---

AgriParam

BharatGen introduces AgriParam, a domain-specialized large language model fine-tuned from Param-1-2.9B-Instruct on a high-quality, India-centric agriculture dataset.
AgriParam is designed to understand and generate contextually rich responses for agricultural queries, farmer advisories, policy information, research insights, and rural knowledge dissemination.

---

🌱 Motivation

Agriculture is the backbone of India’s economy, yet existing language models lack deep domain knowledge tailored to Indian contexts, languages, and cultural nuances.
AgriParam bridges this gap by combining Param-1’s bilingual capabilities with a meticulously curated agricultural knowledge base.

---

🏗 Model Architecture

AgriParam inherits the architecture of Param-1-2.9B-Instruct:

• Hidden size: 2048
• Intermediate size: 7168
• Attention heads: 16
• Hidden layers: 32
• Key-value heads: 8
• Max position embeddings: 2048
• Activation: SiLU
• Positional Embeddings: Rotary (RoPE, theta=10000)
• Attention Mechanism: Grouped-query attention
• Precision: bf16-mixed
• Base model: Param-1-2.9B-Instruct

---

📚 Data Preparation

AgriParam’s training corpus was carefully crafted to ensure deep agricultural knowledge, cultural relevance, and bilingual (English-Hindi) accessibility.

Steps involved:

1. Source Gathering

   • 17k open-source, India-focused agricultural news & information passages.

2. Question Generation

   • Generated 5 curated Q&A pairs per passage using an open-source LLM.

3. Domain Taxonomy & Personas

   • Built an exhaustive, India-specific agricultural taxonomy.
   • Defined farmer, policy-maker, scientist, and agri-business personas.

4. Dataset Construction

   • 2M Q&A pairs grounded in taxonomy and personas.
   • Complete dataset translated into Hindi.
   • 6M multi-turn conversation samples created.

---

🏋️ Training Setup

• Base model: Param-1-2.9B-Instruct
• Training framework: Hugging Face + torchrun multi-node setup
• Prompt template: Custom-designed for agricultural inference
• Scheduler: Linear with warmup
• Epochs: 3
• Total training samples: 12M
• Test samples: 1.2M
• Base learning rate: 5e-6
• Minimum learning rate: 0
• Additional tokens: <user>, <assistant>, <context>, <system_prompt>
• Vocab size: 256k + 4
• Global batch size: 1024
• Micro batch size: 4
• Gradient accumulation steps: 32

---

🚀 Inference Example

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "bharatgenai/AgriParam"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=False)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.bfloat32,
    device_map="auto"
)

# Example agricultural query
user_input = "What are the best practices for organic wheat farming in Uttar Pradesh?"

# 3 types of prompt
# 1. Generic QA: <user> ... <assistant>
# 2. Context based QA: <context> ... <user> ... <assistant>
# 3. Multi-turn conversation (supports upto 5 conversations): <user> ... <assistant> ... <user> ... <assistant>

# Based on your requirements use the type of prompt (refere the above examples)
prompt = f"<user> {user_input} <assistant>"
# prompt = f"<context> {user_context} <user> {user_input} <assistant>"
# prompt = f"<user> {user_input1} <assistant> {user_input2} <user> {user_input3} <assistant>..."
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=300,
        do_sample=True,
        top_k=50,
        top_p=0.95,
        temperature=0.6,
        eos_token_id=tokenizer.eos_token_id,
        use_cache=False
    )

print(tokenizer.decode(output[0], skip_special_tokens=True))

📊 Evaluation

• Crop-specific Q&A
• Policy & scheme awareness
• Rural advisory & extension services
• Bilingual (English/Hindi) capability

BhashaBench-Krishi (BBK)

Model	BBK	BBK_English	BBK_Hindi
Llama-3.2-1B	28.91	29.71	25.21
Llama-3.2-1B-Instruct	28.65	29.16	26.33
Llama-3.2-3B	31.96	32.68	28.69
granite-3.1-3b-a800m-base	32.17	33.36	26.70
sarvam-2b-v0.5	27.68	28.14	25.57
sarvam-1	30.24	30.82	27.57
AgriParam	32.18	33.10	27.97

---

Subject Domain Performance

Subject Domain	Llama-3.2-1B	Llama-3.2-1B-Instruct	Llama-3.2-3B	granite-3.1-3b-a800m-base	sarvam-2b-v0.5	sarvam-1	AgriParam
Agri-Environmental & Allied Disciplines	31.82	32.95	25.00	36.93	29.55	30.11	27.27
Agricultural Biotechnology	31.11	28.63	34.35	43.13	30.34	36.64	36.64
Agricultural Chemistry & Biochemistry	27.05	22.78	31.32	35.94	27.05	34.52	34.16
Agricultural Economics & Policy	29.98	25.52	35.09	34.77	27.75	30.78	32.54
Agricultural Engineering & Technology	27.46	26.23	32.79	30.33	27.46	29.51	27.87
Agricultural Extension Education	30.88	29.46	32.30	29.84	28.17	29.97	34.50
Agricultural Microbiology	34.23	36.04	31.53	34.23	17.12	26.13	34.23
Agriculture Communication	33.07	28.35	29.53	34.25	25.59	33.07	32.68
Agriculture Information Technology	30.53	31.58	44.21	36.84	27.89	32.11	27.89
Agronomy	27.92	28.77	31.84	31.51	28.67	29.60	32.49
Animal Sciences	25.68	34.46	36.49	37.84	35.14	29.05	40.54
Crop Sciences	31.15	26.41	29.87	35.15	26.59	29.33	32.42
Dairy & Poultry Science	35.96	31.46	30.34	44.94	33.71	32.58	29.21
Entomology	29.02	27.59	35.49	29.31	27.59	27.87	31.75
Fisheries and Aquaculture	29.41	41.18	38.24	26.47	20.59	14.71	23.53
General Knowledge & Reasoning	28.44	27.53	33.13	32.38	26.17	30.56	31.92
Genetics and Plant Breeding	30.59	30.08	28.02	29.05	26.99	31.62	29.82
Horticulture	27.05	28.60	31.21	32.17	27.00	29.76	31.40
Natural Resource Management	28.50	26.42	29.02	32.64	26.42	26.94	27.46
Nematology	22.83	28.26	28.26	27.17	21.20	24.46	23.91
Plant Pathology	28.97	30.48	27.96	29.97	25.44	33.50	25.44
Plant Sciences & Physiology	28.68	31.78	37.98	26.36	20.93	30.23	31.01
Seed Science and Technology	29.70	28.71	27.72	29.21	29.70	34.65	27.23
Soil Science	31.25	29.92	31.69	29.99	27.49	30.21	34.93
Veterinary Sciences	27.08	14.58	37.50	39.58	20.83	41.67	43.75

---

Question Level Difficulty

Difficulty	Llama-3.2-1B	Llama-3.2-1B-Instruct	Llama-3.2-3B	granite-3.1-3b-a800m-base	sarvam-2b-v0.5	sarvam-1	AgriParam
Easy	29.43	30.22	36.44	36.08	28.26	32.20	36.94
Hard	27.72	26.37	25.61	26.02	28.01	27.54	25.91
Medium	28.68	27.69	29.17	29.88	27.03	28.99	29.09

---