It's beating Claude 3.7 on (competitive) programming –a domain Anthropic has been historically really strong at– 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
We find that OlympicCoder models outperform Claude 3.7 Sonnet, as well as others over 100x larger 💪
Together with the models, we are releasing:
📊CodeForces-CoTs: new dataset of code problems from the most popular competitive coding platform, with R1 traces in C++ and Python open-r1/codeforces-cots
🏆 IOI'2024: a new benchmark of VERY hard programming problems where even frontier models struggle to match human performance open-r1/ioi
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. ✅
🏆 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. 💪
(Note that results will be added manually, so the leaderboard might not always have the latest LLMs)
Extremely bullish on @CohereForAI's Aya Vision (8B & 32B) - new SOTA open-weight VLMs
- 8B wins up to 81% of the time in its class, better than Gemini Flash - 32B beats Llama 3.2 90B! - Covers 23 languages, excels in image captioning, VQA & more - Integrated on transformers from Day 0!
We now have a Deep Research for academia: SurveyX automatically writes academic surveys nearly indistinguishable from human-written ones 🔥
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.
🎯 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 “AttributeTree” object that structures key information from citations. Ablating these AttributeTrees significantly decreased structure and synthesis scores, so they were really useful!
📝 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 🏆
Google just released PaliGemma 2 Mix: new versatile instruction vision language models 🔥
> Three new models: 3B, 10B, 28B with res 224, 448 💙 > Can do vision language tasks with open-ended prompts, understand documents, and segment or detect anything 🤯
Less is More for Reasoning (LIMO): a 32B model fine-tuned with 817 examples can beat o1-preview on math reasoning! 🤯
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 —no 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.
⚡ The Less-is-More Reasoning Hypothesis: ‣ Minimal but precise examples that showcase optimal reasoning patterns matter more than sheer quantity ‣ Pre-training knowledge plus sufficient computational resources at inference levels up math skills
➡️ Core techniques: ‣ High-quality reasoning chains with self-verification steps ‣ 817 handpicked problems that encourage deeper reasoning ‣ Enough inference-time computation to allow extended reasoning
💪 Efficiency gains: ‣ Only 817 examples instead of 100k+ ‣ 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 🚀
👀 Multimodal > OpenGVLab released InternVideo 2.5 Chat models, new video LMs with long context > AIDC released Ovis2 model family along with Ovis dataset, new vision LMs in different sizes (1B, 2B, 4B, 8B, 16B, 34B), with video and OCR support > ColQwenStella-2b is a multilingual visual retrieval model that is sota in it's size > Hoags-2B-Exp is a new multilingual vision LM with contextual reasoning, long context video understanding
💬 LLMs A lot of math models! > Open-R1 team released OpenR1-Math-220k large scale math reasoning dataset, along with Qwen2.5-220K-Math fine-tuned on the dataset, OpenR1-Qwen-7B > Nomic AI released new Nomic Embed multilingual retrieval model, a MoE with 500 params with 305M active params, outperforming other models > DeepScaleR-1.5B-Preview is a new DeepSeek-R1-Distill fine-tune using distributed RL on math > LIMO is a new fine-tune of Qwen2.5-32B-Instruct on Math
🗣️ Audio > Zonos-v0.1 is a new family of speech recognition models, which contains the model itself and embeddings
🖼️ Vision and Image Generation > We have ported DepthPro of Apple to transformers for your convenience! > illustrious-xl-v1.0 is a new illustration generation model
𝗚𝗿𝗲𝗮𝘁 𝗳𝗲𝗮𝘁𝘂𝗿𝗲 𝗮𝗹𝗲𝗿𝘁: you can now share agents to the Hub! 🥳🥳
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.
The community has been busy distilling DeepSeek-R1 from inference providers, but we decided to have a go at doing it ourselves from scratch 💪
What’s new compared to existing reasoning datasets?
♾ 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.
🐳 800k R1 reasoning traces: We generate two answers for 400k problems using DeepSeek R1. The filtered dataset contains 220k problems with correct reasoning traces.
📀 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.
⏳ 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 can’t be verified with a rules-based parser)
📊 We match the performance of DeepSeek-Distill-Qwen-7B by finetuning Qwen-7B-Math-Instruct on our dataset.
🤖 Robotics > Pi0, first open-source foundation vision-language action model was released in Le Robot (Apache 2.0)
💬 LLMs > Groundbreaking: s1 is simpler approach to test-time scaling, the release comes with small s1K dataset of 1k question-reasoning trace pairs (from Gemini-Thinking Exp) they fine-tune Qwen2.5-32B-Instruct to get s1-32B, outperforming o1-preview on math 🤯 s1-32B and s1K is out! > Adyen released DABstep, a new benchmark along with it's leaderboard demo for agents doing data analysis > Krutrim released Krutrim-2 instruct, new 12B model based on NeMo12B trained and aligned on Indic languages, a new multilingual sentence embedding model (based on STSB-XLM-R), and a translation model for Indic languages
👀 Multimodal > PKU released Align-DS-V, a model aligned using their new technique called LLF for all modalities (image-text-audio), along with the dataset Align Anything > OLA-7B is a new any-to-any model by Tencent that can take text, image, video, audio data with context window of 32k tokens and output text and speech in English and Chinese > Krutrim released Chitrarth, a new vision language model for Indic languages and English
🖼️ Vision > BiRefNet_HR is a new higher resolution BiRefNet for background removal
🗣️ Audio > kyutai released Hibiki, it's a real-time speech-to-speech translation model 🤯 it's available for French-English translation > Krutrim released Dhwani, a new STT model for Indic languages > They also release a new dataset for STT-TTS
🖼️ Image Generation > Lumina released Lumina-Image-2.0, a 2B parameter-flow based DiT for text to image generation > Tencent released Hunyuan3D-2, a 3D asset generation model based on DiT and Hunyuan3D-Paint > boreal-hl-v1 is a new boring photorealistic image generation LoRA based on Hunyuan
➡️ 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.
👎 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.
🧐 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! 🚀