parameter>

3. atem: a internal joke of meta in reverse?

https://huggingface.co/meta-models/Muse-Glimmer-30B/blob/mai...

One thing I hate about Hsinchu was its windy weather. I do agree Taipei are much better and Kaoshiung is certainly in my second place compared to Hsinchu.

It’s not as good as GPT-4, the only feature I used often was their PDF QA. The rest ChatGPT plus ( GPT-4 ) still miles ahead

This. I recently bought an "second hand" iPhone XS advertised as those came from shop display phones. It was in good condition: no burn ins, battery at 80% health. When the screen broke after dropping it, I went to some shop for repair just to found out its remix of parts from other iPhone ( apparently the motherboard came from South Korea ). This made the device very fragile as some of the screws maybe missing and screen are not properly mounted. The iPhone XS I bought had one missing screw and no display adhesive.

Unless they are able to introduce smaller module like “GPU” version, it will only be a nice demo product of what we can do with state of the art semiconductor manufacturing.

I recently found this paper[1] claiming near GPT-3 performance with only a fraction of parameters. They seems to simply reformulate the input sequence to change classification to a sequence generation task.

Disclaimer, I am not affiliated to any of the authors

[1] https://arxiv.org/pdf/2009.07118.pdf