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
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
Presenting a simple re-implementation of "Inference-time scaling diffusion models beyond denoising steps" by Ma et al.
I did the simplest random search strategy, but results can potentially be improved with better-guided search methods.
Supports Gemini 2 Flash & Qwen2.5 as verifiers for "LLMGrading" ๐ค
The steps are simple:
For each round:
1> Starting by sampling 2 starting noises with different seeds. 2> Score the generations w.r.t a metric. 3> Obtain the best generation from the current round.
If you have more compute budget, go to the next search round. Scale the noise pool (2 ** search_round) and repeat 1 - 3.
This constitutes the random search method as done in the paper by Google DeepMind.
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.
We have been cooking a couple of fine-tuning runs on CogVideoX with finetrainers, smol datasets, and LoRA to generate cool video effects like crushing, dissolving, etc.
We are also releasing a LoRA extraction utility from a fully fine-tuned checkpoint. I know that kind of stuff has existed since eternity, but the quality on video models was nothing short of spectacular. Below are some links:
We are reproducing the full DeepSeek R1 data and training pipeline so everybody can use their recipe. Instead of doing it in secret we can do it together in the open!
๐งช Step 1: replicate the R1-Distill models by distilling a high-quality reasoning corpus from DeepSeek-R1.
๐ง Step 2: replicate the pure RL pipeline that DeepSeek used to create R1-Zero. This will involve curating new, large-scale datasets for math, reasoning, and code.
๐ฅ Step 3: show we can go from base model -> SFT -> RL via multi-stage training.
Cosmos is a family of pre-trained models purpose-built for generating physics-aware videos and world states to advance physical AI development. The release includes Tokenizers nvidia/cosmos-tokenizer-672b93023add81b66a8ff8e6
I was initially pretty sceptical about Meta's Coconut paper [1] because the largest perf gains were reported on toy linguistic problems. However, these results on machine translation are pretty impressive!
* Iteratively sample CoTs from the model, using a mix of different search strategies. This gives you something like Stream of Search via prompting. * Verify correctness of each CoT using GPT-4o (needed because exact match doesn't work well in medicine where there are lots of aliases) * Use GPT-4o to reformat the concatenated CoTs into a single stream that includes smooth transitions like "hmm, wait" etc that one sees in o1 * Use the resulting data for SFT & RL * Use sparse rewards from GPT-4o to guide RL training. They find RL gives an average ~3 point boost across medical benchmarks and SFT on this data already gives a strong improvement.
Applying this strategy to other domains could be quite promising, provided the training data can be formulated with verifiable problems!
* 4 new video models * Multiple image models, including SANA & Flux Control * New quantizers -> GGUF & TorchAO * New training scripts Enjoy this holiday-special Diffusers release ๐ค Notes: https://github.com/huggingface/diffusers/releases/tag/v0.32.0