Ask HN: How are thinking efforts implemented?

Posted by simianwords 2 days ago

Claude and ChatGPT have thinking efforts where you can tune the amount of thinking allowed.

Like low, medium, high, xhigh and so on.

But are they different models underneath? Or same model with different parameter?

The reason I ask is because, if I change the effort param mid conversation in Claude code, I get a warning suggesting I’m breaking the cache.

I don’t think this happens in Codex because when I change the effort, the responses are still quick.

Comments

Comment by pyentropy 2 days ago

Take a look at the harmony repo which specifies the internal OpenAI format - the effort level is specified in the context after the <|start|> tag - https://github.com/openai/harmony

Note that inference libs also have parsers that put hard limits on reasoning tokens with separate counters (similar to how you can put a limit on token generation per completion versus waiting for an <eos>). For that, take a look at vllm reasoning docs.

Comment by pyentropy 2 days ago

Examples with inference of different reasoning effort levels is in the OpenAI docs as well - https://developers.openai.com/cookbook/articles/openai-harmo...

https://docs.vllm.ai/en/latest/features/reasoning_outputs/#a...

https://developers.openai.com/api/docs/guides/reasoning

Comment by simianwords 2 days ago

I think you have the right answer but I'm struggling to understand: does changing the effort change the prompt at the start of the conversation? I wonder why come up with this way at all? Why not just add a parameter at the end or something? At least it won't break cache.

Maybe like: add a secret suffix to your chat in the conversation to think more like

   conversation....

   Hey please help
   [think more]

Comment by pyentropy 2 days ago

I'm considering the possibility that it's good to break the prefix and cache because the LLM itself was rewarded (during post-training) with different prefixes/system prompts, each containing reasoning traces of the correct size.

I might be very very wrong though and LLMs disagree with me, insisting that cache is preserved and the system message doesn't have to change (even though it often contains effort level in context) if effort level changes across turns, and that all you have to do is tell the inference lib that parses think tags to early-close think tags that are too long.

Comment by simianwords 1 day ago

This seems correct but again I would like to think post training could have been also done by checking only the string in the last message sent.

Comment by aabdi 2 days ago

Different models do slight variants.

Usually it’s done in post training to enforce behavior based on prompt. Ie. System prompt with thinking:max or low or wtv.

Enforcement then goes via constrained decoding, checking for think token start and end with max lengths, or other variations

Comment by bjourne 2 days ago

LLMs work by generating the most likely continuation to a prompt. But they can also generate multiple likely continuations. This create multiple branches which in turn can generate even more branches. The LLM can then evaluate the branches, prune the unpromising ones, and merge the best ones. More branches means more tokens, means more effort.

Comment by simianwords 2 days ago

this has nothing to do with the thinking effort however

Comment by bjourne 2 days ago

Yes, it does. Breadth of search is exactly what the effort setting controls.

Comment by pyentropy 2 days ago

LLM-judge/parallel branching ≠ multi-token prediction ≠ reasoning effort.

See https://developers.openai.com/cookbook/articles/openai-harmo... and src/openai/types/shared/reasoning_effort.py

Comment by bjourne 1 day ago

Yes it is Dumbo. Your links are irrelevant.

Comment by simianwords 1 day ago

No it doesn't and lets not call people names. You can verify this using ChatGPT or anything else. You are mistaken and there are no "branches" happening.

Comment by bjourne 14 hours ago

And u can verify with actual facts that u are embarassingly wrong.

Comment by __patchbit__ 2 days ago

At a guess. May be associated with token length context window. Down selecting is consistent with warning message, forcing cutoff to context window. The technical term cache being a synonym. Increasing the headroom for more "thinking" should allow the implementation to access more resources without warning about the cache breaking.

Comment by sometimelurker 2 days ago

they use multitoken prediction behind the scenes, that might interact with the CoT in a strange way. maybe for different thinking modes they have different MTP models? if so thats interesting

Comment by pyentropy 2 days ago

The number of tokens you predict at time (multi or not) has nothing to do with whether the model wants to emit any, some or a lot of reasoning tokens in reasoning tag -- similar to how branch prediction will not really change the for loop iteration count.

Comment by sometimelurker 1 day ago

no it might. a high reasoning task is probably harder than a low reasoning task, so the same MTP LLM will predict more correct tokens on the low reasoning task. to compensate for this, big labs likely have different MTP LLMs for different cases. it would make sense for them to do this