A Token is the most fundamental and common concept in large language models.
It's well known that the training data volume, context limit, and generation speed of large language models are all measured in tokens.
For example:
Tongyi Qianwen-7B is pre-trained with over 2.4 trillion tokens of data.
The numbers 8k or 32k after a model's name indicate the maximum text length it can generate or predict.
TPS (tokens per second) measures the generation speed of large models, indicating how many tokens are output per second.
