Zayne Sprague

Have that eerie feeling of déjà vu when reading model-generated text 👀, but can’t pinpoint the specific words or phrases 👀? ✨We introduce QUDsim, to quantify discourse similarities beyond lexical, syntactic, and content overlap.

🌟Job ad🌟 We (@gregd_nlp, @mattlease and I) are hiring a postdoc fellow within the CosmicAI Institute, to do galactic work with LLMs and generative AI! If you would like to push the frontiers of foundation models to help solve myths of the universe, please apply!

📢📢📢 Releasing OpenThinker3-1.5B, the top-performing SFT-only model at the 1B scale! 🚀 OpenThinker3-1.5B is a smaller version of our previous 7B model, trained on the same OpenThoughts3-1.2M dataset.

Announcing OpenThinker3-7B, the new SOTA open-data 7B reasoning model: improving over DeepSeek-R1-Distill-Qwen-7B by 33% on average over code, science, and math evals. We also release our dataset, OpenThoughts3-1.2M, which is the best open reasoning dataset across all data…

Solving complex problems with CoT requires combining different skills. We can do this by: 🧩Modify the CoT data format to be “composable” with other skills 🔥Train models on each skill 📌Combine those models Lead to better 0-shot reasoning on tasks involving skill composition!

1/🧵Excited to share CLEVER — a new benchmark for end-to-end verified code generation in Lean. Can we go from natural language to a formally verified Lean program? CLEVER puts this to the test. 📄 arxiv.org/abs/2505.13938 💻 github.com/trishullab/cle…

Introducing ChartMuseum🖼️, testing visual reasoning with diverse real-world charts! ✍🏻Entirely human-written questions by 13 CS researchers 👀Emphasis on visual reasoning – hard to be verbalized via text CoTs 📉Humans reach 93% but 63% from Gemini-2.5-Pro & 38% from Qwen2.5-72B

📣Thrilled to announce I’ll join Carnegie Mellon University (@CMU_EPP & @LTIatCMU) as an Assistant Professor starting Fall 2026! Until then, I’ll be a Research Scientist at @AIatMeta FAIR in SF, working with @kamalikac’s amazing team on privacy, security, and reasoning in LLMs!

Extremely excited to announce that I will be joining @UTAustin @UTCompSci in August 2025 as an Assistant Professor! 🎉 I’m looking forward to continuing to develop AI agents that interact/communicate with people, each other, and the multimodal world. I’ll be recruiting PhD…

With the rise of R1, search seems out of fashion? We prove the opposite! 😎 Introducing Retro-Search 🌈: an MCTS-inspired search algorithm that RETROspectively revises R1’s reasoning traces to synthesize untaken, new reasoning paths that are better 💡, yet shorter in length ⚡️.

I’ll be presenting our work, To CoT or not to CoT? Chain-of-Thought Helps Mainly on Math and Symbolic Reasoning at ICLR as a poster today at 3pm #292. Stop by and chat about reasoning, CoT, LongCoTs, math etc!! See you there 🤘

Excited to be at #ICLR2025 🤩 I'll be giving an oral presentation for Creativity Index on Fri 25th 11:06, Garnet 212&219 🎙️ I'll also be presenting posters: 📍ExploreToM, Sat 26th 10:00, Hall 3 + 2B #49 📍CreativityIndex, Fri 25th 10:30, Hall 3 + 2B #618 Hope to see you there!

Check out my work at @bespokelabsai We release Bespoke-MiniChart-7B, a new SOTA in chart understanding of its size Chart understanding is really fun and challenging and requires reasoning skills beyond math reasoning It's a great starting point for open chart model development!

I will be at #ICLR2025 to present the final project of my PhD (🥲): Chain-of-thought prompting elicits an LLM's knowledge for answering a single question. What about a whole ~set~ of questions? We explore ways to build an LLM's discrete microtheory about a topic of questions.

Evaluating language model responses on open-ended tasks is hard! 🤔 We introduce EvalAgent, a framework that identifies nuanced and diverse criteria 📋✍️. EvalAgent identifies 👩‍🏫🎓 expert advice on the web that implicitly address the user’s prompt 🧵👇

Turns out, it’s possible to outperform DeepSeekR1-32B with only SFT on open data and no RL: Announcing OpenThinker2-32B and OpenThinker2-7B. We also release the data, OpenThoughts2-1M, curated by selecting quality instructions from diverse sources. 🧵 (1/n)

Can we generate long text from compressed KV cache? We find existing KV cache compression methods (e.g., SnapKV) degrade rapidly in this setting. We present 𝐑𝐞𝐟𝐫𝐞𝐬𝐡𝐊𝐕, an inference method which ♻️ refreshes the smaller KV cache, which better preserves performance.

PutnamBench: A math benchmark where no reasoning model can solve even a single problem! We evaluated leading LRMs on the Lean 4 version🧵

Pretty happy that our OpenThinker-32B is in no4 position in the General Reasoning Leaderboard. It should also be pointed out which models are open data (post-training data): OpenThinker, LIMO, OpenHermes and DeepScaler.

Announcing OpenThinker-32B: the best open-data reasoning model distilled from DeepSeek-R1. Our results show that large, carefully curated datasets with verified R1 annotations produce SoTA reasoning models. Our 32B model outperforms all 32B models including…