Llama-3-Chinese-8B-Instruct-GGUF
This repository contains Llama-3-Chinese-8B-Instruct-GGUF (llama.cpp/ollama/tgw, etc. compatible), which is the quantized version of Llama-3-Chinese-8B-Instruct.
Note: this is an instruction (chat) model, which can be used for conversation, QA, etc.
Further details (performance, usage, etc.) should refer to GitHub project page: https://github.com/ymcui/Chinese-LLaMA-Alpaca-3
Performance
Metric: PPL, lower is better
Note: Old models have been removed due to its inferior performance (llama.cpp has breaking changes on pre-tokenizer).
| Quant | Size | PPL (old model) | 👍🏻 PPL (new model) |
|---|---|---|---|
| Q2_K | 2.96 GB | 10.3918 +/- 0.13288 | 9.1168 +/- 0.10711 |
| Q3_K | 3.74 GB | 6.3018 +/- 0.07849 | 5.4082 +/- 0.05955 |
| Q4_0 | 4.34 GB | 6.0628 +/- 0.07501 | 5.2048 +/- 0.05725 |
| Q4_K | 4.58 GB | 5.9066 +/- 0.07419 | 5.0189 +/- 0.05520 |
| Q5_0 | 5.21 GB | 5.8562 +/- 0.07355 | 4.9803 +/- 0.05493 |
| Q5_K | 5.34 GB | 5.8062 +/- 0.07331 | 4.9195 +/- 0.05436 |
| Q6_K | 6.14 GB | 5.7757 +/- 0.07298 | 4.8966 +/- 0.05413 |
| Q8_0 | 7.95 GB | 5.7626 +/- 0.07272 | 4.8822 +/- 0.05396 |
| F16 | 14.97 GB | 5.7628 +/- 0.07275 | 4.8802 +/- 0.05392 |
Others
For full model, please see: https://huggingface.co/hfl/llama-3-chinese-8b-instruct
For LoRA-only model, please see: https://huggingface.co/hfl/llama-3-chinese-8b-instruct-lora
If you have questions/issues regarding this model, please submit an issue through https://github.com/ymcui/Chinese-LLaMA-Alpaca-3