|
from typing import Dict, List, Any |
|
import torch |
|
from torch import autocast |
|
from huggingface_hub import hf_hub_download |
|
from diffusers import DiffusionPipeline |
|
import base64 |
|
from io import BytesIO |
|
|
|
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
|
print("device ~>", device) |
|
|
|
|
|
class EndpointHandler: |
|
def __init__(self, path=""): |
|
print("path ~>", path) |
|
|
|
self.pipe = DiffusionPipeline.from_pretrained( |
|
"stabilityai/stable-diffusion-xl-base-1.0", |
|
torch_dtype=torch.float16 if device.type == "cuda" else None, |
|
variant="fp16", |
|
).to(device) |
|
|
|
self.pipe.load_lora_weights("SvenN/sdxl-emoji", weight_name="lora.safetensors") |
|
self.pipe.fuse_lora() |
|
|
|
embedding_path = hf_hub_download( |
|
repo_id="SvenN/sdxl-emoji", filename="embeddings.pti", repo_type="model" |
|
) |
|
state_dict = load_file(embedding_path) |
|
|
|
self.pipe.load_textual_inversion(state_dict["text_encoders_0"], token=["<s0>", "<s1>"], text_encoder=pipeline.text_encoder, tokenizer=pipeline.tokenizer) |
|
self.pipe.load_textual_inversion(state_dict["text_encoders_1"], token=["<s0>", "<s1>"], text_encoder=pipeline.text_encoder_2, tokenizer=pipeline.tokenizer_2) |
|
|
|
def __call__(self, data: Any) -> List[List[Dict[str, float]]]: |
|
""" |
|
Args: |
|
data (:obj:): |
|
includes the input data and the parameters for the inference. |
|
Return: |
|
A :obj:`dict`:. base64 encoded image |
|
""" |
|
inputs = data.pop("inputs", data) |
|
|
|
|
|
full_prompt = f"A <s0><s1> emoji {inputs}" |
|
images = self.pipe( |
|
full_prompt, |
|
cross_attention_kwargs={"scale": 0.8}, |
|
num_inference_steps=25 |
|
).images |
|
image = images[0] |
|
|
|
return image |
|
|
|
|
|
if __name__ == "__main__": |
|
handler = EndpointHandler() |
|
print(handler) |
|
output = handler({"inputs": "emoji of a tiger face, white background"}) |
|
print(output) |
|
|
