You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

DARJYO-Agri-7B

A 7-billion parameter foundation model purpose-built for African soil science, trained from scratch.

Architecture (Unique Design)

Feature DARJYO-Agri-7B Standard GPT-2
Hidden size 4096 4096
Intermediate size 14336 (3.5x) 16384 (4x)
FFN ratio 3.5:1 4:1
Activation SiLU (gated) GELU
Layer strategy Hybrid (4 zones) Uniform
Vocabulary Custom 32k (African soil) 50k general
Tokenizer BPE (soil-optimized) BPE (general)

Layer Strategy

Layers Type Purpose
0-7 Full Global Attention Continental soil patterns
8-15 Local Window Attention Regional texture analysis
16-23 Dilated Attention Long-range chemical gradients
24-31 Hybrid Global/Local Integrated soil profiling

Custom Vocabulary

32000 tokens optimized for African agriculture:

  • African geography (countries, regions, soil zones)
  • Soil science terminology (texture, chemistry, classification)
  • Crop types (cabbage, tomato, capsicum, chillies, marigold flowers etc.)
  • Agricultural practices (irrigation, tillage, fertilization)
  • DARJYO-specific domain tokens

Files

File Description
darjyo-agri-7b-F16.gguf FP16 GGUF model
config.json Custom architecture configuration
vocab.json Custom African soil vocabulary
tokenizer.json Custom tokenizer
tokenizer_config.json Tokenizer configuration
special_tokens_map.json Special token mappings

Usage

from llama_cpp import Llama
llm = Llama("persadian/darjyo-agri-7b-F16.gguf", n_ctx=512)

Unique Metadata

  • darjyo.version: 1.0.0
  • darjyo.region: africa
  • darjyo.domain: soil_science
  • darjyo.ffn_ratio: 3.5
  • darjyo.layer_strategy: hybrid
  • darjyo.vocab_domain: african_agriculture_soil

Development

  • Developer: persadian
  • Built for: DARJYO
  • Corpus: ISRIC WoSIS + DARJYO field samples
  • Region: Sub-Saharan Africa
  • Status: Base model (pre-training weights initialized)

DOI

DOI

@misc{persadian2026,
    author       = {Darshani Persadh (~persadian)},
    title        = {darjyo-agri-7b},
    year         = 2026,
    url          = {https://huggingface.co/persadian/darjyo-agri-7b},
    doi          = {10.57967/hf/9056},
    publisher    = {Hugging Face}
}

Built by persadian for DARJYO - Growing Africa's agricultural intelligence

Downloads last month
581
GGUF
Hardware compatibility
Log In to add your hardware

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support