Instructions to use me-nabi/farmer-advisory-hindi-qwen2.5-3b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use me-nabi/farmer-advisory-hindi-qwen2.5-3b with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="me-nabi/farmer-advisory-hindi-qwen2.5-3b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("me-nabi/farmer-advisory-hindi-qwen2.5-3b", device_map="auto")

PEFT
How to use me-nabi/farmer-advisory-hindi-qwen2.5-3b with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use me-nabi/farmer-advisory-hindi-qwen2.5-3b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "me-nabi/farmer-advisory-hindi-qwen2.5-3b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "me-nabi/farmer-advisory-hindi-qwen2.5-3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/me-nabi/farmer-advisory-hindi-qwen2.5-3b

SGLang

How to use me-nabi/farmer-advisory-hindi-qwen2.5-3b with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "me-nabi/farmer-advisory-hindi-qwen2.5-3b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "me-nabi/farmer-advisory-hindi-qwen2.5-3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "me-nabi/farmer-advisory-hindi-qwen2.5-3b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "me-nabi/farmer-advisory-hindi-qwen2.5-3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use me-nabi/farmer-advisory-hindi-qwen2.5-3b with Docker Model Runner:
```
docker model run hf.co/me-nabi/farmer-advisory-hindi-qwen2.5-3b
```

🌾 HindiKrishi — Farmer Crop Advisory AI

Fine-tuned on 21K examples. Giving Indian farmers crop advice in Hindi.

Model Details

Parameter	Value
Base Model	Qwen/Qwen2.5-3B-Instruct
Method	QLoRA (4-bit quantization + LoRA r=16)
Training Data	21,069 instruction-response pairs
Languages	Hindi (primary), English
Training Hardware	NVIDIA A100 80GB (RunPod)
Training Time	83 minutes
Final Training Loss	0.388
Epochs	2
Trainable Parameters	29.9M / 3.1B (0.96%)

What This Model Does

Answers crop disease identification questions in Hindi with specific chemical names
Recommends pesticides with exact dosage (ml/liter, kg/hectare)
Covers 25+ major Indian crops including wheat, rice, cotton, sugarcane, pulses, vegetables, and fruits
Provides information on government agricultural schemes (PMFBY, PM-KISAN, PMKSY)
Gives season-specific farming advice (Kharif, Rabi, Zayad)
Every response includes a KVK consultation disclaimer for safety

Quick Start

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel

# Load base model in 4-bit
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True,
)

base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-3B-Instruct",
    quantization_config=bnb_config,
    device_map="auto",
)

# Load fine-tuned LoRA adapter
model = PeftModel.from_pretrained(base_model, "me-nabi/farmer-advisory-hindi-qwen2.5-3b")
tokenizer = AutoTokenizer.from_pretrained("me-nabi/farmer-advisory-hindi-qwen2.5-3b")
model.eval()

# Ask a question
SYSTEM = """आप भारतीय किसानों के लिए एक कृषि विशेषज्ञ सहायक हैं।
आप फसल रोग, कीट प्रबंधन, उर्वरक और खेती की सलाह देते हैं।
हमेशा व्यावहारिक सलाह दें और स्थानीय KVK से पुष्टि करने की सलाह दें।"""

question = "सरसों में माहू कीट के लिए कौन सा कीटनाशक छिड़कें?"

prompt = f"<|im_start|>system\n{SYSTEM}<|im_end|>\n<|im_start|>user\n{question}<|im_end|>\n<|im_start|>assistant\n"

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
with torch.no_grad():
    outputs = model.generate(
        **inputs, 
        max_new_tokens=200, 
        temperature=0.7, 
        do_sample=True,
        repetition_penalty=1.1,
    )

print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Example Outputs

Hindi — Crop Disease Identification

Q: गेहूं में पीला रतुआ रोग के लक्षण क्या हैं?

A: गेहूं में पीला रतुआ रोग के लक्षण में पौधों की पत्तियाँ पीली हो जाती हैं। इसे नियंत्रित करने के लिए कापर ऑक्सीक्लोराइड (100 ग्राम प्रति 200 लीटर पानी) का छिड़काव करें। अधिक जानकारी के लिए अपने नजदीकी KVK से संपर्क करें।

Hindi — Pesticide Recommendation

Q: सरसों में माहू कीट के लिए कौन सा कीटनाशक छिड़कें?

A: सरसों में माहू कीट के लिए Imidacloprid का उपयोग करें। इसकी मात्रा 0.5-1 ग्राम प्रति लीटर पानी में मिलाकर छिड़कें। यह छिड़काव बुवाई के 20-30 दिन बाद करें। अधिक जानकारी के लिए अपने नजदीकी KVK से संपर्क करें।

Hindi — Sowing Schedule

Q: रबी सीजन में गेहूं की बुवाई कब करनी चाहिए?

A: रबी सीजन में गेहूं की बुवाई 15-20 अक्टूबर से 10-15 नवंबर के बीच करनी चाहिए। अधिक जानकारी के लिए अपने नजदीकी KVK से संपर्क करें।

English — Pest Control

Q: How to control aphids in mustard crop?

A: Aphid infestations can be controlled by spraying with insecticides like Malathion 50% EC or Quinalphos 25% EC at a rate of 1.0-1.5 liters per hectare.

English — Fertilizer Advice

Q: What fertilizer should I use for wheat?

A: A balanced fertilizer with nitrogen (N), phosphorus (P), and potassium (K) is ideal. The recommended NPK fertilizers are 60 kg/ha of urea (46% N), 20 kg/ha of superphosphate, and 15 kg/ha of muriate of potash. Additionally, apply 50 kg/ha of farmyard manure to enhance soil fertility.

Training Data

The model was trained on 21,069 instruction-response pairs from multiple sources:

Source	Examples	Type
GPT-4o-mini generated from ICAR PDFs	4,335	Hindi crop disease Q&A with specific dosages
GPT-4o-mini targeted generation (crops × subtopics)	4,726	Hindi — 25 crops, diseases, pesticides, schemes
KisanVaani agriculture Q&A	22,615 → deduplicated	English agriculture Q&A
Mahesh2841/Agriculture	5,916	English agriculture Q&A
DigiGreen + CGIAR	2,274	Hindi agriculture translations
Vikaspedia (scraped via Playwright)	295	Hindi government advisory content
Indic Anudesh (AI4Bharat)	7,577	Hindi instruction-following data

All data was cleaned, deduplicated, and converted to Qwen ChatML format for training.

Training Configuration

LoRA Config:
  r: 16
  lora_alpha: 16
  target_modules: [q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj]
  dropout: 0
  bias: none

Training Args:
  per_device_train_batch_size: 32
  gradient_accumulation_steps: 1
  learning_rate: 2e-4
  num_train_epochs: 2
  optimizer: adamw_8bit
  lr_scheduler: cosine
  warmup_steps: 20
  max_seq_length: 512
  fp16: true
  gradient_checkpointing: true

Training Loss Curve

Step 10: 1.495 Step 50: 0.645 Step 100: 0.555 Step 200: 0.477 Step 500: 0.434 Step 800: 0.408 Step 1000: 0.388 Step 1250: 0.348 Final: 0.388

Crops Covered

The model has been trained on data covering these Indian crops:

Cereals: Wheat (गेहूं), Rice (धान), Maize (मक्का), Bajra, Jowar

Pulses: Chickpea (चना), Pigeon pea (अरहर), Black gram (उड़द), Green gram (मूंग)

Oilseeds: Mustard (सरसों), Soybean (सोयाबीन), Groundnut (मूंगफली)

Cash Crops: Cotton (कपास), Sugarcane (गन्ना)

Vegetables: Tomato (टमाटर), Onion (प्याज), Chili (मिर्च), Brinjal (बैंगन), Okra (भिंडी), Potato (आलू), Cauliflower (गोभी), Spinach (पालक)

Fruits: Mango (आम), Banana (केला), Grapes (अंगूर), Orange (संतरा), Guava (अमरूद)

Spices: Turmeric (हल्दी), Ginger (अदरक), Garlic (लहसुन), Coriander (धनिया), Cumin (जीरा)

Limitations

Hindi responses occasionally suggest incorrect chemical names for specific diseases — always verify with local KVK before application
Government scheme details (amounts, percentages, dates) may not reflect the latest policy updates
Model performs better on common crops (wheat, rice, cotton) than less common ones
Responses are advisory only — not a substitute for professional agricultural consultation
Dosage recommendations should be cross-checked with product labels and local KVK guidelines
Not designed for emergency pest outbreak situations

Intended Use

Intended users:

Agricultural extension workers providing advice to farmers
Developers building mobile apps for offline crop advisory in rural India
Researchers studying multilingual domain-specific LLM fine-tuning
Students and educators demonstrating AI applications in agriculture

Not intended for:

Direct pesticide application without KVK verification
Medical or veterinary advice
Legal or financial agricultural decisions
Replacing trained agricultural scientists

Environmental Impact

Hardware: NVIDIA A100 80GB
Training Time: 83 minutes
Cloud Provider: RunPod
Estimated Carbon Emitted: ~0.15 kg CO2eq

Technical Architecture

Base Model: Qwen 2.5 3B Instruct (3,115,872,256 parameters) ↓ 4-bit NF4 Quantization (BitsAndBytes) ↓ LoRA Adapters (29,933,568 trainable parameters) ↓ Fine-tuned on 21,069 agricultural Q&A pairs ↓ ChatML format: system + user + assistant

Related Projects

Aksara (CropIn AI): India-specific crop advisory model
KisanVaani: Agricultural Q&A dataset
AI4Bharat Indic Instruct: Hindi instruction-following data

Citation

@misc{hindikrishi2026,
  title={HindiKrishi: A Fine-tuned Hindi Crop Advisory Model for Indian Farmers Using QLoRA},
  author={Md Ehtasham Nabi},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/me-nabi/farmer-advisory-hindi-qwen2.5-3b}
}

Author

Md Ehtasham Nabi

Founding AI/ML Engineer at Ksham (ksham.in)

Disclaimer

⚠️ This model is for informational purposes only. Agricultural advice varies significantly by region, soil type, climate, and season. Always confirm pesticide recommendations, dosages, and application methods with your local Krishi Vigyan Kendra (KVK) or qualified agricultural extension officer before application. Incorrect pesticide usage can damage crops, harm the environment, and pose health risks.

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for me-nabi/farmer-advisory-hindi-qwen2.5-3b

Base model

Qwen/Qwen2.5-3B

Finetuned

Qwen/Qwen2.5-3B-Instruct