Comment on Contextual RetrievalComments−msp261yIf your knowledge base is smaller than 200,000 tokens (about 500 pages of material)I would prefer that anthropic just release their tokeniser so we don't have to make guesses.−whereismyacc1yshouldn't this be possible to reverse-engineer since they stream the responses token-by-token?
Comments
I would prefer that anthropic just release their tokeniser so we don't have to make guesses.
shouldn't this be possible to reverse-engineer since they stream the responses token-by-token?