From 896fb6d8d7c10001eb2a92568be7b4bd3d5ddea3 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E3=81=A1=E3=81=8F=E3=82=8F=E3=81=B6?= <55102558+Prgckwb@users.noreply.github.com> Date: Thu, 13 Jun 2024 16:52:35 +0900 Subject: [PATCH] Fix duplicate variable assignments in SD3's JointAttnProcessor (#8516) * Fix duplicate variable assignments. * Fix duplicate variable assignments. --- src/diffusers/models/attention_processor.py | 5 +---- utils/update_metadata.py | 1 + 2 files changed, 2 insertions(+), 4 deletions(-) diff --git a/src/diffusers/models/attention_processor.py b/src/diffusers/models/attention_processor.py index 39c9a1b484..d363194939 100644 --- a/src/diffusers/models/attention_processor.py +++ b/src/diffusers/models/attention_processor.py @@ -1132,9 +1132,7 @@ class JointAttnProcessor2_0: key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2) value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2) - hidden_states = hidden_states = F.scaled_dot_product_attention( - query, key, value, dropout_p=0.0, is_causal=False - ) + hidden_states = F.scaled_dot_product_attention(query, key, value, dropout_p=0.0, is_causal=False) hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim) hidden_states = hidden_states.to(query.dtype) @@ -1406,7 +1404,6 @@ class XFormersAttnProcessor: class AttnProcessorNPU: - r""" Processor for implementing flash attention using torch_npu. Torch_npu supports only fp16 and bf16 data types. If fp32 is used, F.scaled_dot_product_attention will be used for computation, but the acceleration effect on NPU is diff --git a/utils/update_metadata.py b/utils/update_metadata.py index 5aa3f289e7..103a2b9ab0 100644 --- a/utils/update_metadata.py +++ b/utils/update_metadata.py @@ -24,6 +24,7 @@ python utils/update_metadata.py Script modified from: https://github.com/huggingface/transformers/blob/main/utils/update_metadata.py """ + import argparse import os import tempfile