app.py

"""
Adapted from https://huggingface.co/spaces/stabilityai/stable-diffusion
"""

from tensorflow import keras

import time

import gradio as gr
import keras_cv

from constants import css, examples, img_height, img_width, num_images_to_gen
from share_btn import community_icon_html, loading_icon_html, share_js

from huggingface_hub import from_pretrained_keras

import requests

# MODEL_CKPT = "chansung/textual-inversion-pipeline@v1673026791"
# MODEL = from_pretrained_keras(MODEL_CKPT)

# model = keras_cv.models.StableDiffusion(
#     img_width=img_width, img_height=img_height, jit_compile=True
# )
# model._text_encoder = MODEL
# model._text_encoder.compile(jit_compile=True)

# # Warm-up the model.
# _ = model.text_to_image("Teddy bear", batch_size=num_images_to_gen)

def generate_image_fn(prompt: str, unconditional_guidance_scale: int) -> list:
    start_time = time.time()
    # `images is an `np.ndarray`. So we convert it to a list of ndarrays.
    # Each ndarray represents a generated image.
    # Reference: https://gradio.app/docs/#gallery
    images = model.text_to_image(
        prompt,
        batch_size=num_images_to_gen,
        unconditional_guidance_scale=unconditional_guidance_scale,
    )
    end_time = time.time()
    print(f"Time taken: {end_time - start_time} seconds.")
    return [image for image in images]

demoInterface = gr.Interface(
    generate_image_fn,
    inputs=[
        gr.Textbox(
            label="Enter your prompt",
            max_lines=1,
#            placeholder="cute Sundar Pichai creature",
        ),
        gr.Slider(value=40, minimum=8, maximum=50, step=1),
    ],
    outputs=gr.Gallery().style(grid=[2], height="auto"),
    # examples=[["cute Sundar Pichai creature", 8], ["Hello kitty", 8]],
    allow_flagging=False,
)

def avaliable_providers():
    providers = []
    
    headers = {
        "Content-Type": "application/json",
    }
    endpoint_url = "https://api.endpoints.huggingface.cloud/provider"
    response = requests.get(endpoint_url, headers=headers)

    for provider in response.json()['items']:
        if provider['status'] == 'avaliable':
            providers.append(provider['vendor'])
    
    return providers
    

with gr.Blocks() as demo:
    gr.Markdown(
    """
    # Your own Stable Diffusion on Google Cloud Platform
    """)
    
    with gr.Row():
        gcp_project_id = gr.Textbox(
            label="GCP project ID",
        )
        gcp_region = gr.Dropdown(
            ["us-central1", "asia‑east1", "asia-northeast1"],
            value="us-central1",
            interactive=True,
            label="GCP Region"
        )

    gr.Markdown(
    """
    Configurations on scalability
    """)        
    with gr.Row():
        min_nodes = gr.Slider(
            label="minimum number of nodes",
            minimum=1,
            maximum=10)
        
        max_nodes = gr.Slider(
            label="maximum number of nodes",
            minimum=1,
            maximum=10)
    
    btn = gr.Button(value="Ready to Deploy!")
    # btn.click(mirror, inputs=[im], outputs=[im_2])    

def update_regions(provider):
    avalialbe_regions = []
    
    headers = {
        "Content-Type": "application/json",
    }
    endpoint_url = f"https://api.endpoints.huggingface.cloud/provider/{provider}/region"
    response = requests.get(endpoint_url, headers=headers)

    for region in response.json()['items']:
        if region['status'] == 'avaliable':
            avalialbe_regions.append(f"{region['region']}/{region['label']}")

    return gr.Dropdown.update(
        choices=avalialbe_regions,
        value=avalialbe_regions[0] if len(avalialbe_regions) > 0 else None
    )

def update_compute_options(provider, regeion):
    avalialbe_compute_options = []

    headers = {
        "Content-Type": "application/json",
    }
    endpoint_url = f"https://api.endpoints.huggingface.cloud/provider/{provider}/region/{region}/compute"
    response = requests.get(endpoint_url, headers=headers)

    for compute in response.json()['items']:
        if compute['status'] == 'avaliable':
            type = f"{compute['numAccelerators']}vCPU {compute['memoryGb'].replace("Gi", "GB")} · {compute['architecture']}" 
                   if compute['accelerator'] == "cpu"
                   else  f"{compute['numAccelerators']x {compute['architecture']}}"
            
            avalialbe_compute_options.append(
                f"{compute['accelerator'].upper()} [{compute['instanceSize']}] · {type}"
            )

    return gr.Dropdown.update(
        choices=avalialbe_compute_options,
        value=avalialbe_compute_options[0] if len(avalialbe_compute_options) > 0 else None
    )


with gr.Blocks() as demo2:
    gr.Markdown(
    """
    ## Your own Stable Diffusion on Hugging Face 🤗 Endpoint
    """)
    hf_token_input = gr.Textbox(
        label="enter your Hugging Face Access Token"
    )
        
    providers = avaliable_providers()
    
    with gr.Row():
        provider_selector = gr.Dropdown(
            providers
            label="select cloud provider",
            interactive=True,
        )
        
        region_selector = gr.Dropdown(
            [],
            value="",
            interactive=True,
            label="select a region"
        )
        
        provider_selector.change(update_regions, inputs=provider_selector, outputs=region_selector)

    compute_selector = gr.Dropdown(
        [],
        value="",
        label="select compute instance type"
        interactive=True
    )
    region_selector.change(update_compute_options, inputs=[provider_selector, region_selector], outputs=compute_selector)

gr.TabbedInterface(
    [demoInterface, demo, demo2], ["Try-out", "🚀 Deploy on GCP", " Deploy on 🤗 Endpoint"]
).launch(enable_queue=True)