Base Model: unsloth/Llama-3.2-1B-Instruct (meta-llama/Llama-3.2-1B-Instruct)
Method: QLoRA (4-bit quantization)
LoRA Rank: 16
LoRA Alpha: 32
Training Framework: Unsloth + HuggingFace Transformers
Fine-tuned for: Kwanzaa cultural expertise, historical accuracy, citation generation

Training Details

This adapter was trained to provide accurate, well-cited responses about:

Kwanzaa principles (Nguzo Saba) and their applications
Historical context and cultural significance
Cultural contributions and community practices
Proper citation of primary and secondary sources

Dataset Composition

Citation Examples: Proper source attribution and formatting
Grounded Answers: Factual responses with evidence
Cultural Contributions: Historical and contemporary contributions
Format Compliance: Consistent response formatting
Refusal Patterns: Appropriate handling of out-of-scope queries

Training Configuration

Epochs: 3-4
Learning Rate: 2e-4
Batch Size: 2
Gradient Accumulation: 4-8
Max Sequence Length: 2048
Optimizer: AdamW (8-bit)
Scheduler: Cosine with warmup
LoRA Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Usage

With PEFT

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = "unsloth/Llama-3.2-1B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    device_map="auto",
    torch_dtype="auto"
)
tokenizer = AutoTokenizer.from_pretrained(base_model)

# Load adapter
model = PeftModel.from_pretrained(model, "ainativestudio/ainative-adapter-v1")

# Generate response
prompt = """What is the principle of Umoja and how is it applied in daily life?

Please provide citations from primary sources."""

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.9,
    do_sample=True
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

With Unsloth (for training/inference)

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="ainativestudio/ainative-adapter-v1",
    max_seq_length=2048,
    dtype=None,  # Auto-detect
    load_in_4bit=True,
)

FastLanguageModel.for_inference(model)  # Enable inference mode

# Use the model
inputs = tokenizer("What are the Seven Principles of Kwanzaa?", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Performance

The adapter is optimized for:

Accurate cultural knowledge representation
Proper citation formatting
Grounded, evidence-based responses
Appropriate scope handling (refusals for out-of-domain queries)

Limitations

Trained specifically on Kwanzaa cultural knowledge
May not perform well on general-purpose tasks
Requires base model knowledge for broader reasoning
Best used with retrieval-augmented generation (RAG) for up-to-date information

Integration with AINative Platform

This adapter is designed to work with:

ZeroDB: Vector database for semantic search
RAG Pipeline: Enhanced with retrieved primary sources
Agent Swarm: Multi-agent coordination for complex queries

Citation Format

Responses include citations in this format:

[Source: Author Last Name, "Title" (Year), page/section]

License

Apache 2.0

Citation

If you use this adapter, please cite:

bibtex

@misc{ainative-kwanzaa-adapter-v1,
  title={AINative Platform Adapter v1 - Kwanzaa Knowledge},
  author={AINative Studio},
  year={2026},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/ainativestudio/ainative-adapter-v1}}
}

Contact

For questions or issues:

Repository: kwanzaa-project
Platform: AINative