Use encode_prompt()
Browse files
app.py
CHANGED
|
@@ -129,16 +129,8 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
| 129 |
if not high_vram:
|
| 130 |
fake_diffusers_current_device(text_encoder, gpu) # since we only encode one text - that is one model move and one encode, offload is same time consumption since it is also one load and one encode.
|
| 131 |
load_model_as_complete(text_encoder_2, target_device=gpu)
|
| 132 |
-
|
| 133 |
-
llama_vec, clip_l_pooler
|
| 134 |
-
|
| 135 |
-
if cfg == 1:
|
| 136 |
-
llama_vec_n, clip_l_pooler_n = torch.zeros_like(llama_vec), torch.zeros_like(clip_l_pooler)
|
| 137 |
-
else:
|
| 138 |
-
llama_vec_n, clip_l_pooler_n = encode_prompt_conds(n_prompt, text_encoder, text_encoder_2, tokenizer, tokenizer_2)
|
| 139 |
-
|
| 140 |
-
llama_vec, llama_attention_mask = crop_or_pad_yield_mask(llama_vec, length=512)
|
| 141 |
-
llama_vec_n, llama_attention_mask_n = crop_or_pad_yield_mask(llama_vec_n, length=512)
|
| 142 |
|
| 143 |
# Processing input image
|
| 144 |
|
|
@@ -174,10 +166,6 @@ def worker(input_image, prompt, n_prompt, seed, total_second_length, latent_wind
|
|
| 174 |
|
| 175 |
# Dtype
|
| 176 |
|
| 177 |
-
llama_vec = llama_vec.to(transformer.dtype)
|
| 178 |
-
llama_vec_n = llama_vec_n.to(transformer.dtype)
|
| 179 |
-
clip_l_pooler = clip_l_pooler.to(transformer.dtype)
|
| 180 |
-
clip_l_pooler_n = clip_l_pooler_n.to(transformer.dtype)
|
| 181 |
image_encoder_last_hidden_state = image_encoder_last_hidden_state.to(transformer.dtype)
|
| 182 |
|
| 183 |
# Sampling
|
|
|
|
| 129 |
if not high_vram:
|
| 130 |
fake_diffusers_current_device(text_encoder, gpu) # since we only encode one text - that is one model move and one encode, offload is same time consumption since it is also one load and one encode.
|
| 131 |
load_model_as_complete(text_encoder_2, target_device=gpu)
|
| 132 |
+
|
| 133 |
+
[llama_vec, clip_l_pooler, llama_vec_n, clip_l_pooler_n, llama_attention_mask, llama_attention_mask_n] = encode_prompt(prompt, n_prompt)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 134 |
|
| 135 |
# Processing input image
|
| 136 |
|
|
|
|
| 166 |
|
| 167 |
# Dtype
|
| 168 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
image_encoder_last_hidden_state = image_encoder_last_hidden_state.to(transformer.dtype)
|
| 170 |
|
| 171 |
# Sampling
|