experimental/clip_app_client.py

# File name: graph_client.py
from concurrent.futures import ThreadPoolExecutor
import json
import os
import numpy as np
import requests
from concurrent.futures import ThreadPoolExecutor, as_completed
import time

import torch

# hack for debugging, set HTTP_ADDRESS to "http://127.0.0.1:8000/"
# os.environ["HTTP_ADDRESS"] = "http://192.168.7.79:8000"

test_image_url = "https://static.wixstatic.com/media/4d6b49_42b9435ce1104008b1b5f7a3c9bfcd69~mv2.jpg/v1/fill/w_454,h_333,fp_0.50_0.50,q_90/4d6b49_42b9435ce1104008b1b5f7a3c9bfcd69~mv2.jpg"
english_text = (
    "It was the best of times, it was the worst of times, it was the age "
    "of wisdom, it was the age of foolishness, it was the epoch of belief"
)

clip_model="ViT-L/14"
clip_model_id ="laion5B-L-14"
device = "cuda:0" if torch.cuda.is_available() else "cpu"
print ("using device", device)
from clip_retrieval.load_clip import load_clip, get_tokenizer 
# from clip_retrieval.clip_client import ClipClient, Modality
model, preprocess = load_clip(clip_model, use_jit=True, device=device)
tokenizer = get_tokenizer(clip_model)

def preprocess_image(image_url):
    # download image from url
    import requests
    from PIL import Image
    from io import BytesIO
    response = requests.get(test_image_url)
    input_image = Image.open(BytesIO(response.content))
    input_image = input_image.convert('RGB')
    # convert image to numpy array
    input_image = np.array(input_image)
    input_im = Image.fromarray(input_image)
    prepro = preprocess(input_im).unsqueeze(0).cpu()
    return prepro

preprocessed_image = preprocess_image(test_image_url)

def send_text_request(number):
    payload = {
        "text": ('str', english_text, 'application/octet-stream'),
    }
    url = os.environ.get("HTTP_ADDRESS", "http://127.0.0.1:8000/")
    response = requests.post(url, files=payload)
    embeddings = response.text
    return number, embeddings

def send_image_url_request(number):
    payload = {
        "image_url": ('str', test_image_url, 'application/octet-stream'),
    }
    url = os.environ.get("HTTP_ADDRESS", "http://127.0.0.1:8000/")
    response = requests.post(url, files=payload)
    embeddings = response.text
    return number, embeddings

def send_preprocessed_image_request(number):
    key = "preprocessed_image"
    data_bytes = preprocessed_image.numpy().tobytes()
    shape_bytes = np.array(preprocessed_image.shape).tobytes()
    dtype_bytes = str(preprocessed_image.dtype).encode()
    payload = {
        key: ('tensor', data_bytes, 'application/octet-stream'),
        'shape': ('shape', shape_bytes, 'application/octet-stream'),
        'dtype': ('dtype', dtype_bytes, 'application/octet-stream'),
    }
    url = os.environ.get("HTTP_ADDRESS", "http://127.0.0.1:8000/")
    response = requests.post(url, files=payload)
    embeddings = response.text
    return number, embeddings

def process(numbers, send_func, max_workers=10):
    with ThreadPoolExecutor(max_workers=max_workers) as executor:
        futures = [executor.submit(send_func, number) for number in numbers]
        for future in as_completed(futures):
            n_result, result = future.result()
            result = json.loads(result)
            print (f"{n_result} : {len(result[0])}")

# def process_text(numbers, max_workers=10):
#     for n in numbers:
#         n_result, result = send_text_request(n)
#         result = json.loads(result)
#         print (f"{n_result} : {len(result[0])}")

if __name__ == "__main__":
    n_calls = 300

    # test text
    # n_calls = 1
    numbers = list(range(n_calls))
    start_time = time.monotonic()
    process(numbers, send_text_request)
    end_time = time.monotonic()
    total_time = end_time - start_time
    avg_time_ms = total_time / n_calls * 1000
    calls_per_sec = n_calls / total_time
    print(f"Text...")
    print(f" Average time taken: {avg_time_ms:.2f} ms")
    print(f" Number of calls per second: {calls_per_sec:.2f}")    

    # test image url
    # n_calls = 1
    numbers = list(range(n_calls))
    start_time = time.monotonic()
    process(numbers, send_image_url_request)
    end_time = time.monotonic()
    total_time = end_time - start_time
    avg_time_ms = total_time / n_calls * 1000
    calls_per_sec = n_calls / total_time
    print(f"Image passing url...")
    print(f" Average time taken: {avg_time_ms:.2f} ms")
    print(f" Number of calls per second: {calls_per_sec:.2f}")    

    # test image as vector
    # n_calls = 1
    numbers = list(range(n_calls))
    start_time = time.monotonic()
    process(numbers, send_preprocessed_image_request)
    end_time = time.monotonic()
    total_time = end_time - start_time
    avg_time_ms = total_time / n_calls * 1000
    calls_per_sec = n_calls / total_time
    print(f"Preprocessed image...")
    print(f" Average time taken: {avg_time_ms:.2f} ms")
    print(f" Number of calls per second: {calls_per_sec:.2f}")