VictorMylle/Kohya-ss-sd-scripts
1
0
Fork 1
mirror of https://github.com/kohya-ss/sd-scripts.git synced 2026-04-08 22:35:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix clip_skip not work in weight capt, sample gen

Browse Source
This commit is contained in:
Kohya S
2023-06-08 22:02:46 +09:00
parent 4e24733f1c
commit 045cd38b6e
2 changed files with 12 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
library/custom_train_functions.py
View File

@@ -265,11 +265,6 @@ def get_unweighted_text_embeddings(
text_embedding = enc_out["hidden_states"][-clip_skip]
text_embedding = text_encoder.text_model.final_layer_norm(text_embedding)
# cover the head and the tail by the starting and the ending tokens
text_input_chunk[:, 0] = text_input[0, 0]
text_input_chunk[:, -1] = text_input[0, -1]
text_embedding = text_encoder(text_input_chunk, attention_mask=None)[0]
if no_boseos_middle:
if i == 0:
# discard the ending token
@@ -284,7 +279,12 @@ def get_unweighted_text_embeddings(
text_embeddings.append(text_embedding)
text_embeddings = torch.concat(text_embeddings, axis=1)
else:
if clip_skip is None or clip_skip == 1:
text_embeddings = text_encoder(text_input)[0]
else:
enc_out = text_encoder(text_input, output_hidden_states=True, return_dict=True)
text_embeddings = enc_out["hidden_states"][-clip_skip]
text_embeddings = text_encoder.text_model.final_layer_norm(text_embeddings)
return text_embeddings

10
library/lpw_stable_diffusion.py
View File

@@ -245,11 +245,6 @@ def get_unweighted_text_embeddings(
text_embedding = enc_out["hidden_states"][-clip_skip]
text_embedding = pipe.text_encoder.text_model.final_layer_norm(text_embedding)
# cover the head and the tail by the starting and the ending tokens
text_input_chunk[:, 0] = text_input[0, 0]
text_input_chunk[:, -1] = text_input[0, -1]
text_embedding = pipe.text_encoder(text_input_chunk, attention_mask=None)[0]
if no_boseos_middle:
if i == 0:
# discard the ending token
@@ -264,7 +259,12 @@ def get_unweighted_text_embeddings(
text_embeddings.append(text_embedding)
text_embeddings = torch.concat(text_embeddings, axis=1)
else:
if clip_skip is None or clip_skip == 1:
text_embeddings = pipe.text_encoder(text_input)[0]
else:
enc_out = pipe.text_encoder(text_input, output_hidden_states=True, return_dict=True)
text_embeddings = enc_out["hidden_states"][-clip_skip]
text_embeddings = pipe.text_encoder.text_model.final_layer_norm(text_embeddings)
return text_embeddings