Model Details

• Developed by: [Your Name/Organization]
• Model type: Multimodal Large Language Model (Vision-Language)
• Language(s): English
• Finetuned from model: unsloth/Qwen2-VL-7B-Instruct
• Finetuning approach: LoRA (Low-Rank Adaptation)

Training Details

Training Data

Fine-tuned on the images split of the Docmatix dataset, which focuses on document understanding and visual question answering.

Training Hyperparameters

• Method: SFT (Supervised Fine-Tuning) with LoRA
• LoRA Rank (r): 8
• LoRA Alpha: 16
• Optimizer: AdamW (8-bit)
• Learning Rate: 1e-4
• Batch Size: 1 (with Gradient Accumulation steps: 8)
• Max Steps: 200
• Precision: fp16/bf16 (depending on hardware compatibility)

How to Get Started with the Model

Loading the LoRA Adapter

python
Copy code
from unsloth import FastVisionModel
import torch

model, tokenizer = FastVisionModel.from_pretrained(
    "unsloth/Qwen2-VL-7B-Instruct",
    load_in_4bit=True,
)

model = FastVisionModel.load_adapter(model, "path_to_your_lora_files")

Inference Example

python
Copy code
from transformers import TextStreamer

FastVisionModel.for_inference(model)

# Standard Qwen2-VL inference code follows...

Framework versions

• PEFT 0.19.1
• Unsloth 2026.5.8
• Transformers 5.0.0
• PyTorch 2.11.0