`n_min`>0 gives an error


https://github.com/fallenshock/FlowEdit/blob/f9d988c139ea130845cbedf153a64b3dbae31e29/FlowEdit_utils.py#L207-L225

For an obvious reason:
```
File "FlowEdit/FlowEdit_utils.py", line 223, in FlowEditSD3
    prev_sample = prev_sample.to(noise_pred_tar.dtype)
NameError: name 'noise_pred_tar' is not defined
```

I would also suggest changing 
https://github.com/fallenshock/FlowEdit/blob/f9d988c139ea130845cbedf153a64b3dbae31e29/FlowEdit_utils.py#L215
to
`src_tar_latent_model_input = torch.cat([xt_src, xt_src, xt_tar, xt_tar]) if pipe.do_classifier_free_guidance else (xt_src, xt_tar)` for consistency, and to avoid misleading code readers


	else: # i >= T_steps-n_min # regular sampling for last n_min steps

	if i == T_steps-n_min:
	# initialize SDEDIT-style generation phase
	fwd_noise = torch.randn_like(x_src).to(x_src.device)
	xt_src = scale_noise(scheduler, x_src, t, noise=fwd_noise)
	xt_tar = zt_edit + xt_src - x_src

	src_tar_latent_model_input = torch.cat([xt_tar, xt_tar, xt_tar, xt_tar]) if pipe.do_classifier_free_guidance else (xt_src, xt_tar)

	_, Vt_tar = calc_v_sd3(pipe, src_tar_latent_model_input,src_tar_prompt_embeds, src_tar_pooled_prompt_embeds, src_guidance_scale, tar_guidance_scale, t)

	xt_tar = xt_tar.to(torch.float32)

	prev_sample = xt_tar + (t_im1 - t_i) * (Vt_tar)

	prev_sample = prev_sample.to(noise_pred_tar.dtype)

	xt_tar = prev_sample

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`n_min`>0 gives an error #16

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

n_min>0 gives an error #16

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

`n_min`>0 gives an error #16