Qwen3-VL-4B-Instruct-Flowers102-Open-QA

This model is a fine-tuned version of Qwen/Qwen3-VL-4B-Instruct on a custom dataset.

Model Details

• Base Model: Qwen/Qwen3-VL-4B-Instruct
• Fine-tuned from checkpoint: 2500
• Model Type: Vision-Language Model
• Training Framework: LMMS-Engine with FSDP2

Training Details

Training Data

Custom dataset based on Oxford Flowers102 found here.

Training Hyperparameters

• Per Device Train Batch Size: 1
• Learning Rate: 2e-05
• Weight Decay: 0.0
• Gradient Accumulation Steps: 8
• Gradient Checkpointing: True
• Warmup Ratio: 0.1
• Max Steps: 5000
• Number of Epochs: 1
• LR Scheduler: cosine
• Optimizer: adamw_torch_fused
• Mixed Precision (BF16): True
• FSDP2: True
• Save Steps: 500
• Logging Steps: 1

Training Configuration

• Checkpoint Step: 2500
• Training Framework: LMMS-Engine
• Distributed Training: FSDP2 (Fully Sharded Data Parallel v2)

Usage

from transformers import AutoProcessor, AutoModel
import torch

# Load model and processor
processor = AutoProcessor.from_pretrained("oscarqjh/Qwen3-VL-4B-Instruct-Flowers102-Open-QA")
model = AutoModel.from_pretrained("oscarqjh/Qwen3-VL-4B-Instruct-Flowers102-Open-QA", torch_dtype=torch.bfloat16)

# Example usage
# Your inference code here

Training Framework

This model was trained using LMMS-Engine, a framework for training large multimodal models.

Limitations and Bias

This model inherits the limitations and potential biases from its base model Qwen/Qwen3-VL-4B-Instruct. Please refer to the base model's documentation for detailed information about limitations and ethical considerations.

Citation

If you use this model, please cite the original base model and the training framework used.