Berkeley Researchers Replicate DeepSeek R1's Core Tech for Just $30: A Small Mod

https://xyzlabs.substack.com/p/berkeley-researchers-replicate-deepseek

by semking • 1 year ago

163 66 1 year ago

xyzlabs.substack.com

aurareturn3• 1 year ago

This is truly the biggest breakthrough from DeepSeek - that an LLM can teach itself to reason, no human feedback needed.

That’s nuts and brings forward the idea that an AI is close to self improvement.

EGreg• 1 year ago

We already knew that kind of stuff from AlphaZero vs AlphaGo

AlphaGo Zero is a version of DeepMind's Go software AlphaGo. AlphaGo's team published an article in Nature in October 2017 introducing AlphaGo Zero, a version created without using data from human games, and stronger than any previous version.[1] By playing games against itself, AlphaGo Zero: surpassed the strength of AlphaGo Lee in three days by winning 100 games to 0; reached the level of AlphaGo Master in 21 days; and exceeded all previous versions in 40 days.[2]

Training artificial intelligence (AI) without datasets derived from human experts has significant implications for the development of AI with superhuman skills, as expert data is "often expensive, unreliable, or simply unavailable."[3] Demis Hassabis, the co-founder and CEO of DeepMind, said that AlphaGo Zero was so powerful because it was "no longer constrained by the limits of human knowledge".[4] Furthermore, AlphaGo Zero performed better than standard deep reinforcement learning models (such as Deep Q-Network implementations[5]) due to its integration of Monte Carlo tree search. David Silver, one of the first authors of DeepMind's papers published in Nature on AlphaGo, said that it is possible to have generalized AI algorithms by removing the need to learn from humans.[6]

Google later developed AlphaZero, a generalized version of AlphaGo Zero that could play chess and Shōgi in addition to Go.[7] In December 2017, AlphaZero beat the 3-day version of AlphaGo Zero by winning 60 games to 40, and with 8 hours of training it outperformed AlphaGo Lee on an Elo scale. AlphaZero also defeated a top chess program (Stockfish) and a top Shōgi program (Elmo).[8][9]

Source: https://en.wikipedia.org/wiki/AlphaGo_Zero

aurareturn3• 1 year ago

Indeed. But DeepSeek is the first to release a working repro for LLMs as far as I know.

simlevesque2• 1 year ago

Yeah but LLM and game AIs are very different. An AI can easily tell if it won the game but an LLM can't know by itself if what they said was useful.

thorum• 1 year ago

It’s great, but it only works for problems where there is exactly one correct solution and it’s possible to automatically verify the solution - like math and programming. So far these reasoning models have not shown much transfer learning of reasoning to other domains, and are often worse at non-math/code tasks than standard models.

datameta• 1 year ago

Not sure if this a trivial or naive thought - is that perhaps because in non-discrete ideas there is more granularity of information encoded compared to a numerical solution?

Separatelt, but on a related note - do we need analog or quantum computing to "truly" scale?

onlyrealcuzzo2• 1 year ago

There are infinite solutions to coding problems.

You could have lots of comments and pass statements and unnecessary conditionals.

I don't think it matters that there is only one correct answer.

It matters that you can verify reliably if the answer is correct enough.

JKCalhoun• 1 year ago

So, we have a very good left-brain model.

vintermann• 1 year ago

We can probably make it work at more nebulous problems by using a big LLM to judge the quality of the answers as well. It should be easier to recognise e.g. a really good poetic translation than to make one, and as long as that's true it could benefit from internal monologue "reasoning" in those domains as well.

aurareturn3• 1 year ago

LLMs can verify math problems if you give it a calculator tool. Coding problems if you give it a Linux environment.

So why can’t we give LLMs other virtual environments to verify if the solution is correct? For example, a stock simulator, a physics simulator, a driving simulator, etc.

onlypassingthru• 1 year ago

Once the computer learns to apply the logic of Tic-Tac-Toe to Global Thermonuclear War, we will finally have the W.O.P.R.[0]

[0]https://en.wikipedia.org/wiki/WarGames

segasaturn• 1 year ago

This makes ScaleAI obsolete right?

ricksunny• 1 year ago

A physics simulator's rules will be derivative of known physics. If we ask it to push the boundaries of known physics, then we can't verify it without real-world experiment. Absent that then it's basically Chegg on steroids: "Tell me, based on what we believe about how the universe works, about X." is the implicit preface to every physics question.

To the extent that someone can figure out a way out of this epistemic box, I'm interested.

cluckindan3• 1 year ago

If the current hubbub around DeepSeen is really because they ”created their model” with like $5M when previously ”creating a model” cost $500B, it is rather obvious that ”creating the model” with just $30 implies the meanings of the three ”creating a model” expressions are highly divergent.

simlevesque2• 1 year ago

Not really, the two first examples you named are the same meaning.

drawnwren2• 1 year ago

They aren’t though. $5M is the cost of a single training run. $500B includes the cost of operations, data center, a lot more failed runs because they weren’t sure that they’d were on the right path etc.

bilbo0s• 1 year ago

The first two are not the same thing.

DeepSeek never claimed to have trained the base models they used. Now maybe a lot of people inferred that they did, but that's not what they claimed.

Their breakthrough was more along the lines of, "given a model, we can train it to reason using a certain class of RL techniques". Which is, to my mind, more useful in any case.

But yeah, if there are people out there thinking they can train base models from scratch for USD6 Million with no data, they're likely to be disagreeably surprised when they make the attempt.

TheTaytay• 1 year ago

So far it is literally 3 different uses of the phrase “creating a model”. I could further obfuscate it by fine tuning the work of these researchers on a single step and claimed I “created a model” for $0.05. :( I feel like that’s what everyone who is naming their model “R1” is doing right now…

This article is trying to make the headline sound as if they replicated the model, when they “just” replicated the theory of one of their findings. It’s a great result even without the author’s obfuscation.

PaulHoule• 1 year ago

These models are highly specialized models that reason similarly to DeepSeek's new model except instead of reasoning over the general domain they do something specific like arithmetic.

They're using a similar model and a similar training method to do something simple. It is to DeepSeek what DeepSeek is to ChatGPT-4o, and this kind of reduction to the simplest case is exactly what makes progress go.

(e.g. ChatGPT-4 is a dead end because it is so expensive to train that you can't do training experiments. At $30 a model it is reasonable to train a model 1000s of times)

nyrikki• 1 year ago

That may be the current hubub,

But how they built on other open concepts like Cold Start + RL + Rejection Sampling etc... is the big thing.

I am not at all connected to whomever this is, but as I am on a "confidential project for a confidential client" it seems to simplify what I would overcomplicate anyway.

https://youtu.be/Pabqg33sUrg

Some of us need to improve domain specificity, and are more interested in targeted capabilities.

DeepSeek-R1-Distill-Qwen-32B itself is pretty good to be honest, but it is more aspirational how easy it is to add reasoning to any base model IMHO.

Those of us who remember Simon’s satisficing principle, realize that often complex problems sub-optimal solutions.

We don't need, want or expect some central omniscient and omnipotent AGI, we want tools that are sufficient to solve real world problems.

Obviously having good reliable domain specific CoT for cold start etc... is hard.

But not nearly as hard as hitching the companies success to MS/OpenAI.

While there is a lot of noise...I think some people are missing the forrest for the trees here. Deepseek is more about moving the decision to use OpenAI for some needs from a single source vendor to one of convenience for us.

For those of us who are in the camp that OpenAI was never going to reach their "AGI" claims because of fundamental issues like the frame problem, this is huge.

Ya...we won't be training our own models for some silly budget of ~$5M like the press is claiming.

But we won't have to for some use cases...

I think it will take time and someone on a project with not such a strict NDA to demonstrate just how much this 'completes' things in ways that RAG, functions etc... didn't.

Or I just possibly got lucky with my needs? who knows.

highfrequency• 1 year ago

First graph tells the story - below a certain model size (500m params), reinforcement learning is close to useless. Above this (task-dependent) model size threshold, reinforcement learning basically works.

I suspect this is what we saw play out with math/coding reasoning models - until recently, the base models were not good enough for ~random output search to hit on a correct path with any reasonable frequency. Below this threshold of base model intelligence, the only efficient way forward was to collect plain supervised data (either through human labeled math problem solutions [1] or meticulous filtering of web text [2].

But as soon the base model (in this case Deepseek V3) breaks through and can actually solve a decent fraction of math problems, then reinforcement learning (plus other simple tricks like chain-of-thought prompting, simple ensemble voting, etc.) can easily juice the results through the following loop:

1) random search through different solution paths

2) identify the correct solution paths based on the final answer

3) train on the correct solution paths

The exciting thing is that not only can RL bump up the performance of the current base model, but it can be used to generate new high-quality reasoning trace data, which was in painfully short-supply for training the initial models. This leads to a new wave of base models with better one-pass intuition, which leads to more efficient reinforcement learning search on harder problems, which leads to better training data...

Note that this was basically impossible for non-LLM models in the past. You could always juice ImageNet classification performance with a simple ensemble of identically trained models, but that path didn't lead anywhere interesting because a juiced model didn't allow the creation of new synthetic data that was superior to the data it was trained on. The key difference is that LLMs not only output the solution but also output a solution path with all the intermediate steps - and these searched-and-filtered solution paths are much more valuable than the vast majority of the model's initial training data.

[1] https://arxiv.org/abs/2305.20050

[2] https://arxiv.org/abs/2402.03300 and https://arxiv.org/abs/2206.14858

krackers• 1 year ago

Yeah that is the hypothesis some people have floated for as to why we only "just now" discovered this simple idea:

https://news.ycombinator.com/item?id=42837349

DannyPage• 1 year ago

Unless I missed it, it seems strange that the article wouldn’t link to the Github repo for the TinyZero model.

https://github.com/Jiayi-Pan/TinyZero

semking8• 1 year ago

You are right! I commented here and on their substack to give the credit to the OPs!

fp64• 1 year ago

$30 is “less than a dinner for two”?

jedberg• 1 year ago

For a grad student. It's about 2 burritos in Berkeley. :)

nick34433• 1 year ago

Would it be correct to summarize that the general conceptual shift is optimizing MOEs on more specific smaller tasks? It smells like borderline overfitting to me for some reason.

whimsicalism3• 1 year ago

the core insight is that you can train on extremely sparse deterministic reward signal and it just works

that MoEs are better for a given compute budget has been known for a while.

nick34433• 1 year ago

Thank you for the reply. Maybe I'm daft or out of the loop, I don't see the difference between this method and what I have seen previously as "fine tuning on synthetic data".

They also mention in the source tweet that the quality of the base model is important, which is not really accounted for in the $30 figure.

whimsicalism3• 1 year ago

'replication' requires matching benchmark performance, definitionally.

more like 'demonstrates the technique generalizes' here. HN has really been inundated with blogspam recently

ipsum23• 1 year ago

Reposting the comment from https://news.ycombinator.com/item?id=42843959:

This is blogspam of https://github.com/Jiayi-Pan/TinyZero and https://nitter.lucabased.xyz/jiayi_pirate/status/18828393705.... This also doesn't mention that it's for one specific domain (playing Countdown). See also https://news.ycombinator.com/item?id=42819262.

semking8• 1 year ago

I sincerely do not understand this comment so please clarify: are you saying I'm spamming? Is there an issue with the source?

ipsum23• 1 year ago

The title of the post is misleading and the contents are blogspam and do not link to the original source.

oytis• 1 year ago

Is it some kind of a joke?

xigency• 1 year ago

This reads like an AI hallucination. I'm willing to steak-out this ground even if I'm wrong because of the glaring lack of skepticism. I don't even know when dollars became a concrete compute measure. We used to use FLOPs before we were trying to pull headlines like it were a claw machine game.

semking8• 1 year ago

The title is click-baity if you ask me. I didn't change it.

snake_doc• 1 year ago

@dang please link to either the GitHub https://github.com/Jiayi-Pan/TinyZero

or the primary source twitter thread: https://x.com/jiayi_pirate/status/1882839370505621655

semking8• 1 year ago

Thanks, I just commented on their sub to give credit!

nightpool2• 1 year ago

Commented where? I don't see any source links in the substack article.

semking8• 1 year ago

Guys I'm sorry but it appears the substack did NOT link to the original authors which is NOT acceptable!

Credit

GitHub: https://github.com/Jiayi-Pan/TinyZero

Source on X: https://x.com/jiayi_pirate/status/1882839370505621655

semking8• 1 year ago

I just left a comment on their Substack to give the credit to the OPs.

excalibur• 1 year ago

Good to know that our AI overlords will be built as cheaply as possible. If there's one thing I can't stand about bondage it's inefficiency.

SubiculumCode• 1 year ago

Sell Nvidia last week? Seriously. Or is it that now we can make smaller models more powerful, and then run more of them to get more work done.

UncleOxidant• 1 year ago

"TinyZero is a reproduction of DeepSeek R1 Zero in countdown and multiplication tasks." Does that mean that this has very limited utility (to certain math problems)?

Berkeley Researchers Replicate DeepSeek R1's Core Tech for Just $30: A Small Mod

Related Stories