---
license: apache-2.0
language:
- en
base_model: []
pipeline_tag: text-generation
datasets:
- HuggingFaceTB/cosmopedia
- tiiuae/falcon-refinedweb
library_name: gguf
tags:
- text-generation
- gguf
- quantized
- 1b
- llama-cpp
---

# PT1S-1B-Q8.gguf

This model is a 1-billion parameter text generation model trained on a high-quality mixture of synthetic and web-crawled data. It is optimized for efficiency and performance in a small footprint.

## Model Details

- **Model Type:** Text Generation
- **Parameters:** 1B
- **Quantization:** Q8_0 (8-bit quantization for high precision with reduced memory)
- **Training Data:** 
  - [HuggingFaceTB/cosmopedia](https://huggingface.co/datasets/HuggingFaceTB/cosmopedia)
  - [tiiuae/falcon-refinedweb](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)
- **Language(s):** English
- **License:** Apache 2.0

## Training Information

The model was trained on a curated blend of:
1. **Cosmopedia**: A large-scale synthetic dataset designed to provide high-quality educational content across various domains.
2. **Falcon RefinedWeb**: A massive, filtered web dataset that provides broad world knowledge and linguistic diversity.

This combination allows the model to have both structured knowledge from synthetic sources and a natural "web-aware" conversational style.

## Usage

### llama.cpp

You can use this model with [llama.cpp](https://github.com/ggerganov/llama.cpp) by running:

```bash
./main -m PT1S-1B-Q8.gguf -p "Once upon a time," -n 128
```

### Python (via llama-cpp-python)

```python
from llama_cpp import Llama

llm = Llama(model_path="./PT1S-1B-Q8.gguf")
output = llm("Q: What is the importance of cosmopedia dataset? A:", max_tokens=100)
print(output)
```

## Intended Use

This model is ideal for:
- Lightweight text generation tasks.
- Educational applications.
- On-device inference where memory is limited.
- Research into small language models (SLMs).

## Limitations and Bias

While trained on filtered data, small models may still exhibit biases or generate incorrect information (hallucinations). Users should always verify the output of the model for critical applications.