Show HN: AutoThink – Boosts local LLM performance with adaptive reasoning

[codelion]

I built AutoThink, a technique that makes local LLMs reason more efficiently by adaptively allocating computational resources based on query complexity.

The core idea: instead of giving every query the same "thinking time," classify queries as HIGH or LOW complexity and allocate thinking tokens accordingly. Complex reasoning gets 70-90% of tokens, simple queries get 20-40%.

I also implemented steering vectors derived from Pivotal Token Search (originally from Microsoft's Phi-4 paper) that guide the model's reasoning patterns during generation. These vectors encourage behaviors like numerical accuracy, self-correction, and thorough exploration.

Results on DeepSeek-R1-Distill-Qwen-1.5B:

- GPQA-Diamond: 31.06% vs 21.72% baseline (+43% relative improvement)

- MMLU-Pro: 26.38% vs 25.58% baseline

- Uses fewer tokens than baseline approaches

Works with any local reasoning model - DeepSeek, Qwen, custom fine-tuned models. No API dependencies.

The technique builds on two things I developed: an adaptive classification framework that can learn new complexity categories without retraining, and an open source implementation of Pivotal Token Search.

Technical paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5253327

Code and examples: https://github.com/codelion/optillm/tree/main/optillm/autoth...

PTS implementation: https://github.com/codelion/pts

I'm curious about your thoughts on adaptive resource allocation for AI reasoning. Have you tried similar approaches with your local models?

[codelion]

The motivation for AutoThink came from watching how current reasoning models waste computation - they spend the same amount of "thinking time" on "what's 2+2?" as they do on complex mathematical proofs. This seemed obviously inefficient.

The breakthrough was combining two techniques I'd been working on separately: adaptive classification (which can learn new categories without retraining) and an open source implementation of Pivotal Token Search from Microsoft's Phi-4 paper. When I put them together with dynamic token budgeting, the performance gains were much better than expected.

What surprised me most was that the technique actually uses fewer tokens on average while improving performance. The adaptive allocation means simple queries finish faster, offsetting the extra computation on complex ones.

A few technical notes:

- The steering vectors are small (typically <1MB per pattern) and add minimal memory overhead

- Classification adds about 10ms latency, which is negligible

- Target layer selection matters - I found middle layers (15-20) work best for most models

I'd love feedback on:

- Have you tried similar adaptive approaches with your models?

- What other reasoning patterns would be useful to steer toward?

- Ideas for automatically detecting the optimal target layer?

Thanks for checking it out! Happy to answer any questions about the implementation or results.

[behnamoh]

> they spend the same amount of "thinking time" on "what's 2+2?" as they do on complex mathematical proofs.

Not anymore. Have you seen Gemini 2.5 Pro? Ask it simple questions and it almost doesn't "think". Ask it a coding question and it'll write a long reasoning article. I think the same goes for o3.

[Abishek_Muthian]

Congratulations! Any work to optimise efficiency w.r.t LLMs is much appreciated.

So far I’ve taken only lazy approach to optimising local LLMs by sending small queries to my M4 Mac Mini running MLX models and sending larger queries to my Nvidia 4090; it’s remarkable how efficient M4 is compared to Nvidia and I think Apple is in the right direction with MLX.

I would read about AutoThink and try to integrate it with my workflow.

[Lerc]

I have thought it might be worth seeding responses with the output of non-reasoning models, so after the user prompt, inject a block of "a non-reasoning model thought this:... stuff ....Was that what the user wanted?" For the instances where the non reasoning version was sufficient it might help the reasoning model get to the point earlier.

[waffletower]

Claude Sonnet 3.5 (not even the latest iterations: 3.7 or 4) clearly adapts processing time to query complexity -- processing time is dynamic.

[bufferoverflow]

But how do you classify a question as high vs low complexity? Some seemingly simple questions can turn out to be very very complex. For example, integer solution to

    x³ + y³ + z³ = 42 

took over a hundred years of compute time to find.

Or another seemingly simple equation with positive integers x,y,z

    x/(y+z)+y/(z+x)+z/(x+y) = 4

requires elliptic curve knowledge, and the solution is huge

    x = 154476802108746166441951315019919837485664325669565431700026634898253202035277999

    y = 36875131794129999827197811565225474825492979968971970996283137471637224634055579

    z = 4373612677928697257861252602371390152816537558161613618621437993378423467772036

(Solution is discussed here: https://www.quora.com/How-do-you-find-the-positive-integer-s...)

[codelion]

Query complexity in this context is based on how many tokens it took for the model to respond to a query correctly based on a ground truth dataset like GSM8k. The adaptive classifier learns over this dataset and then we use it at inference for classification.

[MrManatee]

I think there exists a separate skill for classifying problems by difficulty, apart from being able to solve them. This skill can be developed from both directions by learning which problems have been solved and which haven't been.

If someone asked me to find solutions to these example equations, there are three complications that I would immediately notice:

1. We are looking for solutions over integers. 2. There are three variables. 3. The degree of the equation is 3.

Having all three is a deadly combination. If we were looking for solutions over reals or complex numbers? Solvable. Less than three variables? Solvable. Degree less than 3? Solvable. With all three complications, it's still not necessarily hard, but now it might be. We might even be looking at an unsolved problem.

I haven't studied enough number theory to actually solve either of these problems, but I have studied enough to know where to look. And because I know where to look, it only takes me a few seconds to recognize the "this might be very difficult" vibe that both of these have. Maybe LLMs can learn to pick up on similar cues to classify problems as difficult or not so difficult without having needing to solve them. (Or, maybe they have already learned?)

[NiloCK]

I, too, built a POC autothink shortly after the Claude 3.7 release that included the `extended thinking` toggle. It's literally also called autothink:

https://github.com/NiloCK/autothink

https://www.paritybits.me/think-toggles-are-dumb/

My own version took a first pass with an LLM whose job was to assign a 0-100 complexity rating, and then there was more or less a linear scaling of the allocated thinking budget.

The OP effort here is obviously higher grade, and I'm really tickled to see quantitative results. Well done.

[nssnsjsjsjs]

This is an obvious optimisation. Surprised this isn't been done already. Good job writing it up and showing how it can be done.

[mentalgear]

It's great how small models help small teams and individual researchers everywhere now compete with big AI labs by allowing them to demonstrate new innovative approaches on small experiments.

Also, as small language models (SML) become more competent, it's amazing what they can do on-device !

[chrisweekly]

> "small language models (SML)"

that should be SLM, right?

[CMay]

In terms of reasoning models like QwQ or Qwen 3 I didn't waste too much time trying to improve their results aside from coming up with various ways to constrain their reasoning token output with prompts.

Even though Gemma 3 27B QAT is not a reasoning model, it's so good at instruction following and being used in LLM chains/routes that it can be used for classifying/language optimization steps before instructing it how to reason about the prompt in the next step. You can even have it output intermediate answers interspersed between multiple think tags in the same response. In many ways for these models I just define thinking as any tokens that are helping the model arrive at the conclusion, but are not fully formed parts of the answer.

Instructing it to use certain words (tokens) and types of phrasing preferentially is something that is known to improve results in general, not just in LLMs and I've seen improved results by encouraging certain types of language to be used. AutoThink using the highest performing tokens out of a dataset _could_ be a nice way to optimize towards that in a more general way.

It seems like there's a risk of using so many pivotal tokens that it almost overfits responses to benchmark questions, though. So, while I have personally seen careful word/token selection improve result quality and also see it as a potential low cost high return optimization, I'd still want to see how AutoThink generalizes.

[vintermann]

If host models for others, then sure, I'm happy to save some computation time for really simple queries. Sure the cost is that the model will be effectively dismissive of questions it judges to be "easy", but I'm not the one carrying that cost I suppose.

However, for a local model, answering my own queries? That's the last thing I want. I already spent way too much money on that GPU, might as well get use out of it.

[GENIXUS]

I’m very new to the world of LLMs and AI, but this project really caught my attention.

From what I understood, AutoThink helps the AI “think more wisely” by adjusting how much effort it spends based on how hard the question is. That makes a lot of intuitive sense — like how people don’t spend 10 minutes figuring out what 2+2 is, but do take time with tricky problems.

Even though I don’t know the technical parts (like token budgeting or steering vectors), it’s fascinating to see how these methods can make the AI both faster and smarter at the same time.

Thanks for sharing — I’m definitely going to follow this kind of work more closely from now on.

[shah_akshat]

Surprised this didn't exist. Great work @codelion

[SamScout]

Great food for thought! We will discuss this approach as we find our evolving AI-crawler should ideally be able to recognize when a site we visit needs more vs. less queries.

For context, we're samaritanscout.org a search engine that is attempting to provide a comprehensive view into all local volunteering opportunities posted on a range of nonprofit websites.

[casenmgreen]

It seems to me inadvisable to say "think" and "reason", because those words have particular meanings, and those particular meanings are not in use by LLMs.

They are a computing method, where we can choose to use more or less run time (and so processor time), to generate results.

[falcor84]

The ship has sailed, just like "computers" once referred to a human profession and now referred to machines.

[dymk]

When you “ping” and IP address, are you bouncing sound waves off of the metal hull of the other computer? No, but the word is used anyways, as it’s a useful metaphor for what’s really going on.

[dgb23]

My worldview is materialist and deterministic in principle. But day to day I'm an existentialist with a touch of spiritualism.

To me, a fairly pragmatic way of characterizing these tools day to day is to anthropomorphize them. One benefit of this heuristic: they simulate conversation and it's much easier to use them with a conversational flow. Another one is to create an approximation of a character, which makes it easier to build a useful intuition for what they can and cannot do.

Obviously these kinds of heuristics do break down. But it's obvious enough when they do so one can switch into a more precise and analytical mode of thinking.

[lostmsu]

Official https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-... says GPQA-Diamond is 33.8

[codelion]

Hey, yes the reported results do not restrict any time limit or token limit for the benchmarks. We run our baseline with the same config 0.6 temp and max_token 32k but we set a timeout after 600 secs. Otherwise it would take forever to benchmark with the resources we had. I have a note in the actual paper on that in the implementation details section.

[Dowwie]

Hey, this is really interesting. What are the features you used to measure the reasoning complexity? In other words, how does one evaluate a query during classification?

[codelion]

We use an adaptive classifier to learn how many tokens the model takes to respond correctly on a known dataset. I used the https://huggingface.co/adaptive-classifier/llm-router for experiments it is based on distilbert.

[shwouchk]

Very interesting, thanks for sharing!

FWIW gemini explicitly told me that it ranks question difficulty from 1 to 100 and depending on the bin allocates more or less resources to answering it

[NitpickLawyer]

> gemini explicitly told me

Do you mean someone from the gemini team? If you "asked" the LLM then it's likely a "hallucinated" answer. They say all sort of things about "themselves" only because they were trained to do so. They likely have 0 knowledge about their true architecture.

[throwaway314155]

Gemini has no access to its internal processes outside of whats in its system prompt - and even then LLM's are known to fabricate information about their inception.

[transfire]

That’s awesome!

Now have it mark blocks of text on or off, so it can ignore irrelevant, or worse erroneous material — no need to include it in the context window.

[codelion]

This sounds like an interesting idea, can you elaborate more may be with a concrete example. I am wondering if this can be implemented easily as a plugin in optillm.

[pkoird]

Back to TF IDF we go.

[shirman]

Hi, it does not work with llama.cpp right?

[codelion]

Optillm works with llama.cpp but this approach is implemented as a decoding strategy in PyTorch so at the moment you will need to use the local inference server in optillm to use it.

[danielhanchen]

Super cool and the results look pretty solid as well! Will give it a try!

[keeganpoppen]

i have definitely observed a similar pattern in the Big Label Foundation Models… so, i’m glad to see it in this realm too <3