The "cost and performance focused ML models " show quite a large range of differently sized LLMs. From 1.6B up to 8x7B, maybe a more narrow overview would be of value. What do you think?
It's something I've considered. The 8x7B version only has 13B parameters active at inference time, so it fits the cost value proposition. Looking forward, as the field expands and we get even more options with small and medium parameter sizes, I'll look into a more granular analysis.
The "cost and performance focused ML models " show quite a large range of differently sized LLMs. From 1.6B up to 8x7B, maybe a more narrow overview would be of value. What do you think?
It's something I've considered. The 8x7B version only has 13B parameters active at inference time, so it fits the cost value proposition. Looking forward, as the field expands and we get even more options with small and medium parameter sizes, I'll look into a more granular analysis.