Find answers from the community

Updated 2 months ago

Starcoder

I am trying to use HuggingFaceLLMPredictor class with starchat-beta model. How do I decide on the value of two params - max_input_size and max_new_tokens?
I tried to look for guidelines in the documentation but did not find any.
Thank you.
L
a
2 comments
max_new_tokens can be anything you want, but the larger it is, the less room there will be in the overall context window. The default is 256, and that's usually good enough

max_input_size can be weird to find, huggingface doesn't make this easy. Looking at the starchat page on huggingface, it appears to be 8192
https://huggingface.co/HuggingFaceH4/starchat-beta/blob/b1bcda690655777373f57ea6614eb095ec2c886f/config.json#L23

You can definitely set this lower though too, to help with speed/memory usage
Thank you so much. That helps.
Add a reply
Sign up and join the conversation on Discord