> What a token is depends on the context of the model you're using, but generally a token is a portion of a word. When doing quick estimates, I just assume eve

> What a token is depends on the context of the model you're using, but generally a token is a portion of a word. When doing quick estimates, I just assume every syllable is a token. It tends to overestimate, which is fine for my OOM mitigation purposes.