Top New Trends Comments Past Ask Show Jobs

sunandcoffee

no profile record

Submissions

K2-65B: the most performant open LLM to date

2 points·by sunandcoffee·il y a 2 ans·0 comments

K2-65B: the first reproducible LLM to outperform Llama 2 70B

1 points·by sunandcoffee·il y a 2 ans·0 comments

CrystalChat – 7B model from LLM360 beating Mistral

3 points·by sunandcoffee·il y a 3 ans·2 comments

LLM360 – transparent LLMs with intermediate checkpoints and data buckets

4 points·by sunandcoffee·il y a 3 ans·0 comments

comments

·il y a 2 ans·discuss

what are you thoughts on projects like these: https://www.llm360.ai/

seems like they make everything available.

·il y a 2 ans·discuss

they mentioned LLM360 in their paper as well: llm360.ai

·il y a 3 ans·discuss

Looks like the model is slightly worse on language ability but better on coding ability. This might be a good model for trying agents.