Llama 3 8B Vision Captioning

Published by @caption_lab · Community adapter

image-to-text #vision #captioning #alt-text #multimodal

Image captioning adapter producing detailed, accessibility-grade alt text. Describes scene, subjects, spatial relationships, mood, and notable visual elements. Trained on COCO and Conceptual Captions.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
base = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
model = PeftModel.from_pretrained(base, "modelforgelab/llama3-8b-vision-caption-lora")

inputs = tokenizer("Your prompt here", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Compatibility

transformers>=4.40