Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix BT starcoder fp16 #1255

Merged
merged 1 commit into from
Aug 4, 2023
Merged

Fix BT starcoder fp16 #1255

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 1 addition & 5 deletions optimum/bettertransformer/models/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -679,11 +679,7 @@ def gpt_bigcode_wrapped_scaled_dot_product(
raise_on_head_mask(head_mask)

# TODO: remove once PyTorch 2.1 is released with the scale argument to SDPA
if self.scale_attn_weights:
softmax_dtype = torch.float32 if self.attention_softmax_in_fp32 else query.dtype
if self.scale_attention_softmax_in_fp32 and query.dtype != softmax_dtype:
query = query / (self.layer_idx + 1)
else:
if not self.scale_attn_weights:
query = query / self.head_dim**0.5

# MQA models: (batch_size, query_length, num_heads * head_dim)
Expand Down
Loading