Optimizing Test-Time Compute via Meta Reinforcement Fine-Tuning
Paper
โข
2503.07572
โข
Published
โข
25
Data Agents
Enterprise
community
AI & ML interests
None defined yet.
Recent Activity
View all activity
data-agents's activity
edbeechingย
authored
a
paper
about 20 hours ago
lewtunย
authored
a
paper
about 20 hours ago
Post
1073
We've kept pushing our Open-R1 project, an open initiative to replicate and extend the techniques behind DeepSeek-R1.
And even we were mind-blown by the results we got with this latest model we're releasing: โก๏ธOlympicCoder ( open-r1/OlympicCoder-7B and open-r1/OlympicCoder-32B)
It's beating Claude 3.7 on (competitive) programming โa domain Anthropic has been historically really strong atโ and it's getting close to o1-mini/R1 on olympiad level coding with just 7B parameters!
And the best part is that we're open-sourcing all about its training dataset, the new IOI benchmark, and more in our Open-R1 progress report #3: https://huggingface.co/blog/open-r1/update-3
Datasets are are releasing:
- open-r1/codeforces
- open-r1/codeforces-cots
- open-r1/ioi
- open-r1/ioi-test-cases
- open-r1/ioi-sample-solutions
- open-r1/ioi-cots
- open-r1/ioi-2024-model-solutions
And even we were mind-blown by the results we got with this latest model we're releasing: โก๏ธOlympicCoder ( open-r1/OlympicCoder-7B and open-r1/OlympicCoder-32B)
It's beating Claude 3.7 on (competitive) programming โa domain Anthropic has been historically really strong atโ and it's getting close to o1-mini/R1 on olympiad level coding with just 7B parameters!
And the best part is that we're open-sourcing all about its training dataset, the new IOI benchmark, and more in our Open-R1 progress report #3: https://huggingface.co/blog/open-r1/update-3
Datasets are are releasing:
- open-r1/codeforces
- open-r1/codeforces-cots
- open-r1/ioi
- open-r1/ioi-test-cases
- open-r1/ioi-sample-solutions
- open-r1/ioi-cots
- open-r1/ioi-2024-model-solutions
freddyaboultonย
posted
an
update
1 day ago
Post
1413
Privacy matters when talking to AI! ๐
We've just added a microphone mute button to FastRTC in our latest update (v0.0.14). Now you control exactly what your LLM hears.
Plus lots more features in this release! Check them out:
https://github.com/freddyaboulton/fastrtc/releases/tag/0.0.14
We've just added a microphone mute button to FastRTC in our latest update (v0.0.14). Now you control exactly what your LLM hears.
Plus lots more features in this release! Check them out:
https://github.com/freddyaboulton/fastrtc/releases/tag/0.0.14
Post
1598
Introducing OlympicCoder: a series of open reasoning models that can solve olympiad-level programming problems ๐งโ๐ป
- 7B open-r1/OlympicCoder-7B
- 32B open-r1/OlympicCoder-32B
We find that OlympicCoder models outperform Claude 3.7 Sonnet, as well as others over 100x larger ๐ช
Together with the models, we are releasing:
๐CodeForces-CoTs: new dataset of code problems from the most popular competitive coding platform, with R1 traces in C++ and Python open-r1/codeforces-cots
๐ IOI'2024: a new benchmark of VERY hard programming problems where even frontier models struggle to match human performance open-r1/ioi
For links to the models and datasets, check out our latest progress report from Open R1: https://huggingface.co/blog/open-r1/update-3
- 7B open-r1/OlympicCoder-7B
- 32B open-r1/OlympicCoder-32B
We find that OlympicCoder models outperform Claude 3.7 Sonnet, as well as others over 100x larger ๐ช
Together with the models, we are releasing:
๐CodeForces-CoTs: new dataset of code problems from the most popular competitive coding platform, with R1 traces in C++ and Python open-r1/codeforces-cots
๐ IOI'2024: a new benchmark of VERY hard programming problems where even frontier models struggle to match human performance open-r1/ioi
For links to the models and datasets, check out our latest progress report from Open R1: https://huggingface.co/blog/open-r1/update-3
- 1 reply
Post
760
Our new Agentic leaderboard is now live!๐ฅ
If you ever asked which LLM is best for powering agents, we've just made a leaderboard that ranks them all! Built with @albertvillanova , this ranks LLMs powering a smolagents CodeAgent on subsets of various benchmarks. โ
๐ GPT-4.5 comes on top, even beating reasoning models like DeepSeek-R1 or o1. And Claude-3.7-Sonnet is a close second!
The leaderboard also allows you to show the scores of vanilla LLMs (without any agentic setup) on the same benchmarks: this shows the huge improvements brought by agentic setups. ๐ช
(Note that results will be added manually, so the leaderboard might not always have the latest LLMs)
If you ever asked which LLM is best for powering agents, we've just made a leaderboard that ranks them all! Built with @albertvillanova , this ranks LLMs powering a smolagents CodeAgent on subsets of various benchmarks. โ
๐ GPT-4.5 comes on top, even beating reasoning models like DeepSeek-R1 or o1. And Claude-3.7-Sonnet is a close second!
The leaderboard also allows you to show the scores of vanilla LLMs (without any agentic setup) on the same benchmarks: this shows the huge improvements brought by agentic setups. ๐ช
(Note that results will be added manually, so the leaderboard might not always have the latest LLMs)
- 1 reply
freddyaboultonย
posted
an
update
16 days ago
Post
3150
Getting WebRTC and Websockets right in python is very tricky. If you've tried to wrap an LLM in a real-time audio layer then you know what I'm talking about.
That's where FastRTC comes in! It makes WebRTC and Websocket streams super easy with minimal code and overhead.
Check out our org: hf.co/fastrtc
That's where FastRTC comes in! It makes WebRTC and Websocket streams super easy with minimal code and overhead.
Check out our org: hf.co/fastrtc
Post
4727
We now have a Deep Research for academia: SurveyX automatically writes academic surveys nearly indistinguishable from human-written ones ๐ฅ
Researchers from Beijing and Shanghai just published the first application of a deep research system to academia: their algorithm, given a question, can give you a survey of all papers on the subject.
To make a research survey, you generally follow two steps, preparation (collect and organize papers) and writing (outline creation, writing, polishing). Researchers followed the same two steps and automated them.
๐ฏ For the preparation part, a key part is find all the important references on the given subject.
Researchers first cast a wide net of all relevant papers. But then finding the really important ones is like distilling knowledge from a haystack of information. To solve this challenge, they built an โAttributeTreeโ object that structures key information from citations. Ablating these AttributeTrees significantly decreased structure and synthesis scores, so they were really useful!
๐ For the writing part, key was to get a synthesis that's both short and true. This is not easy to get with LLMs! So they used methods like LLM-based deduplication to shorten the too verbose listings made by LLMs, and RAG to grab original quotes instead of made-up ones.
As a result, their system outperforms previous approaches by far!
As assessed by LLM-judges, the quality score os SurveyX even approaches this of human experts, with 4.59/5 vs 4.75/5 ๐
I advise you to read the paper, it's a great overview of the kind of assistants that we'll get in the short future! ๐ SurveyX: Academic Survey Automation via Large Language Models (2502.14776)
Their website shows examples of generated surveys ๐ http://www.surveyx.cn/
Researchers from Beijing and Shanghai just published the first application of a deep research system to academia: their algorithm, given a question, can give you a survey of all papers on the subject.
To make a research survey, you generally follow two steps, preparation (collect and organize papers) and writing (outline creation, writing, polishing). Researchers followed the same two steps and automated them.
๐ฏ For the preparation part, a key part is find all the important references on the given subject.
Researchers first cast a wide net of all relevant papers. But then finding the really important ones is like distilling knowledge from a haystack of information. To solve this challenge, they built an โAttributeTreeโ object that structures key information from citations. Ablating these AttributeTrees significantly decreased structure and synthesis scores, so they were really useful!
๐ For the writing part, key was to get a synthesis that's both short and true. This is not easy to get with LLMs! So they used methods like LLM-based deduplication to shorten the too verbose listings made by LLMs, and RAG to grab original quotes instead of made-up ones.
As a result, their system outperforms previous approaches by far!
As assessed by LLM-judges, the quality score os SurveyX even approaches this of human experts, with 4.59/5 vs 4.75/5 ๐
I advise you to read the paper, it's a great overview of the kind of assistants that we'll get in the short future! ๐ SurveyX: Academic Survey Automation via Large Language Models (2502.14776)
Their website shows examples of generated surveys ๐ http://www.surveyx.cn/
Post
3026
Less is More for Reasoning (LIMO): a 32B model fine-tuned with 817 examples can beat o1-preview on math reasoning! ๐คฏ
Do we really need o1's huge RL procedure to see reasoning emerge? It seems not.
Researchers from Shanghai Jiaotong University just demonstrated that carefully selected examples can boost math performance in large language models using SFT โno huge datasets or RL procedures needed.
Their procedure allows Qwen2.5-32B-Instruct to jump from 6.5% to 57% on AIME and from 59% to 95% on MATH, while using only 1% of the data in previous approaches.
โก The Less-is-More Reasoning Hypothesis:
โฃ Minimal but precise examples that showcase optimal reasoning patterns matter more than sheer quantity
โฃ Pre-training knowledge plus sufficient computational resources at inference levels up math skills
โก๏ธ Core techniques:
โฃ High-quality reasoning chains with self-verification steps
โฃ 817 handpicked problems that encourage deeper reasoning
โฃ Enough inference-time computation to allow extended reasoning
๐ช Efficiency gains:
โฃ Only 817 examples instead of 100k+
โฃ 40.5% absolute improvement across 10 diverse benchmarks, outperforming models trained on 100x more data
This really challenges the notion that SFT leads to memorization rather than generalization! And opens up reasoning to GPU-poor researchers ๐
Read the full paper here ๐ย LIMO: Less is More for Reasoning (2502.03387)
Do we really need o1's huge RL procedure to see reasoning emerge? It seems not.
Researchers from Shanghai Jiaotong University just demonstrated that carefully selected examples can boost math performance in large language models using SFT โno huge datasets or RL procedures needed.
Their procedure allows Qwen2.5-32B-Instruct to jump from 6.5% to 57% on AIME and from 59% to 95% on MATH, while using only 1% of the data in previous approaches.
โก The Less-is-More Reasoning Hypothesis:
โฃ Minimal but precise examples that showcase optimal reasoning patterns matter more than sheer quantity
โฃ Pre-training knowledge plus sufficient computational resources at inference levels up math skills
โก๏ธ Core techniques:
โฃ High-quality reasoning chains with self-verification steps
โฃ 817 handpicked problems that encourage deeper reasoning
โฃ Enough inference-time computation to allow extended reasoning
๐ช Efficiency gains:
โฃ Only 817 examples instead of 100k+
โฃ 40.5% absolute improvement across 10 diverse benchmarks, outperforming models trained on 100x more data
This really challenges the notion that SFT leads to memorization rather than generalization! And opens up reasoning to GPU-poor researchers ๐
Read the full paper here ๐ย LIMO: Less is More for Reasoning (2502.03387)
Post
2903
๐๐ฟ๐ฒ๐ฎ๐ ๐ณ๐ฒ๐ฎ๐๐๐ฟ๐ฒ ๐ฎ๐น๐ฒ๐ฟ๐: you can now share agents to the Hub! ๐ฅณ๐ฅณ
And any agent pushed to Hub get a cool Space interface to directly chat with it.
This was a real technical challenge: for instance, serializing tools to export them meant that you needed to get all the source code for a tool, verify that it was standalone (not relying on external variables), and gathering all the packages required to make it run.
Go try it out! ๐ https://github.com/huggingface/smolagents
And any agent pushed to Hub get a cool Space interface to directly chat with it.
This was a real technical challenge: for instance, serializing tools to export them meant that you needed to get all the source code for a tool, verify that it was standalone (not relying on external variables), and gathering all the packages required to make it run.
Go try it out! ๐ https://github.com/huggingface/smolagents
- 2 replies
Post
2514
For those who haven't come across it yet, here's a handy trick to discuss an entire GitHub repo with an LLM:
=> Just replace "github" with "gitingest" in the url, and you get the whole repo as a single string that you can then paste in your LLMs
=> Just replace "github" with "gitingest" in the url, and you get the whole repo as a single string that you can then paste in your LLMs
Post
4821
"๐ฎ๐ฌ๐ฎ๐ฑ ๐๐ถ๐น๐น ๐ฏ๐ฒ ๐๐ต๐ฒ ๐๐ฒ๐ฎ๐ฟ ๐ผ๐ณ ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐": this statement has often been made, here are numbers to support it.
I've plotted the progress of AI agents on GAIA test set, and it seems they're headed to catch up with the human baseline in early 2026.
And that progress is still driven mostly by the improvement of base LLMs: progress would be even faster with fine-tuned agentic models.
I've plotted the progress of AI agents on GAIA test set, and it seems they're headed to catch up with the human baseline in early 2026.
And that progress is still driven mostly by the improvement of base LLMs: progress would be even faster with fine-tuned agentic models.
Post
4864
Introducing OpenR1-Math-220k!
open-r1/OpenR1-Math-220k
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch ๐ช
Whatโs new compared to existing reasoning datasets?
โพ Based on AI-MO/NuminaMath-1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.
๐ณ 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
๐ 512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.
โณ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that canโt be verified with a rules-based parser)
๐ We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
๐ Read our blog post for all the nitty gritty details: https://huggingface.co/blog/open-r1/update-2
open-r1/OpenR1-Math-220k
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch ๐ช
Whatโs new compared to existing reasoning datasets?
โพ Based on AI-MO/NuminaMath-1.5: we focus on math reasoning traces and generate answers for problems in NuminaMath 1.5, an improved version of the popular NuminaMath-CoT dataset.
๐ณ 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
๐ 512 H100s running locally: Instead of relying on an API, we leverage vLLM and SGLang to run generations locally on our science cluster, generating 180k reasoning traces per day.
โณ Automated filtering: We apply Math Verify to only retain problems with at least one correct answer. We also leverage Llama3.3-70B-Instruct as a judge to retrieve more correct examples (e.g for cases with malformed answers that canโt be verified with a rules-based parser)
๐ We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
๐ Read our blog post for all the nitty gritty details: https://huggingface.co/blog/open-r1/update-2
Post
3735
๐๐ฑ๐๐ฒ๐ป'๐ ๐ป๐ฒ๐ ๐๐ฎ๐๐ฎ ๐๐ด๐ฒ๐ป๐๐ ๐๐ฒ๐ป๐ฐ๐ต๐บ๐ฎ๐ฟ๐ธ ๐๐ต๐ผ๐๐ ๐๐ต๐ฎ๐ ๐๐ฒ๐ฒ๐ฝ๐ฆ๐ฒ๐ฒ๐ธ-๐ฅ๐ญ ๐๐๐ฟ๐๐ด๐ด๐น๐ฒ๐ ๐ผ๐ป ๐ฑ๐ฎ๐๐ฎ ๐๐ฐ๐ถ๐ฒ๐ป๐ฐ๐ฒ ๐๐ฎ๐๐ธ๐! โ
โก๏ธ How well do reasoning models perform on agentic tasks? Until now, all indicators seemed to show that they worked really well. On our recent reproduction of Deep Search, OpenAI's o1 was by far the best model to power an agentic system.
So when our partner Adyen built a huge benchmark of 450 data science tasks, and built data agents with smolagents to test different models, I expected reasoning models like o1 or DeepSeek-R1 to destroy the tasks at hand.
๐ But they really missed the mark. DeepSeek-R1 only got 1 or 2 out of 10 questions correct. Similarly, o1 was only at ~13% correct answers.
๐ง These results really surprised us. We thoroughly checked them, we even thought our APIs for DeepSeek were broken and colleagues Leandro Anton helped me start custom instances of R1 on our own H100s to make sure it worked well.
But there seemed to be no mistake. Reasoning LLMs actually did not seem that smart. Often, these models made basic mistakes, like forgetting the content of a folder that they had just explored, misspelling file names, or hallucinating data. Even though they do great at exploring webpages through several steps, the same level of multi-step planning seemed much harder to achieve when reasoning over files and data.
It seems like there's still lots of work to do in the Agents x Data space. Congrats to Adyen for this great benchmark, looking forward to see people proposing better agents! ๐
Read more in the blog post ๐ https://huggingface.co/blog/dabstep
โก๏ธ How well do reasoning models perform on agentic tasks? Until now, all indicators seemed to show that they worked really well. On our recent reproduction of Deep Search, OpenAI's o1 was by far the best model to power an agentic system.
So when our partner Adyen built a huge benchmark of 450 data science tasks, and built data agents with smolagents to test different models, I expected reasoning models like o1 or DeepSeek-R1 to destroy the tasks at hand.
๐ But they really missed the mark. DeepSeek-R1 only got 1 or 2 out of 10 questions correct. Similarly, o1 was only at ~13% correct answers.
๐ง These results really surprised us. We thoroughly checked them, we even thought our APIs for DeepSeek were broken and colleagues Leandro Anton helped me start custom instances of R1 on our own H100s to make sure it worked well.
But there seemed to be no mistake. Reasoning LLMs actually did not seem that smart. Often, these models made basic mistakes, like forgetting the content of a folder that they had just explored, misspelling file names, or hallucinating data. Even though they do great at exploring webpages through several steps, the same level of multi-step planning seemed much harder to achieve when reasoning over files and data.
It seems like there's still lots of work to do in the Agents x Data space. Congrats to Adyen for this great benchmark, looking forward to see people proposing better agents! ๐
Read more in the blog post ๐ https://huggingface.co/blog/dabstep
lewtunย
authored
a
paper
about 1 month ago
loubnabnlย
authored
a
paper
about 1 month ago
anton-lย
authored
a
paper
about 1 month ago
thomwolfย
authored
a
paper
about 1 month ago
lvwerraย
authored
a
paper
about 1 month ago
arjunguhaย
authored
a
paper
about 1 month ago