run.py

from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig
from PIL import Image
import requests
import torch
import json
import re
import os
from torch.nn.utils.rnn import pad_sequence
from datasets import load_dataset
import re


dataset = load_dataset('moondream/analog-clock-benchmark', split='test', streaming=True)


result_location = './molmo-answers.json'
model_path = 'moondream/Molmo-72B-0924'
processor = AutoProcessor.from_pretrained(
    model_path,
    trust_remote_code=True,
    torch_dtype='auto',
    device_map='auto'
)

model = AutoModelForCausalLM.from_pretrained(
    model_path,
    trust_remote_code=True,
    torch_dtype='auto',
    device_map='auto'
)

def extract_points_and_text(input_string):
    # match the points attributes with single or double quotes
    points_pattern = r'(?:x|y)(\d*)=["\']?([\d.]+)["\']?'
    
    matches = re.findall(points_pattern, input_string)
    
    # Group matches by their index (or empty string if no index)
    point_dict = {}
    for index, value in matches:
        point_dict.setdefault(index, []).append(float(value))
    
    # Convert matches to the desired format
    points = [point for point in point_dict.values() if len(point) == 2]

    text_pattern = r'<(?:point|points)[^>]*>(.*?)</(?:point|points)>'
    text_match = re.search(text_pattern, input_string)
    text = text_match.group(1) if text_match else ""
    
    cleaned_string = re.sub(text_pattern, text, input_string)
    
    # Find all integers in the cleaned string
    answers = [int(num) for num in re.findall(r'\b\d+\b', cleaned_string)]
    
    result = {
        "points": points,
        "cleaned_string": cleaned_string,
        "answers": answers
    }
    
    return result

def do_inference(image, text):
    inputs = processor.process(
        images=[image],
        text=text
    )

    # move inputs to the correct device and make a batch of size 1
    inputs = {k: v.to(model.device).unsqueeze(0) for k, v in inputs.items()}

    # generate output; maximum 200 new tokens; stop generation when <|endoftext|> is generated
    inputs["images"] = inputs["images"].to(torch.float16)
    output = model.generate_from_batch(
        inputs,
        GenerationConfig(max_new_tokens=200, stop_strings="<|endoftext|>", attention_type='flash'),
        tokenizer=processor.tokenizer
    )

    # only get generated tokens; decode them to text
    generated_tokens = output[0,inputs['input_ids'].size(1):]
    generated_text = processor.tokenizer.decode(generated_tokens, skip_special_tokens=True)

    return generated_text

    # cleaned_obj = extract_points_and_text(generated_text)

    # return cleaned_obj



# create results file if it doesn't exist
if not os.path.exists(result_location):
    os.makedirs(os.path.dirname(result_location), exist_ok=True)
    with open(result_location, 'w') as file:
        json.dump([], file)

def has_one_clock(image):
    molmo_res_text = do_inference(image, 'How many clocks are in this image?')
    print('Clock count -> ', molmo_res_text)
    
    if molmo_res_text.strip().endswith("1.") and molmo_res_text.strip().startswith(" Counting the"):
        return True
    else:
        return False
    

def get_time(image):
    return do_inference(image, '''Look at this image with a readable clock and do the following:
1. Report the hour hand location
2. Report the minute hand location
3. Report the time based on these locations''')





# Create progress bar
datapoint_index = 0
for datapoint in dataset:
    molmo_time_string = get_time(datapoint['image'])
    print(molmo_time_string)
    print('\n\n\n\n\n')

    #skip if the string is empty or function returned false (molmo could not tell the time)

    # Check if molmo_time_string contains exactly one time formatted as HH:MM
    time_pattern = r'\b\d{1,2}:\d{2}\b'
    times_found = re.findall(time_pattern, molmo_time_string)

    if len(times_found) == 1 and times_found[0] != "10:10":
        obj_to_save = {
            'datapoint_index': datapoint_index,
            'correct_answer': molmo_time_string,
            'reported_times':  times_found,
            'llm_string': molmo_time_string
        }
        with open(result_location, 'r+') as file:
            data = json.load(file)
            data.append(obj_to_save)
            file.seek(0)
            json.dump(data, file, indent=4)
            file.truncate()