aquif-3-micro

A compact and efficient 434M parameter language model optimized for general-purpose text generation tasks. aquif-3-micro delivers competitive performance while maintaining a small footprint, making it ideal for resource-constrained environments and applications requiring fast inference.

Model Details

Parameter Count: 434 million parameters
Base Architecture: GPT-2 style transformer
License: Apache 2.0
Primary Focus: General text generation with emphasis on efficiency

Performance Benchmarks

aquif-3-micro demonstrates strong performance across multiple evaluation benchmarks compared to other models in similar parameter ranges:

Model	Parameters	GSM8K	MMLU	GPQA	Average
aquif-3-micro	434M	52.1	55.6	28.5	45.4
LFM2	742M	46.4	49.9	28.5	41.6
Gemma 3	1B	59.9	40.1	21.1	40.4
Llama 3.2	1B	35.7	46.6	28.8	37.0
Qwen3	752M	36.5	44.9	22.1	34.5

The model achieves the highest average score (45.4) while using fewer parameters than most competitors, demonstrating excellent parameter efficiency.

Key Features

• Compact Size: 434M parameters for efficient deployment
• Strong Performance: Competitive results across diverse benchmarks
• General Purpose: Capable across multiple domains including reasoning, knowledge, and problem-solving
• Resource Efficient: Optimized for inference speed and memory usage

Usage

Load the model using Hugging Face transformers:

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "aquiffoo/aquif-3-micro"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Generate text
inputs = tokenizer("Your prompt here", return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

Intended Use

aquif-3-micro is designed for:

• General text generation tasks
• Applications requiring efficient inference
• Educational and research purposes
• Prototyping and development environments
• Edge deployment scenarios

Limitations

As a compact model, aquif-3-micro may have limitations in:

• Very specialized domain knowledge
• Complex multi-step reasoning tasks
• Tasks requiring extensive world knowledge

License

This model is released under the Apache 2.0 license.