Litxiong/Qv3-Plant_spotter-Vl-8b

Qv3-Plant_spotter-Vl-8b: Specialized Plant Disease Identification

Qv3-Plant_spotter-Vl-8b is an 8.77 billion parameter multimodal vision-language model, specifically fine-tuned for identifying plant diseases from images. Built upon the Qwen3-VL-8B-Instruct architecture, this model processes visual input of plant leaves to detect various conditions and diseases.

Key Capabilities & Features

• Specialized Vision-Language Understanding: Optimized for plant disease identification, accepting an image and a prompt to diagnose the condition.
• LoRA Fine-tuning: Utilizes Low-Rank Adaptation (LoRA) for efficient fine-tuning, targeting all linear layers in both the LLM and vision encoder.
• Comprehensive Training Data: Fine-tuned on diverse plant-related datasets, including PlantVillage and PlantDoc, totaling over 27,000 samples across approximately 100 disease classes.
• Robust Architecture: Features a Qwen3VLForConditionalGeneration architecture with a 262,144 token context length for the language model and a dedicated vision encoder.
• Efficient Performance: Achieved a final training loss of 0.266 in approximately 10.8 minutes on an NVIDIA L40 GPU.

Ideal Use Cases

• Agricultural Diagnostics: Rapid identification of plant diseases in both lab and field settings.
• Crop Monitoring: Assisting farmers and agronomists in early detection of plant health issues.
• Educational Tools: Developing applications for learning about plant pathology.

Limitations

• Primarily trained on specific datasets; performance may vary in highly complex real-world scenarios not represented in the training data.
• Not a replacement for professional agricultural diagnosis.