brimbim/agriculture-crop-disease-model-v2

Agriculture Crop Disease Advisor v2

92% win rate against base model. Built with AutoScientist Round 2 for the Adaption Labs AutoScientist Challenge (Agriculture Category).

A specialized LoRA adapter for smallholder farmers and agronomists, focused on accurate crop disease diagnosis and practical, low-cost recommendations.

Model Details

• Developed by: brimbim (using Adaption Labs AutoScientist)
• Model type: LoRA adapter (PEFT)
• Base model: mistralai/Mixtral-8x7B-Instruct-v0.1
• Language(s): English
• License: Apache 2.0
• Finetuned from: mistralai/Mixtral-8x7B-Instruct-v0.1

Capabilities

• Precise diagnosis from text descriptions of symptoms
• Practical treatment recommendations (strong emphasis on organic & low-cost options)
• Region-aware and climate-resilient advice
• Confidence scores, risk flags, and safety warnings
• Simple, farmer-friendly language

Training Details

• Dataset: ~20,000 high-quality instruction-completion pairs (PlantVillage + synthetic diverse examples)
• Method: LoRA (rank 64, alpha 128)
• Win Rate: 92% (adapted vs base model)
• Quality Score: 9.8/10 (Grade A)

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "brimbim/agriculture-crop-disease-model-v2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.bfloat16)

prompt = """You are an expert agronomist helping smallholder farmers.
A farmer in East Africa reports yellow spots with white centers on tomato leaves after heavy rain."""

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=400, temperature=0.7, top_p=0.9)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Uses

Intended Use:

Helping smallholder farmers and agronomists with crop disease diagnosis and management.

Out-of-Scope:

Medical advice for humans/animals, large-scale commercial farming systems, or any high-stakes legal/financial decisions.

Bias, Risks, and Limitations

• Model performance is strongest on common crops (tomato, maize, potato, etc.)
• May have reduced accuracy on rare diseases or crops not well-represented in training data
• Always verify critical recommendations with local agricultural extension services

Citation

If you use this model, please cite:bibtex

@misc{agriculture-crop-disease-advisor-v2, author = {brimbim}, title = {Agriculture Crop Disease Advisor v2}, year = {2026}, publisher = {Hugging Face}, howpublished = {\url{https://huggingface.co/brimbim/agriculture-crop-disease-model-v2}} }