Skip to content

Conversation

Madrawn
Copy link

@Madrawn Madrawn commented Dec 1, 2023

#180
Potentially fixes the problem where a token optimal length over the minimal value fails when the engine also triggers the onnx export.

I just adjusted the get_batch_dim so, when called with the default engine generation parameters, with the only difference being the optimal and maximal token length being increased, it produces the default engine but with optimal and max token length raised.

And I pass the token_min_length parameter from the ui that was abandoned through the objects so the token_min_length can actually be set. Right now min_length and opt_length are just the same value (opt_length).

Daniel Dengler added 2 commits December 1, 2023 03:25
for unclear reasons opt_batch will later divided by 2 anyways.
For some reason it was replaced with text_optlen at some point and
 the actual minlength value was never used
@contentis contentis changed the base branch from main to dev February 5, 2024 09:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant