qwen3-4b-vl-lora-vision-connector

Vision LoRA Adapter

This is a LoRA adapter for vision-language models, trained to adapt vision tower and connector layers in addition to language model layers.

Model Details

• Base Model: Qwen/Qwen3-VL-4B-Instruct
• LoRA Rank: 16
• LoRA Alpha: 32
• Target Modules:
  • Language Model: ✓
  • Vision Tower: ✓
  • Connector/Projector: ✓

Usage with vLLM

from vllm import LLM
from vllm.lora.request import LoRARequest

# Load model with LoRA support
llm = LLM(
    model="Qwen/Qwen3-VL-4B-Instruct",
    enable_lora=True,
    max_loras=1,
    max_lora_rank=16,
)

# Generate with LoRA
lora_request = LoRARequest("adapter", 1, "prashanth058/qwen3-4b-vl-lora-vision-connector")
outputs = llm.generate(
    prompts=["<your prompt>"],
    lora_request=lora_request,
)

Usage with Transformers + PEFT

from transformers import AutoModelForVision2Seq, AutoProcessor
from peft import PeftModel

# Load base model
model = AutoModelForVision2Seq.from_pretrained("Qwen/Qwen3-VL-4B-Instruct")
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-4B-Instruct")

# Load adapter
model = PeftModel.from_pretrained(model, "prashanth058/qwen3-4b-vl-lora-vision-connector")

# Generate
# ... (process your inputs)
outputs = model.generate(**inputs)

Training Details

This adapter was trained to demonstrate vision layer adaptation capabilities in vLLM.

• Dataset: Synthetic/small-scale training data
• Training: PEFT LoRA with vision layer targeting
• Purpose: Testing and demonstration

License

This adapter follows the license of the base model: Qwen/Qwen3-VL-4B-Instruct

Citation

If you use this adapter, please cite:

@misc{vision-lora-adapter,
  author = {vLLM Team},
  title = {Vision LoRA Adapter for Qwen/Qwen3-VL-4B-Instruct},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/prashanth058/qwen3-4b-vl-lora-vision-connector}},
}