a problem about the code,thanks

[aosong01]

it seems that you change all the basictransformerblock in both down_blocks, mid_blocks and up_blocks. why still change the up_blocks in the unet again?

def register_extended_attention(model):
    for _, module in model.unet.named_modules():
        if isinstance_str(module, "BasicTransformerBlock"):
            module.attn1.forward = sa_forward(module.attn1)

    res_dict = {1: [1, 2], 2: [0, 1, 2], 3: [0, 1, 2]}
    # we are injecting attention in blocks 4 - 11 of the decoder, so not in the first block of the lowest resolution
    for res in res_dict:
        for block in res_dict[res]:
            module = model.unet.up_blocks[res].attentions[block].transformer_blocks[0].attn1
            module.forward = sa_forward(module)

[ShashwatNigam99]

I have the same query

The first for loop modifies the following blocks:

 down_blocks.0.attentions.0.transformer_blocks.0
 down_blocks.0.attentions.1.transformer_blocks.0
 down_blocks.1.attentions.0.transformer_blocks.0
 down_blocks.1.attentions.1.transformer_blocks.0
 down_blocks.2.attentions.0.transformer_blocks.0
 down_blocks.2.attentions.1.transformer_blocks.0
 up_blocks.1.attentions.0.transformer_blocks.0
 up_blocks.1.attentions.1.transformer_blocks.0
 up_blocks.1.attentions.2.transformer_blocks.0
 up_blocks.2.attentions.0.transformer_blocks.0
 up_blocks.2.attentions.1.transformer_blocks.0
 up_blocks.2.attentions.2.transformer_blocks.0
 up_blocks.3.attentions.0.transformer_blocks.0
 up_blocks.3.attentions.1.transformer_blocks.0
 up_blocks.3.attentions.2.transformer_blocks.0
 mid_block.attentions.0.transformer_blocks.0

The second for loop modifies:

up_blocks.1.attentions.1.transformer_blocks.0.attn1
up_blocks.1.attentions.2.transformer_blocks.0.attn1
up_blocks.2.attentions.0.transformer_blocks.0.attn1
up_blocks.2.attentions.1.transformer_blocks.0.attn1
up_blocks.2.attentions.2.transformer_blocks.0.attn1
up_blocks.3.attentions.0.transformer_blocks.0.attn1
up_blocks.3.attentions.1.transformer_blocks.0.attn1
up_blocks.3.attentions.2.transformer_blocks.0.attn1

Which is a subset of the first for loop.

according to the comment, the first block of the lowest resolution shouldn't have extended attention registered. the first for loop registers extended attention for that block as well.

[Zeldalina]

同问

[edward3862]

I think the valid function should be register_extended_attention_pnp where a list injection_schedule is defined.

TokenFlow/tokenflow_utils.py

Lines 203 to 214 in 8ae24e9

	for _, module in model.unet.named_modules():
	if isinstance_str(module, "BasicTransformerBlock"):
	module.attn1.forward = sa_forward(module.attn1)
	setattr(module.attn1, 'injection_schedule', [])
	res_dict = {1: [1, 2], 2: [0, 1, 2], 3: [0, 1, 2]}
	# we are injecting attention in blocks 4 - 11 of the decoder, so not in the first block of the lowest resolution
	for res in res_dict:
	for block in res_dict[res]:
	module = model.unet.up_blocks[res].attentions[block].transformer_blocks[0].attn1
	module.forward = sa_forward(module)
	setattr(module, 'injection_schedule', injection_schedule)

The injection is activated according to injection_schedule.

TokenFlow/tokenflow_utils.py

Lines 124 to 130 in 8ae24e9

	if self.injection_schedule is not None and (self.t in self.injection_schedule or self.t == 1000):
	# inject unconditional
	q[n_frames:2 * n_frames] = q[:n_frames]
	k[n_frames:2 * n_frames] = k[:n_frames]
	# inject conditional
	q[2 * n_frames:] = q[:n_frames]
	k[2 * n_frames:] = k[:n_frames]

TokenFlow/tokenflow_utils.py

Lines 86 to 91 in 8ae24e9

	if self.injection_schedule is not None and (self.t in self.injection_schedule or self.t == 1000):
	source_batch_size = int(hidden_states.shape[0] // 3)
	# inject unconditional
	hidden_states[source_batch_size:2 * source_batch_size] = hidden_states[:source_batch_size]
	# inject conditional
	hidden_states[2 * source_batch_size:] = hidden_states[:source_batch_size]

BTW, I tried removing the first loop in L203-L206 and found the result was not changed. However, when removing the second loop in L208-L214, the result would get worse.