StoryGold/bella-v1.0-13b

Blueprint Extraction Large Learning Agent (BELLA) v1.0

Finetuned on LLaVA 1.6 Vicuna 13B on CADCoder dataset to convert image to CadQuery code.

Evaluation

The model was evaluated on the 100 example test split of the CADCODER/GenCAD-Code dataset.

Metric	Score
Intersection over Union (IoU)	0.7505
Valid Generated Solid Rate	0.98
ROUGE-L	0.8454
BLEU	0.7456

How to Use

import torch
from transformers import AutoProcessor, LlavaNextForConditionalGeneration
from PIL import Image
import os
from datasets import load_dataset

model_path = "StoryGold/bella-v1.0-13b"
dataset_name = "CADCODER/GenCAD-Code"

dataset = load_dataset(dataset_name, split='test', streaming=True)
example = next(iter(dataset))

image = example['image']
prompt_text = example['prompt']
ground_truth_code = example['cadquery']

print("Dataset example loaded successfully.")
print(f"Prompt: '{prompt_text}'")


model = LlavaNextForConditionalGeneration.from_pretrained(
        model_path,
        torch_dtype=torch.bfloat16,
        device_map="auto")
processor = AutoProcessor.from_pretrained(model_path)

chat_template = (
    "A chat between a curious user and an artificial intelligence assistant. "
    "The assistant gives helpful, detailed, and polite answers to the user's questions. "
    f"USER: <image>\n{prompt_text} ASSISTANT:"
)

inputs = processor(text=chat_template, images=image, return_tensors="pt").to(model.device)

with torch.inference_mode():
    output_ids = model.generate(**inputs, max_new_tokens=4096, do_sample=False)

output_text = processor.decode(output_ids[0], skip_special_tokens=True)