If I go to “Configure” and “Token Lenghts” I can select “Max output length:” up to 4096, this means the length of the final text corresponds approximately to 3000 words, since a token is equivalent to about 0.75 words.
In any case the length of the text that I can obtain at most is about 1000 words, even though I asked in the prompt, to create a document of about 3000 words. Do you have any suggestions on how to force an output of about 3000 words?
Under the configure tab the “Max Output Length” corresponds to the maximum allowable length. Putting that slider up or down changes the limit of what the model is allowed to output. However, it does not influence how long the model’s output will be. You must do that in the prompt.
For example, you can slide that limit up to 4000 tokens. But if your prompt is “Write a haiku”, it will only output a haiku (a few words).
So if you want very long outputs, you need to do that within the prompt. I would specify that in the prompt in multiple places. I would also suggest providing the prompt with an outline structure that helps explain how it should reach 3000 words.
i achieved the goal of a long text changed the model with claude sonnet 3.5. i hope it is useful to the community
1 Like
4096 tokens is not enough. Can we increase the limit? My context is translating a document. When using Pickaxe, it keeps stopping midway and ask if I want to continue. When I say proceed, it doesn’t remember where it stopped and starts to invent things that doesn’t exist in the original, even after I set a very high memory.
When I use the same document and do it on Claude, it’s able to translate the entire document in 1 go, clearly going beyond the 4096 token output limit.
Please advise what can I do in Pickaxe to achieve the same, or let it remember where it stopped so it knows where to continue.
1 Like
Claude can now do 8192 token output, but pickaxe can only do 4096 eventhough we’re using 3.5 Sonnet. Can we extend this?
1 Like
@nathaniel check this out for Claude models.
1 Like