transformers
64f94504 - [CB] Add a "max_requests_per_batch" parameter (#46434)

Commit
36 days ago
[CB] Add a "max_requests_per_batch" parameter (#46434) * Broaden the safety margin and add it to the config * Plumbing * IO take attributes from the CB config * Cap max req per batch to max tokens per batch * Update cfg is now a fn * Finalize to model runner and IO * Rename to be grammaticaly correct * Auto review * Small fixes * Fix imports and remove debub print * Fix wrong max value for padding * Solved an aleady solved TODO * Add progress bar now that there is less info * Typos * Add a warning for persistent + sized manager * Add TODO * Fix wrong bound * Fix logit processors * Fix eager crash of FA2 * Test an edge case * Better comment
Author
Parents
Loading