--- pipeline_tag: text-generation inference: false license: apache-2.0 library_name: transformers tags: - language - aquif - gpt2 - text-generation-inference - small - efficient language: - en --- # 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: Benchmark Performance Comparison | 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. Parameter Efficiency Scatter Plot ## 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: ```python 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.