Chinese-LLaMA-2-13B-16K-GGUF
This repository contains the GGUF-v3 models (llama.cpp compatible) for Chinese-LLaMA-2-13B-16K.
Performance
Metric: PPL, lower is better
| Quant | original | imatrix (-im) |
|---|---|---|
| Q2_K | 11.8958 +/- 0.20739 | 13.0017 +/- 0.23003 |
| Q3_K | 9.7130 +/- 0.17037 | 9.3443 +/- 0.16582 |
| Q4_0 | 9.2002 +/- 0.16219 | - |
| Q4_K | 9.0055 +/- 0.15918 | 8.9848 +/- 0.15908 |
| Q5_0 | 8.8441 +/- 0.15690 | - |
| Q5_K | 8.8999 +/- 0.15751 | 8.8983 +/- 0.15753 |
| Q6_K | 8.8944 +/- 0.15776 | 8.8833 +/- 0.15760 |
| Q8_0 | 8.8745 +/- 0.15745 | - |
| F16 | 8.8687 +/- 0.15729 | - |
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-2-13b-16k
Please refer to https://github.com/ymcui/Chinese-LLaMA-Alpaca-2/ for more details.