Chinese-LLaMA-2-1.3B-GGUF
This repository contains the GGUF-v3 models (llama.cpp compatible) for Chinese-LLaMA-2-1.3B.
Performance
Metric: PPL, lower is better
| Quant | original | imatrix (-im) |
|---|---|---|
| Q2_K | 21.7422 +/- 0.47160 | 19.0472 +/- 0.42128 |
| Q3_K | 14.8538 +/- 0.32874 | 14.2966 +/- 0.31684 |
| Q4_0 | 13.8498 +/- 0.30349 | - |
| Q4_K | 16.0222 +/- 0.35555 | 13.0832 +/- 0.28717 |
| Q5_0 | 12.5743 +/- 0.27428 | - |
| Q5_K | 12.5865 +/- 0.27759 | 12.6305 +/- 0.27711 |
| Q6_K | 12.6588 +/- 0.27682 | 12.6944 +/- 0.27805 |
| Q8_0 | 12.7243 +/- 0.27883 | - |
| F16 | 12.6122 +/- 0.27616 | - |
The model with -im suffix is generated with important matrix, which has generally better performance (not always though).
Others
For Hugging Face version, please see: https://huggingface.co/hfl/chinese-llama-2-1.3b
Please refer to https://github.com/ymcui/Chinese-LLaMA-Alpaca-2/ for more details.