riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 33 Packages Projects Releases Wiki Activity

Fix copies for DBRX - neuron fix (#30610)

Browse Source
This commit is contained in:
amyeroberts 2024-05-02 11:00:26 +01:00 committed by GitHub
parent f95302584b
commit 4ad5adaf1d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/transformers/models/dbrx/modeling_dbrx.py
View File

@ -1256,8 +1256,11 @@ class DbrxModel(DbrxPreTrainedModel):
causal_mask = causal_mask.clone() # copy to contiguous memory for in-place edit
if attention_mask.dim() == 2:
mask_length = attention_mask.shape[-1]
padding_mask = causal_mask[..., :mask_length].eq(0.0) * attention_mask[:, None, None, :].eq(0.0)
causal_mask[..., :mask_length] = causal_mask[..., :mask_length].masked_fill(padding_mask, min_dtype)
padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]
padding_mask = padding_mask == 0
causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
padding_mask, min_dtype
)
elif attention_mask.dim() == 4:
# backwards compatibility: we allow passing a 4D attention mask shorter than the input length with
# cache. In that case, the 4D attention mask attends to the newest tokens only.