mirror of
https://github.com/facebookresearch/blt.git
synced 2025-09-17 01:29:42 +00:00
Fix realtime entropy patching (#26)
* allow loading of the entropy model directly * remove unused argument * remove spammy warning * allow patch_batch_size to be adjusted in the forward() method * revert to original patcher style, fix warning * allow grads when calculating entropies * fix grad flow * return preds from calculate_entropies() * remove legacy arg * fix an error with monotonicity and small sequence lengths * ensure patcher is serializable * revert patcher to original * remove unused import
This commit is contained in:
parent
6ffeb66b53
commit
392117bff2
4 changed files with 26 additions and 12 deletions
|
@ -162,9 +162,6 @@ def create_causal_mask(
|
|||
return "causal"
|
||||
|
||||
if BLT_SUPPRESS_ATTN_ERROR == 1:
|
||||
logging.warning(
|
||||
"SDPA attention being used, which doesn't have specialized attention implementations for block_causal and local_block_causal attention. Allowing model to run since BLT_SUPPRESS_ATTN_ERROR=1"
|
||||
)
|
||||
return "causal"
|
||||
else:
|
||||
raise ValueError(
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue