BeejX_CoT

Model Card: BeejX CoT Agriculture Model

Model Information

• Model Name: BeejX CoT Agriculture Model (SmolLM3-3B)
• Base Architecture: SmolLM3 (3.08B parameters)
• Specialty: Agricultural knowledge with Chain-of-Thought reasoning
• Language: English
• License: Apache 2.0

Files in this Repository

BeejX_cot_model.gguf          (5.74 GB) - Original model, F16 precision
BeejX_cot_model_q8_0.gguf     (3.05 GB) - Q8_0 quantization, ~99% quality
BeejX_cot_model_q5_0.gguf     (2.01 GB) - Q5_0 quantization, recommended
chat_with_BeejX_cot.bat                - Windows batch script for easy use

Quick Usage Examples

Example 1: Disease Identification

Q: What are common diseases in maize and how can they be managed?

A: Understanding that maize (corn) is susceptible to various diseases...
[Provides detailed management practices including irrigation, fertilization, 
pest control, and disease prevention strategies]

Example 2: Late Blight in Potato

Q: What is late blight in potato and how can it be managed?

A: Potatoes are susceptible to fungal diseases such as late blight, 
powdery mildew, and stem canker that can cause significant yield loss...
[Provides information about disease characteristics and management]

Example 3: General Agriculture

Q: What are common diseases in guava and how can they be prevented?

A: Guava can suffer from various fungal, bacterial, and viral diseases, 
as well as nematode and root rot problems, especially in poorly drained soils...
[Continues with prevention and treatment methods]

Technical Specifications

Model Architecture

• Parameters: 3.08 billion
• Layers: 36 transformer layers
• Embedding Dimension: 2048
• Attention Heads: 16 (4 KV heads)
• Feed-forward Dimension: 11,008
• Context Window: 65,536 tokens (training), 4,096 typical usage
• Vocabulary Size: 128,256 tokens
• RoPE Frequency Base: 5,000,000

Quantization Details

• Q8_0: 8-bit quantization, minimal quality loss
• Q5_0: 5-bit quantization, balanced size/quality trade-off
• Quantization Tool: llama.cpp quantize

Performance Benchmarks

Inference Speed (CPU - 12 threads)

• Original (F16): ~15-20 tok/s
• Q8_0: ~20-25 tok/s
• Q5_0: ~25-30 tok/s

Memory Requirements

• Original: ~6 GB RAM
• Q8_0: ~3.5 GB RAM
• Q5_0: ~2.5 GB RAM

Training Information

Dataset

• Primary Focus: Agricultural knowledge
• Sources: ICAR data, agricultural research papers, farming guides
• Format: Chain-of-Thought reasoning format
• Training Method: Fine-tuning on SmolLM3-3B base

Training Features

• Step-by-step reasoning for complex questions
• Agricultural domain specialization
• Crop disease identification and management
• Farming best practices
• General knowledge retention

Ethical Considerations

Intended Use

• Educational purposes
• Agricultural research
• Farming knowledge assistance
• Decision support (with expert verification)

Limitations

1. Knowledge Cutoff: Based on training data available at time of fine-tuning
2. Regional Specificity: May not account for all regional agricultural practices
3. Verification Required: Always verify critical decisions with local agricultural experts
4. Language: Primarily English, may have limited multilingual capabilities

How to Use on HuggingFace

1. Download the model:

   • Choose your preferred quantization (Q5_0 recommended)
   • Download using HuggingFace Hub or direct download

2. Run with llama.cpp:

   llama-cli -m BeejX_cot_model_q5_0.gguf -p "What are diseases in maize?" -n 512
   

3. Use in Python:

   from llama_cpp import Llama
   llm = Llama(model_path="BeejX_cot_model_q5_0.gguf")
   response = llm("What are diseases in maize?", max_tokens=512)
   

Community & Support

• Issues: Report bugs or issues
• Discussions: Ask questions and share experiences
• Contributions: Suggestions for improvements welcome

Version History

• v1.0: Initial release with Q5_0 and Q8_0 quantizations
  • Base: SmolLM3-3B
  • Specialty: Agricultural CoT reasoning
  • Context: 4096 tokens

Citation

If you use this model in your research or applications, please cite:

@misc{BeejX-cot-agriculture-model,
  title={BeejX CoT Agriculture Model: Chain-of-Thought Reasoning for Agricultural Knowledge},
  author={[Sky1701/BeejX]},
  year={2025},
  publisher={HuggingFace},
  base_model={HuggingFaceTB/SmolLM3-3B}
}

---

Disclaimer: This model is for educational and research purposes. Always consult with agricultural experts and local extension services for critical farming decisions.