|
|
--- |
|
|
library_name: diffusers |
|
|
--- |
|
|
created using the following code based on https://github.com/huggingface/diffusers/blob/main/tests/pipelines/hidream/test_pipeline_hidream.py |
|
|
|
|
|
```python |
|
|
import numpy as np |
|
|
import torch |
|
|
from transformers import ( |
|
|
AutoTokenizer, |
|
|
CLIPTextConfig, |
|
|
CLIPTextModelWithProjection, |
|
|
CLIPTokenizer, |
|
|
LlamaForCausalLM, |
|
|
T5EncoderModel, |
|
|
) |
|
|
|
|
|
from diffusers import ( |
|
|
AutoencoderKL, |
|
|
FlowMatchEulerDiscreteScheduler, |
|
|
HiDreamImagePipeline, |
|
|
HiDreamImageTransformer2DModel, |
|
|
) |
|
|
def get_dummy_components(): |
|
|
torch.manual_seed(0) |
|
|
transformer = HiDreamImageTransformer2DModel( |
|
|
patch_size=2, |
|
|
in_channels=4, |
|
|
out_channels=4, |
|
|
num_layers=1, |
|
|
num_single_layers=1, |
|
|
attention_head_dim=8, |
|
|
num_attention_heads=4, |
|
|
caption_channels=[32, 16], |
|
|
text_emb_dim=64, |
|
|
num_routed_experts=4, |
|
|
num_activated_experts=2, |
|
|
axes_dims_rope=(4, 2, 2), |
|
|
max_resolution=(32, 32), |
|
|
llama_layers=(0, 1), |
|
|
).eval() |
|
|
torch.manual_seed(0) |
|
|
vae = AutoencoderKL(scaling_factor=0.3611, shift_factor=0.1159) |
|
|
clip_text_encoder_config = CLIPTextConfig( |
|
|
bos_token_id=0, |
|
|
eos_token_id=2, |
|
|
hidden_size=32, |
|
|
intermediate_size=37, |
|
|
layer_norm_eps=1e-05, |
|
|
num_attention_heads=4, |
|
|
num_hidden_layers=5, |
|
|
pad_token_id=1, |
|
|
vocab_size=1000, |
|
|
hidden_act="gelu", |
|
|
projection_dim=32, |
|
|
max_position_embeddings=128, |
|
|
) |
|
|
|
|
|
torch.manual_seed(0) |
|
|
text_encoder = CLIPTextModelWithProjection(clip_text_encoder_config) |
|
|
|
|
|
torch.manual_seed(0) |
|
|
text_encoder_2 = CLIPTextModelWithProjection(clip_text_encoder_config) |
|
|
|
|
|
torch.manual_seed(0) |
|
|
text_encoder_3 = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5") |
|
|
|
|
|
torch.manual_seed(0) |
|
|
text_encoder_4 = LlamaForCausalLM.from_pretrained("hf-internal-testing/tiny-random-LlamaForCausalLM") |
|
|
text_encoder_4.generation_config.pad_token_id = 1 |
|
|
tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") |
|
|
tokenizer_2 = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") |
|
|
tokenizer_3 = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5") |
|
|
tokenizer_4 = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-LlamaForCausalLM") |
|
|
|
|
|
scheduler = FlowMatchEulerDiscreteScheduler() |
|
|
|
|
|
components = { |
|
|
"scheduler": scheduler, |
|
|
"vae": vae, |
|
|
"text_encoder": text_encoder, |
|
|
"tokenizer": tokenizer, |
|
|
"text_encoder_2": text_encoder_2, |
|
|
"tokenizer_2": tokenizer_2, |
|
|
"text_encoder_3": text_encoder_3, |
|
|
"tokenizer_3": tokenizer_3, |
|
|
"text_encoder_4": text_encoder_4, |
|
|
"tokenizer_4": tokenizer_4, |
|
|
"transformer": transformer, |
|
|
} |
|
|
return components |
|
|
|
|
|
|
|
|
if __name__ == "__main__": |
|
|
components = get_dummy_components() |
|
|
pipeline = HiDreamImagePipeline(**components) |
|
|
pipeline.push_to_hub("hf-internal-testing/tiny-hidream-i1-pipe") |
|
|
|
|
|
``` |
|
|
|
