import gradio as gr
import requests
import os
import base64
from PIL import Image
import io
import json
def resize_image(image_path, max_size=(512, 512), quality=85):
with Image.open(image_path) as img:
img.thumbnail(max_size, Image.Resampling.LANCZOS)
buffer = io.BytesIO()
img.save(buffer, format="JPEG", quality=quality)
return buffer.getvalue()
def filepath_to_base64(image_path):
img_bytes = resize_image(image_path)
img_base64 = base64.b64encode(img_bytes)
return f"data:image/jpeg;base64,{img_base64.decode('utf-8')}"
api_key = os.getenv('API_KEY')
def call_neva_22b_api(image_path, content, temperature=0.2, top_p=0.7, max_tokens=512, quality=6, humor=0, creativity=6, helpfulness=6):
print(f"Caminho da imagem recebida: {image_path}")
print(f"Conteúdo: {content}")
# Imprime os novos parâmetros
print(f"Quality: {quality}, Humor: {humor}, Creativity: {creativity}, Helpfulness: {helpfulness}")
image_base64 = filepath_to_base64(image_path)
invoke_url = "https://api.nvcf.nvidia.com/v2/nvcf/pexec/functions/8bf70738-59b9-4e5f-bc87-7ab4203be7a0"
headers = {
"Authorization": f"Bearer {api_key}",
"accept": "text/event-stream",
"content-type": "application/json",
}
payload = {
"messages": [
{
"content": f"{content} ",
"role": "user"
},
{
"labels": {
"creativity": creativity,
"helpfulness": helpfulness,
"humor": humor,
"quality": quality
},
"role": "assistant"
}
],
"temperature": temperature,
"top_p": top_p,
"max_tokens": max_tokens,
"stream": True
}
response = requests.post(invoke_url, headers=headers, json=payload, stream=True)
if response.status_code != 200:
print(f"Erro na requisição: {response.status_code}")
try:
error_details = response.json()
print(error_details)
except ValueError:
print(response.text)
else:
response_text = ""
for line in response.iter_lines():
if line:
try:
# Decode the line from bytes to string
decoded_line = line.decode('utf-8')
# Remove the "data: " prefix
if decoded_line.startswith('data: '):
json_str = decoded_line[6:] # Remove the first 6 characters ('data: ')
json_line = json.loads(json_str)
# Assuming the structure is consistent with the examples you provided.
content_parts = json_line.get("choices", [{}])[0].get("delta", {}).get("content", "")
response_text += content_parts
else:
print(f"Unexpected line format: {decoded_line}")
except json.JSONDecodeError as e:
print(f"Error decoding JSON from response line: {e}")
print(f"Faulty line: {line}")
return response_text
content_input = gr.Textbox(lines=2, placeholder="Enter your content here...", label="Content")
image_input = gr.Image(type="filepath", label="Upload Image")
quality_input = gr.Slider(minimum=0, maximum=9, step=1, value=6, label="Quality")
humor_input = gr.Slider(minimum=0, maximum=9, step=1, value=0, label="Humor")
creativity_input = gr.Slider(minimum=0, maximum=9, step=1, value=6, label="Creativity")
helpfulness_input = gr.Slider(minimum=0, maximum=9, step=1, value=6, label="Helpfulness")
temperature_input = gr.Slider(minimum=0, maximum=1, step=0.01, value=0.2, label="Temperature")
top_p_input = gr.Slider(minimum=0, maximum=1, step=0.01, value=0.7, label="Top P")
max_tokens_input = gr.Slider(minimum=1, maximum=512, step=1, value=512, label="Max Tokens")
iface = gr.Interface(fn=call_neva_22b_api,
inputs=[image_input, content_input, temperature_input, top_p_input, max_tokens_input, quality_input, humor_input, creativity_input, helpfulness_input],
outputs="text",
title="NEVA 22B DEMO",
description="""
Dive into the next generation of AI with NeVA-22B, an advanced multi-modal vision-language model that redefines the boundaries of technology. Developed with a 22 billion parameter architecture, NeVA-22B excels in understanding and generating responses that incorporate both text and images, offering a groundbreaking platform for multi-modal AI exploration.
How to Use:
Empowered by NVIDIA's cutting-edge AI technologies, NeVA-22B API Explorer opens up new avenues for engaging with multi-modal AI, accessible to everyone at no cost.
HF Created by: @artificialguybr (Twitter)
Explore further: artificialguy.com
""" ) iface.launch()