Brand New Mistral 16k Context Size Models got released last night from NurtureAI!

perlthoughts@alien.top · 2 years ago

Brand New Mistral 16k Context Size Models got released last night from NurtureAI!

permalip@alien.top · 2 years ago

I’m not sure who told who that Mistral models are only 8k or 4k. The sliding window is not the context size, it is the embedding positions that is the context size which is 32k.

TeamPupNSudz@alien.top · 2 years ago

I’m not sure who told who that Mistral models are only 8k

The official Mistral product information.

Our very first foundational model: 7B parameters, fast-deployed and easily customisable. Small, yet powerful for a variety of use cases. Supports English and code, and a 8k context length. link

Does Mistral themselves actually mention 32k anywhere?

permalip@alien.top · 2 years ago

It has 32k, they mention it in their config “max_position_embeddings”: 32768. This is the sequence length.

https://preview.redd.it/5r2c9592vr0c1.png?width=256&format=png&auto=webp&s=be88f25168e3cec16cbe7f9aad15f678edf97e99

mcmoose1900@alien.top · 2 years ago

But “true” 16K-32K models like MistralLite seem to perform much better at long context than the default Mistral config.

permalip@alien.top · 2 years ago

There is nothing “true” context length about MistralLite. You are essentially removing the sliding window by doing what Amazon or Yarn is doing.

https://preview.redd.it/rqe1hwc1vr0c1.png?width=256&format=png&auto=webp&s=79f14a98c097d2e8fb5718ffa4d524353b059a10