README.md

metadata

license: mit
base_model:
  - microsoft/Florence-2-large
datasets:
  - Ejafa/ye-pop
tags:
  - art
pipeline_tag: image-to-text
language:
  - en

microsoft/Florence-2-large tuned on Ejafa/ye-pop captioned with CogVLM2

This repository contains a fine-tuned version of the microsoft/Florence-2-large model. The model has been tuned on a 40,000 image subset of the Ejafa/ye-pop dataset, with captions generated using THUDM/cogvlm2-llama3-chat-19B.

Training Details

• Vision Encoder: The vision encoder was frozen during training.
• Batch Size: 64
• Gradient Accumulation Steps: 16
• Learning Rate: 5.12e-05
• Optimizer: AdamW
• Scheduler: polynomial
• Epochs: 7.37

Dataset

The fine-tuning process utilized a 40,000 image subset from the Ejafa/ye-pop dataset. This dataset contains a wide array of images with varying subjects, providing a robust training ground for improving the model's captioning abilities.

Captioning

The captions were generated using THUDM/cogvlm2-llama3-chat-19B.

Usage

To use this model, you can load it directly from the Hugging Face Model Hub:

from transformers import AutoModelForCausalLM, AutoProcessor, AutoConfig
import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = AutoModelForCausalLM.from_pretrained("thwri/CogFlorence-2.1-Large", trust_remote_code=True).to(device).eval()
processor = AutoProcessor.from_pretrained("thwri/CogFlorence-2.1-Large", trust_remote_code=True)
# Function to run the model on an example
def run_example(task_prompt, image):
    prompt = task_prompt
    # Ensure the image is in RGB mode
    if image.mode != "RGB":
        image = image.convert("RGB")
    inputs = processor(text=prompt, images=image, return_tensors="pt").to(device)
    generated_ids = model.generate(
        input_ids=inputs["input_ids"],
        pixel_values=inputs["pixel_values"],
        max_new_tokens=1024,
        num_beams=3,
        do_sample=True
    )
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
    parsed_answer = processor.post_process_generation(generated_text, task=task_prompt, image_size=(image.width, image.height))
    return parsed_answer
from PIL import Image
import requests
import copy
url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg?download=true"
image = Image.open(requests.get(url, stream=True).raw)
result = run_example("<MORE_DETAILED_CAPTION>" , image)
print(result)
# {'<MORE_DETAILED_CAPTION>': 'A vivid, close-up photograph of a classic car, specifically a Volkswagen Beetle, parked on a cobblestone street. The car is painted in a striking shade of turquoise, with a glossy finish that reflects the surrounding environment. The vehicle's rounded shape is accentuated by its rounded tires and chrome detailing. The background reveals a weathered yellow wall with a rustic wooden door, adding to the rustic charm of the scene. The sky above is clear, suggesting a sunny day. The overall style of the image is candid, capturing a moment in time without any posed or staged elements.'}