deadbear34/qwen35-4b-plantdisease-merged

VISIONConcurrency Cost:1Model Size:4.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:May 13, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

The deadbear34/qwen35-4b-plantdisease-merged model is a 4.54 billion parameter Qwen3.5-4B architecture, developed by deadbear34, specifically fine-tuned for plant disease diagnostics and agricultural advice. This merged model combines a Continued Pre-Training (CPT) base with a Supervised Fine-Tuning (SFT) LoRA adapter, supporting both English and Indonesian languages. It excels in domain-specific knowledge, offering specialized responses for identifying and addressing plant health issues.

Loading preview...

Model Overview

This model, deadbear34/qwen35-4b-plantdisease-merged, is a specialized Qwen3.5-4B variant with approximately 4.54 billion parameters. It integrates a Continued Pre-Training (CPT) base (deadbear34/qwen35-4b-plantdisease-cpt) and a Supervised Fine-Tuning (SFT) LoRA adapter (deadbear34/qwen35-4b-plantdisease-sft-lora) into a single set of weights. The tokenizer is sourced from Qwen/Qwen3.5-4B-Base, maintaining a vocabulary of 248,044.

Key Capabilities

  • Plant Disease Diagnostics: Highly specialized in identifying and providing information on plant diseases.
  • Agricultural Advice: Offers guidance related to agricultural practices, particularly concerning plant health.
  • Multilingual Support: Capable of processing and generating responses in both English and Indonesian.
  • GatedDeltaNet Architecture: Utilizes a hybrid architecture for efficient performance.

Benchmark Performance

  • GSM8K (Math Reasoning): Achieved 90.00% on 300 samples.
  • MMLU (Biology/Medical-skewed): Scored 76.00% on a subset of 12 subjects (out of 57 total).
  • PlantDisease ROUGE-1: Demonstrated 38.39% on 100 domain-specific samples in English and Indonesian.
  • HumanEval (Coding): Scored 42.68%, indicating a regression in coding capabilities compared to its base model.

Intended Use Cases

This model is ideal for applications requiring precise, domain-specific knowledge in plant pathology and agriculture, such as automated diagnostic tools for farmers, educational resources on plant health, or agricultural support systems.

Finetunes
1 models