---
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:
| 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:
```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.