app.py

import gradio as gr
from transformers import AutoProcessor, AutoModelForVision2Seq, TextIteratorStreamer
from threading import Thread
import re
import time
from PIL import Image
import torch
import spaces
import subprocess
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)

import requests
import torch
from PIL import Image
from transformers import AutoModelForCausalLM, AutoProcessor

from decord import VideoReader
from decord import cpu
from PIL import Image
import numpy as np

def load_video(video_path, frames=32):
    """
    Load a video and extract a specified number of frames as PIL.Image objects.
    
    Parameters:
    - video_path (str): Path to the video file.
    - frames (int): Number of frames to extract.
    
    Returns:
    - List[PIL.Image]: A list of PIL.Image objects for the extracted frames.
    """
    # Initialize VideoReader
    vr = VideoReader(video_path, ctx=cpu())
    total_frames = len(vr)
    
    # Select frame indices evenly spaced throughout the video
    frame_indices = np.linspace(0, total_frames - 1, frames, dtype=int)
    
    # Extract frames and convert to PIL.Images
    images = []
    for idx in frame_indices:
        frame = vr[idx]  # Get the frame as a NumPy array
        image = Image.fromarray(frame.asnumpy())  # Convert to PIL.Image
        images.append(image)
    
    return images

model_id_or_path = "teowu/Aria-Chat-Preview"

model = AutoModelForCausalLM.from_pretrained(model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16, 
        trust_remote_code=True)

processor = AutoProcessor.from_pretrained(model_id_or_path, trust_remote_code=True)

@spaces.GPU
def model_inference(
    input_dict, history, decoding_strategy, temperature, max_new_tokens, top_p
): 
    text = input_dict["text"]
    print(input_dict["files"])
    if len(input_dict["files"]) > 1:
      images = [Image.open(image).convert("RGB") for image in input_dict["files"]]
    elif len(input_dict["files"]) == 1:
      if input_dict["files"][0].endswith(".mp4") or input_dict["files"][0].endswith(".avi"):
        images = load_video(input_dict["files"][0])
      else:
        images = [Image.open(input_dict["files"][0]).convert("RGB")]   
    else:
      images = []
    

    if text == "" and not images:
        gr.Error("Please input a query and optionally image(s).")

    if text == "" and images:
        text = "Please provide a detailed description."
        #gr.Error("Please input a text query along the image(s).")

    


    resulting_messages = [
                {
                    "role": "user",
                    "content": [{"type": "image", "text": None} for _ in range(len(images))] + [
                        {"type": "text", "text": "\n" + text}
                    ]
                }
            ]
    prompt = processor.apply_chat_template(resulting_messages, add_generation_prompt=True)
    inputs = processor(text=prompt, images=images, return_tensors="pt")
    inputs = {k: v.to("cuda") for k, v in inputs.items()}
    generation_args = {
        "max_new_tokens": max_new_tokens,
        "repetition_penalty": repetition_penalty,
    }

    assert decoding_strategy in [
        "Greedy",
        "Top P Sampling",
    ]
    if decoding_strategy == "Greedy":
        generation_args["do_sample"] = False
    elif decoding_strategy == "Top P Sampling":
        generation_args["temperature"] = temperature
        generation_args["do_sample"] = True
        generation_args["top_p"] = top_p

    generation_args.update(inputs)
    # Generate
    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens= True)
    generation_args = dict(inputs, streamer=streamer, max_new_tokens=max_new_tokens)
    generated_text = ""

    thread = Thread(target=model.generate, kwargs=generation_args)
    thread.start()

    yield "..."
    buffer = ""
    
      
    for new_text in streamer:
    
      buffer += new_text
      generated_text_without_prompt = buffer#[len(ext_buffer):]
      time.sleep(0.01)
      yield buffer


examples=[
              [{"text": "What art era do these artpieces belong to?", "files": ["example_images/rococo.jpg", "example_images/rococo_1.jpg"]}, "Greedy", 0.4, 512, 1.2, 0.8],
              [{"text": "I'm planning a visit to this temple, give me travel tips.", "files": ["example_images/examples_wat_arun.jpg"]},  "Greedy", 0.4, 512, 1.2, 0.8],
              [{"text":  "What is the due date and the invoice date?", "files": ["example_images/examples_invoice.png"]}, "Greedy", 0.4, 512, 1.2, 0.8],
              [{"text": "What is this UI about?", "files": ["example_images/s2w_example.png"]},   "Greedy", 0.4, 512, 1.2, 0.8],
              [{"text":  "Where do the severe droughts happen according to this diagram?", "files": ["example_images/examples_weather_events.png"]},  "Greedy", 0.4, 512, 1.2, 0.8],
      ]
demo = gr.ChatInterface(fn=model_inference, title="Aria-Chat: Improved Real-world Abilties for Open-source LMMs on Images and Videos", 
                description="Play with [rhymes-ai/Aria-Chat-Preview](https://huggingface.co/rhymes-ai/Aria-Chat-Preview) in this demo. To get started, upload an image (or a video) and text or try one of the examples. This checkpoint works best with single turn conversations, so clear the conversation after a single turn.",
                examples=examples,
                textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image"], file_count="multiple"), stop_btn="Stop Generation", multimodal=True, 
                        additional_inputs=[gr.Radio(["Top P Sampling",
              "Greedy"],
          value="Greedy",
          label="Decoding strategy",
          #interactive=True,
          info="Higher values is equivalent to sampling more low-probability tokens.",
                                                  
      ), gr.Slider(
          minimum=0.0,
          maximum=5.0,
          value=0.4,
          step=0.1,
          interactive=True,
          label="Sampling temperature",
          info="Higher values will produce more diverse outputs.",
      ),
                                            gr.Slider(
          minimum=8,
          maximum=1024,
          value=512,
          step=1,
          interactive=True,
          label="Maximum number of new tokens to generate",
      ), 
         gr.Slider(
          minimum=0.01,
          maximum=0.99,
          value=0.8,
          step=0.01,
          interactive=True,
          label="Top P",
          info="Higher values is equivalent to sampling more low-probability tokens.",
      )],cache_examples=False
                )
     
      
      

demo.launch(debug=True)