Unverified50% confidenceFactExact time
WikiText-103是包含约1亿token的英文维基百科语料，是语言模型基准测试的标准数据集之一
1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTraining data for large language models includes both authoritative sources like Wikipedia and unreliable content like forum spam and deliberate fabrications60% similarUnverifiedGitHub user jjlalli compiled 136 entries of Tunisian dialect NLP resources, covering five categories: text corpora, speech data, pre-trained models, evaluation benchmarks, and dictionaries58% similarUnverifiedThe vocabulary of a standard autoregressive language model typically contains 32,000 to 128,000 tokens.57% similarVerified大语言模型通过在数千亿乃至数万亿个词元（Token）的文本数据上进行自监督预训练56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/795597API
curl https://kongchang.com/api/v1/knowledge/claims/795597MCP
get_claim(id=795597)