README.md

metadata

license: apache-2.0
base_model: unsloth/Qwen2.5-VL-3B-Instruct-bnb-4bit
merges:
  - adapter: final_model
    method: merge_and_unload
    date: 2025-06-04T00:00:00.000Z
tags:
  - vision-language
  - lora-merged
  - qwen2.5-vl

🐍 Merged Qwen2.5-VL Model (LoRA + Base)

This repository contains the merged weights of LoRA adapter located at ./final_model and the base model unsloth/Qwen2.5-VL-3B-Instruct-bnb-4bit.

The merge was performed with peft.merge_and_unload() on 2025-06-04.

Usage

from unsloth import FastVisionModel

model, tokenizer = FastVisionModel.from_pretrained("Litian2002/Qwen2.5-VL-3B-Spatial-bnb-4bit")
model = FastVisionModel.for_inference(model) # Enable native 2x faster inference

# Your inference code here

Or with transformers:

from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("Litian2002/Qwen2.5-VL-3B-Spatial-bnb-4bit", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("Litian2002/Qwen2.5-VL-3B-Spatial-bnb-4bit", trust_remote_code=True)