Unverified50% confidenceBenchmarkExact time
At the 64k vocabulary setting, the new tokenizer averaged 1.147 tokens per word, beating the best existing tokenizer's 1.208 and XLM-RoBERTa's 1.512.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/17/2026
First Seen
Valid until: 2/15/2027
Sources
Related Entities
Related Claims
UnverifiedTinyLlama-1.1B's tokenizer produces an average of 1.34 subword tokens per word when processing Latin script, and an average of 22.60 tokens per word when processing Kannada.74% similarUnverifiedAt the 64k vocabulary setting, 88.6% of words were retained as a single token.74% similarVerifiedThe vocabulary of a standard autoregressive language model typically contains 32,000 to 128,000 tokens.69% similarUnverifiedRico Sennrich et al. quantified in their 2016 paper "Neural Machine Translation of Rare Words with Subword Units": on a 100-million-word German corpus, keeping whole words requires a vocabulary of 1.75 million entries, and the test set still leaves 1,079 unknown words; splitting into characters requires about 3,000 symbols, but the corpus balloons to 550 million positions; the subword method BPE uses about 63,000 symbols, 112 million positions, and zero unknown words.67% similarUnverifiedWordCraft is currently in alpha; the author estimates that by command count it covers 88% of Word's ribbon features, but it is only about 60% of the way to truly replacing Word, and for everyday basic needs that figure is 63%.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1178558API
curl https://kongchang.com/api/v1/knowledge/claims/1178558MCP
get_claim(id=1178558)