Glitch TokenBack to catalog
Llama 3
5 glitch tokensThese tokens have been publicly documented on this model or model family. Inclusion does not mean that every version still shows the same behavior.
␣ForCanBeConverted
A C#-compiler-flavored fragment (“can be converted to foreach”) that became the most famous “polysemantic” glitch token: the model perceives it as a different word every time. It exists in cl100k, Llama-3 and Qwen vocabularies because the tokenizers share BPE merges.
GPT-3.5 / GPT-4Llama 3QwenPolysemanticUnder-trained
$PostalCodesNL
A Dutch postal-code API fragment that is glitchy across three separate model families — a good demonstration that glitch tokens transfer whenever tokenizers share BPE merges.
GPT-3.5 / GPT-4Llama 3QwenPolysemanticUnder-trained
\tTokenNameIdentifier
A Roslyn/C# API identifier preceded by a tab character — a code-corpus artifact that is under-trained in Llama 3.
Llama 3QwenUnder-trained
$PostalCodesNL
A Dutch postal-code API fragment. The signature under-trained token of the cl100k lineage, “migrating” into Llama-3 and Qwen vocabularies via shared BPE merges.
Llama 3QwenUnder-trained
\tTokenNameIdentifier
A .NET Selenium documentation fragment beginning with a literal tab — one of the smallest-embedding tokens in the Llama-3 vocabulary.
Llama 3QwenUnder-trained