yizaochen/deepseek-ocr-3b-docvqa-lora

DeepSeek OCR 3B - DocVQA LoRA Adapter

This is a LoRA adapter trained on the DocVQA dataset for the DeepSeek OCR 3B model.

Model Description

• Base Model: unsloth/DeepSeek-OCR
• Training Dataset: DocVQA
• Training Method: LoRA (Low-Rank Adaptation) using Unsloth
• Task: Document Visual Question Answering

Usage

from peft import PeftModel
from transformers import AutoModelForVision2Seq, AutoTokenizer

# Load base model
base_model = AutoModelForVision2Seq.from_pretrained("unsloth/DeepSeek-OCR")
tokenizer = AutoTokenizer.from_pretrained("unsloth/DeepSeek-OCR")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "yizaochen/deepseek-ocr-3b-docvqa-lora")

# Use the model for inference
# ... your inference code here ...

Training Details

• Training samples: 2000 (from DocVQA training set)
• Validation: DocVQA validation set
• Framework: Unsloth + PyTorch

Citation

If you use this model, please cite the original DeepSeek OCR paper and the DocVQA dataset.