hilarl/naturecode-mangrove-small

Naturecode Mangrove Small

Expert AI model for mangrove ecosystem knowledge with function calling capabilities.

Model Description

Naturecode Mangrove Small is a specialized language model trained on comprehensive mangrove ecosystem data. It combines deep domain knowledge about mangrove ecology, conservation, and climate science with function calling capabilities, making it ideal for building mangrove monitoring and research applications.

Key Features

• Domain Expert: Trained on extensive mangrove ecosystem knowledge
• Function Calling: Can invoke tools for data retrieval and analysis
• Conservation Focus: Understands mangrove importance for climate and biodiversity
• Lightweight: 270M parameters for edge deployment

Usage

General Knowledge

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("google/functiongemma-270m-it")
model = PeftModel.from_pretrained(base_model, "hilarl/naturecode-mangrove-small")
tokenizer = AutoTokenizer.from_pretrained("google/functiongemma-270m-it")

prompt = "<start_of_turn>user\nWhat is a mangrove ecosystem?<end_of_turn>\n<start_of_turn>model\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=150)
print(tokenizer.decode(outputs[0]))

Function Calling

prompt = """<start_of_turn>user
[mangrove_api]
{"name": "get_species_info", "description": "Get information about a mangrove species", "parameters": {"species_name": "string"}}

Tell me about Rhizophora mangle<end_of_turn>
<start_of_turn>model
"""

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
# Output: <start_function_call>call:get_species_info{species_name:<escape>Rhizophora mangle<escape>}<end_function_call>

Knowledge Domains

• Species Information: Mangrove tree species, characteristics, distribution
• Ecosystem Services: Carbon sequestration, coastal protection, biodiversity
• Conservation: Threats, restoration techniques, protected areas
• Climate Science: Blue carbon, sea level rise adaptation
• Monitoring: Remote sensing, field survey methods

Training Details

• Base Model: google/functiongemma-270m-it
• Training Method: Mixed training (40% function calling, 60% general knowledge)
• Training Data: Curated mangrove knowledge base + function calling dataset

Intended Use

• Mangrove research assistants
• Conservation planning tools
• Educational applications about coastal ecosystems
• Mangrove monitoring system integration
• Climate science applications

Limitations

• Focused on mangrove ecosystems; general knowledge is limited
• Function calling requires proper tool descriptions
• Knowledge based on training data cutoff

License

Apache 2.0

Citation

@misc{naturecode-mangrove-small,
  author = {Naturecode},
  title = {Naturecode Mangrove Small},
  year = {2025},
  publisher = {HuggingFace},
}