中文模型很弱智
#10
by
Jerry-SDUA
- opened
"有几个汉字/字母" 这种问题根据LLM的tokenization原理来说就很难回答，建议稍微了解一下BPE tokenier的知识。
此外，我们模型没有针对中国历史/文化知识特定训练过，Llama3本身pretraining的时候也很少有这方面的数据，因此这方面确实会弱一些，这个在这个discussion里有讨论过。
By the way, I'm not your servant. I've dedicated a significant amount of time and effort to developing these LLMs and have made them available for free. Please remain respectful in discussions when using these models.
shenzhi-wang
changed discussion status to
closed