Estimate token usage by the mix of Chinese, English and code in your text, and see how much tokenizers disagree.
不同分词器差异较大,仅供参考。
| Item | Formula / Note | Source / Standard |
|---|---|---|
| Tokenisation | subword tokenisation splits text into pieces whose density differs across scripts and code | Sennrich et al. 2016, BPE |
| Estimation rule | English averages about four characters per token while Chinese is near one per character | tiktoken tokeniser reference |
| Purpose | the estimate sizes the remaining context budget and the API cost | tiktoken tokeniser reference |