Ravin Kumar
Ravin Kumar
Subclassing sampler and adding a break there for the tokens you want to stop on would be a good way to do this! https://github.com/google-deepmind/gemma/blob/main/gemma/sampler.py#L280
This is an odd one, can you try again when the library gets updated (in the next couple of weeks)
@nvasilevv sure thing! Do you know which file(s) you want to start with? I'll open up a subissue for you so this can be address in parts
@nvasilevv discrete distributions sounds great. I'll add a note to the issue to indicate your assignment. Thanks for your continued interest
hey hector, all the ones that are applicable!
Hey folks, Thank you for your interest It will be released soon and contain all the v2 updates such as sliding window attention and GQA. I'll update this issue in...
I also would like to be able to remove or disable dark mode entirely
Here is it explained in text as well.
Fix heading https://ravinkumar.com/GenAiGuidebook/language_models/evaluation.html#evaluation-considerations