---
language:
- ur
- pa
- sd
- ps
- skr
license: apache-2.0
tags:
- agriculture
- pakistan
- urdu
- punjabi
- sindhi
- pashto
- saraiki
- instruction-tuning
- gemma
- farming
- multilingual
base_model: google/gemma-3-12b
datasets:
- KhetLink/Fasal-1B-Pak-Agri
pipeline_tag: text-generation
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: instruction
    dtype: string
  - name: output
    dtype: string
  - name: language
    dtype: string
  splits:
  - name: train
    num_bytes: 4034008
    num_examples: 2008
  download_size: 1915898
  dataset_size: 4034008
---

🌾 Fasal-1B — Pakistan's First Multilingual Agricultural AI

Fasal-1B is a fine-tuned instruction model for agricultural advisory in Pakistani languages. Built on Gemma 3-12B, it is the first AI model specifically trained to serve Pakistani farmers in their native languages.

 فصل (Fasal) means "crop" in Urdu.



Model Details

Base Model google/gemma-3-12b
Fine-tuning Instruction tuning (JSONL format) |
Dataset Version v5 
Dataset Size 2,857+ instruction entries 
Languages Urdu, Punjabi, Sindhi, Pashto, Saraiki 
Domain  Agriculture, crop advisory, pest management 
Training  Kaggle (in progress → v5) 
License | Apache 2.0 



 Languages Supported


 Urdu  `ur`  Nastaliq  
 Punjabi  `pa`  Shahmukhi 
 Sindhi  `sd`  Sindhi Arabic 
 Pashto  `ps`  Pashto Arabic 
 Saraiki  `skr`  Shahmukhi 



What It Does

Fasal-1B answers farm advisory questions including:

- 🌱 Crop diagnosis — identify diseases from symptoms
- 🐛 Pest management — identify pests, recommend treatments with dosages
- 💧 Irrigation advice — water scheduling by crop and season
- 🌿 Fertilizer guidance — NPK recommendations, deficiency diagnosis
- 🌦️ Weather-based advice — planting and harvest timing
- 💰 Market information — mandi prices, sell/hold decisions

---

 Example Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("KhetLink/Fasal-1B-Pak-Agri")
model = AutoModelForCausalLM.from_pretrained("KhetLink/Fasal-1B-Pak-Agri")

# Urdu query
prompt =<|system|>
آپ فصل AI ہیں، پاکستانی کسانوں کے لیے ایک ماہر زرعی مشیر۔
<|user|>
گندم کی فصل میں پیلے دھبے آ رہے ہیں، کیا کریں؟
<|assistant|>

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

Sample Output (Urdu):
گندم کی فصل میں پیلے دھبے آنا ایک عام مسئلہ ہے۔ یہ دھبے مختلف وجوہات کی 
بنا پر آ سکتے ہیں:

1. پانی کی کمی — فصل کو مناسب مقدار میں پانی دیں
2. نائٹروجن کی کمی — نائٹروجن کھاد استعمال کریں  
3. فنگل بیماری — Propiconazole فنگیسائیڈ کا چھڑکاؤ کریں
4. کیڑوں کا حملہ — کیڑوں کی پہچان کر کے مناسب علاج کریں
```

---

 Dataset

Training data: [`KhetLink/Fasal-1B-Pak-Agri`](https://huggingface.co/datasets/KhetLink/Fasal-1B-Pak-Agri)

Dataset composition (v5):
2,857+ instruction-response pairs
Covers: wheat, cotton, rice, sugarcane, vegetables, fruits
Topics: diseases, pests, fertilizers, irrigation, harvest, market
Format: JSONL instruction tuning
Generated with: Claude Haiku API + expert validation


Performance

Tested on real farmer queries via WhatsApp and Android app:


| Urdu | ⭐⭐⭐⭐⭐ | 100% |
| Punjabi | ⭐⭐⭐⭐ | 100% |
| Sindhi | ⭐⭐⭐⭐ | 100% |
| Pashto | ⭐⭐⭐⭐ | 100% |
| Saraiki | ⭐⭐⭐⭐ | 100% (v5) |

 Intended Use

 Farm advisory chatbots for Pakistani farmers  
 WhatsApp agricultural helplines  
 Android offline advisory apps  
 Government agriculture extension services  
 Research on low-resource Pakistani NLP  



 Limitations

Training data is AI-generated — real expert validation ongoing
Saraiki is lowest-resource language, may have occasional errors
Does not have real-time market prices (requires external API)
Not a substitute for certified agronomist advice in critical cases


Built By
KhetLink / Fasal Team 🇵🇰  
- HuggingFace: [KhetLink](https://huggingface.co/KhetLink)  
- GitHub: [sarfrazkhan18/fasal-android](https://github.com/sarfrazkhan18/fasal-android)  
- X: [@sarfrazkhan18](https://x.com/sarfrazkhan18)  

Pakistan's first multilingual agricultural AI — built by a farmer-region auditor turned AI entrepreneur from Sindh.
---

Citation

```bibtex
@misc{fasal1b2026,
  title={Fasal-1B: Pakistan's First Multilingual Agricultural AI},
  author={KhetLink Team},
  year={2026},
  url={https://huggingface.co/KhetLink/Fasal-1B-Pak-Agri}
}


If Fasal-1B helps you, please ⭐ star the dataset and model on HuggingFace.  
Every star helps Pakistani farmers get better AI tools.