quyet498/fine-turn-ColQwen2-vn-direct

ColPali Fine-tuned for Vietnamese Documents

This model is a fine-tuned version of vidore/colqwen2.5-v0.2 for Vietnamese document retrieval.

Model Details

• Base Model: vidore/colqwen2.5-v0.2
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Target Language: Vietnamese
• Task: Document Retrieval

Usage

from transformers import AutoModel, AutoConfig
from colpali_engine.models import ColQwen2, ColQwen2Processor

# Method 1: AutoModel
model = AutoModel.from_pretrained("quyet498/fine-turn-ColQwen2-vn-direct")

# Method 2: ColQwen2 (recommended)
model = ColQwen2.from_pretrained("quyet498/fine-turn-ColQwen2-vn-direct")
processor = ColQwen2Processor.from_pretrained("quyet498/fine-turn-ColQwen2-vn-direct")

Training Details

• Original adapter repository: quyet498/fine-turn-ColQwen2-vn
• Fine-tuning approach: LoRA with rank 32
• Target modules: (.(model)(?!.visual).(down_proj|gate_proj|up_proj|k_proj|q_proj|v_proj|o_proj).$|.(custom_text_proj).$)

Notes

This repository contains both the base model files and the fine-tuned adapter weights, making it a complete, self-contained model ready for use.