NanoLlama is a small-scale language model based on the Llama architecture, designed for lightweight applications and resource-constrained environments. This model provides a good balance between performance and computational efficiency.

Developed by: sailesh-duddupudi-nutanix
Model type: Causal Language Model
Language(s): English
License: Apache 2.0
Finetuned from: Llama architecture
Access: Public (Open Access)

Model Architecture

Architecture: LlamaForCausalLM
Hidden Size: 4096
Number of Layers: 4
Number of Attention Heads: 4
Number of Key-Value Heads: 2
Vocabulary Size: 32000
Max Position Embeddings: 4096
Hidden Activation: SiLU

Uses

Direct Use

This model can be used for:

Text generation
Conversational AI
Code completion
Creative writing
Question answering

Downstream Use

The model can be fine-tuned for specific tasks such as:

Domain-specific text generation
Task-specific instruction following
Specialized conversational agents

How to Get Started with the Model

python

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model and tokenizer
model_name = "sailesh-duddupudi-nutanix/nanollama-public"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Generate text
input_text = "Hello, how are you?"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, temperature=0.7)
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_text)

Technical Specifications

Model Architecture and Objective

The model uses the standard Llama architecture with:

RMSNorm for layer normalization
RoPE (Rotary Position Embedding) for positional encoding
SwiGLU activation function
Grouped Query Attention (GQA)

Performance Characteristics

Model Size: Compact design for efficient deployment
Memory Requirements: Optimized for low-memory environments
Inference Speed: Fast inference suitable for real-time applications

Limitations

Limited context length (4096 tokens)
May not perform as well as larger models on complex reasoning tasks
Primarily trained/fine-tuned for English text

Citation

If you use this model, please cite:

bibtex

@misc{nanollama2024,
  title={NanoLlama: A Compact Llama-based Language Model},
  author={sailesh-duddupudi-nutanix},
  year={2024},
  url={https://huggingface.co/sailesh-duddupudi-nutanix/nanollama-public}
}

Contact

For questions or issues, please contact: sailesh-duddupudi-nutanix