#OpenThoughts
LightNews LightNews
ahochlehnert.bsky.social
CuratedThoughts: Data Curation for RL Datasets 🚀

Since DeepSeek-R1 introduced reasoning-based RL, datasets like Open-R1 & OpenThoughts emerged for fine-tuning & GRPO. Our deep dive found major flaws — 25% of OpenThoughts needed elimination by data curation.

Here's why 👇🧵
February 17, 2025 at 6:22 PM
sungkim.bsky.social
Open Thoughts project

They are building the best reasoning datasets out in the open.

Building off their work with Stratos, today they are releasing OpenThoughts-114k and OpenThinker-7B.

Repo: github.com/open-thought...
January 29, 2025 at 6:49 AM
sungkim.bsky.social
Reasoning Datasets collections by @philschmid.bsky.social

1️⃣ ServiceNow-AI/R1-Distill-SFT
2️⃣ open-thoughts/OpenThoughts-114k
3️⃣ bespokelabs/Bespoke-Stratos-17k
4️⃣ EricLu/SCP-116K
5️⃣ cognitivecomputations/dolphin-r1

huggingface.co/collections/...
February 2, 2025 at 4:26 PM
cottonbranch.org
Good morning. 🌸 Let your thoughts open wide so the day can meet you with fresh insight. 💛
#OpenThoughts #InsightAhead #MorningExpansion
January 29, 2026 at 12:05 PM
alexdimakis.bsky.social
Qwen32B. post-training data is open and available as openthoughts 114k huggingface.co/datasets/ope...
open-thoughts/OpenThoughts-114k · Datasets at Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
February 12, 2025 at 10:31 PM
bayesiankitten.bsky.social
CuratedThoughts: Data curation focus for RL post-training! (Update 1) 🚀

25% of Openthoughts-114k-math filtered — issues included proofs, missing figures, and multiple questions with one answer.

Check out work by
@ahochlehnert.bsky.social & @hrdkbhatnagar.bsky.social
below 👇
ahochlehnert.bsky.social Andreas Hochlehnert @ahochlehnert.bsky.social · Feb 17
CuratedThoughts: Data Curation for RL Datasets 🚀

Since DeepSeek-R1 introduced reasoning-based RL, datasets like Open-R1 & OpenThoughts emerged for fine-tuning & GRPO. Our deep dive found major flaws — 25% of OpenThoughts needed elimination by data curation.

Here's why 👇🧵
February 17, 2025 at 6:30 PM
anandraghavan.bsky.social
OpenThoughts-Agent: Data Recipes for Agentic Models arxiv.org/abs/2606.24855
OpenThoughts-Agent: Data Recipes for Agentic Models
Agentic language models dramatically expand the applications of AI yet little is publicly known about how to curate training data for broadly capable agents. Existing open efforts such as SWE-Smith,…
arxiv.org
July 28, 2026 at 6:02 PM
natolambert.bsky.social
2025-05-21 - Hunyuan-TurboS - arxiv.org/abs/2505.15431
2025-05-28 - Skywork OR-1 - arxiv.org/abs/2505.22312
2025-06-04 - Xiaomi MiMo VL - arxiv.org/abs/2506.03569
2025-06-04 - OpenThoughts - arxiv.org/abs/2506.04178
2025-06-10 - Magistral - arxiv.org/abs/2506.10910
Hunyuan-TurboS: Advancing Large Language Models through Mamba-Transformer Synergy and Adaptive Chain-of-Thought
As Large Language Models (LLMs) rapidly advance, we introduce Hunyuan-TurboS, a novel large hybrid Transformer-Mamba Mixture of Experts (MoE) model. It synergistically combines Mamba's long-sequence…
arxiv.org
January 2, 2026 at 4:38 PM
paper.bsky.social
[30/30] 192 Likes, 2 Comments, 3 Posts
2506.04178, cs․LG, 05 Jun 2025

🆕OpenThoughts: Data Recipes for Reasoning Models

Etash Guha, Ryan Marten, Sedrick Keh, Negin Raoof, Georgios Smyrnis, Hritik Bansal, Marianna Nezhurina, Jean Mercat, Trung Vu, Zayne Sprague, Ashima Suvarna, Benjamin Feuer, ...
Reasoning models have made rapid progress on many benchmarks involving math, code, and science. Yet, there are still many open questions about the best training recipes for reasoning since state-of-the-art models often rely on proprietary datasets with little to no public information available. To address this, the goal of the OpenThoughts project is to create open-source datasets for training reasoning models. After initial explorations, our OpenThoughts2-1M dataset led to OpenThinker2-32B, the first model trained on public reasoning data to match DeepSeek-R1-Distill-32B on standard reasoning benchmarks such as AIME and LiveCodeBench. We then improve our dataset further by systematically investigating each step of our data generation pipeline with 1,000+ controlled experiments, which led to OpenThoughts3. Scaling the pipeline to 1.2M examples and using QwQ-32B as teacher yields our OpenThoughts3-7B model, which achieves state-of-the-art results: 53% on AIME 2025, 51% on LiveCodeBench 06/24-01/25, and 54% on GPQA Diamond - improvements of 15.3, 17.2, and 20.5 percentage points compared to the DeepSeek-R1-Distill-Qwen-7B. All of our datasets and models are available on https://openthoughts.ai.
June 28, 2025 at 12:06 AM
abursuc.bsky.social
I recommend checking out very recent works from Jenia & collaborators:
- Scaling Laws for Robust Comparison of Open Foundation Language-Vision Models and Datasets arxiv.org/abs/2506.04598
- OpenThoughts: Data Recipes for Reasoning Models arxiv.org/abs/2506.04178
July 10, 2025 at 10:33 PM
drtimos.bsky.social
Thoughts on this? >> OpenThoughts: A Scalable Supervised Fine-Tuning SFT Data
Curation Pipeline for Reasoning Models >> Comment below! #healthtech #industry40 #IoT #AI #mhealth
OpenThoughts: A Scalable Supervised Fine-Tuning SFT Data Curation Pipeline for Reasoning Models
The Growing Complexity of Reasoning Data Curation Recent reasoning models, such as DeepSeek-R1 and o3, have shown outstanding performance in mathematical, coding, and scientific areas, utilizing post-training techniques like supervised fine-tuning (SFT) and reinforcement learning (RL). However, the complete methodologies behind these frontier reasoning models are not public, which makes research for building reasoning models […] The post OpenThoughts: A Scalable Supervised Fine-Tuning SFT Data Curation Pipeline for Reasoning Models appeared first on MarkTechPost.
dlvr.it
June 14, 2025 at 5:55 AM
shangshang-wang.bsky.social
We also want to express our gratitude to the broader open-source community. This research was made possible by leveraging numerous publicly available resources from DeepScaleR, STILL, OpenThoughts @bespokelabs.bsky.social , OpenR1 @hf.co , LIMR, and OpenRS.
April 23, 2025 at 5:10 PM
justinhjohnson.com
Open-sourcing reasoning just levelled-up! 🧠 📜 OpenThoughts3 releases a 1.2 M example recipe + OpenThinker3-7B that tops AIME 25 (53 %) & GPQA-D (54 %), beating previous open models by 15-20 pp. Full data + code at the link—dig in! bit.ly/3TCvk9l #OpenSource #LLM #Reasoning
OpenThoughts: Data Recipes for Reasoning Models
Reasoning models have made rapid progress on many benchmarks involving math, code, and science. Yet, there are still many open questions about the best training recipes for reasoning since…
bit.ly
July 14, 2025 at 11:02 AM
deepseek.activitypub.awakari.com.ap.brid.gy
OpenThoughts: A Scalable Supervised Fine-Tuning SFT Data Curation Pipeline for Reasoning Models The Growing Complexity of Reasoning Data Curation Recent reasoning models, such as DeepSeek-R1 and o3...

#AI #Paper #Summary #AI #Shorts #Applications […]

[Original post on marktechpost.com]
June 14, 2025 at 3:57 AM
nanoatp.bsky.social
[30/30] 174 Upvotes, 0 Comments, 2 Posts, arXiv:2606.24855

🆕OpenThoughts-Agent: Data Recipes for Agentic Models

Negin Raoof, Richard Zhuang, Marianna Nezhurina, Etash Guha, Atula Tejaswi, Ryan Marten
Agentic language models dramatically expand the applications of AI yet little is publicly known about how to curate training data for broadly capable agents. Existing open efforts such as SWE-Smith, SERA, and Nemotron-Terminal typically target a single benchmark, leaving open the question of how to train models that generalize across diverse agentic tasks. The OpenThoughts-Agent (OT-Agent) project addresses this gap with a fully open data curation pipeline for training agentic models. We conduct more than 100 controlled ablation experiments to systematically investigate each stage of the pipeline, yielding insights on the importance of task sources and diversity. We then assemble a training set of 100K examples from our pipeline and fine-tune Qwen3-32B on this dataset, which yields an average accuracy of 44.8% across seven agentic benchmarks and a 3.9 percentage point improvement over the strongest existing open data agentic model (Nemotron-Terminal-32B, 40.9%). Moreover, our training data exhibits strong scaling properties, outperforming alternative open datasets at every training set size in compute-controlled comparisons. We publicly release our training sets, data pipeline, experimental data, and models at openthoughts.ai to support future open research on agentic model training.
July 18, 2026 at 12:04 AM
kabirkumar.bsky.social
right now, the ai plans website looks dogshit. by the morning, i will have changed this
the https://beta.ai-plans.com website AI-Plans website interface showing a peer review platform for AI alignment research. The left sidebar contains navigation menu items including Papers, People, Guides, Courses, Join our Team, News, About, Submit, and Donate buttons. The main content area displays category filters for "Plans" and "Papers", topic filters including Implementation, Conception, Prosaic Alignment, and Value Learning, and a search bar. Two research papers are listed: "OpenThoughts: Data Recipes for Reasoning Models" and "TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference", both showing 0.00 scores with 0 strengths and 0 flaws. A welcome popup on the right explains that AI-Plans is a peer review platform where users can comment on and vote on AI alignment plans to earn reputation points.
June 7, 2025 at 1:29 AM
nanoatp.bsky.social
[30/30] 174 Upvotes, 0 Comments, 2 Posts, arXiv:2606.24855

🆕OpenThoughts-Agent: Data Recipes for Agentic Models

Negin Raoof, Richard Zhuang, Marianna Nezhurina, Etash Guha, Atula Tejaswi, Ryan Marten
Agentic language models dramatically expand the applications of AI yet little is publicly known about how to curate training data for broadly capable agents. Existing open efforts such as SWE-Smith, SERA, and Nemotron-Terminal typically target a single benchmark, leaving open the question of how to train models that generalize across diverse agentic tasks. The OpenThoughts-Agent (OT-Agent) project addresses this gap with a fully open data curation pipeline for training agentic models. We conduct more than 100 controlled ablation experiments to systematically investigate each stage of the pipeline, yielding insights on the importance of task sources and diversity. We then assemble a training set of 100K examples from our pipeline and fine-tune Qwen3-32B on this dataset, which yields an average accuracy of 44.8% across seven agentic benchmarks and a 3.9 percentage point improvement over the strongest existing open data agentic model (Nemotron-Terminal-32B, 40.9%). Moreover, our training data exhibits strong scaling properties, outperforming alternative open datasets at every training set size in compute-controlled comparisons. We publicly release our training sets, data pipeline, experimental data, and models at openthoughts.ai to support future open research on agentic model training.
July 19, 2026 at 12:03 AM
teardropluv.bsky.social
This Real Talk #OpenThoughts and #Wave interview with Kentrell #NBAYoungboy is top tier
August 12, 2026 at 4:36 AM
hkauthor.bsky.social
www.tiktok.com/t/ZP8tsnduM/
#clips #marco #dijonaicarrington #funnymarco #openthoughts
TikTok video by CaughtOnClip
www.tiktok.com
July 31, 2026 at 4:48 PM
nanoatp.bsky.social
(2/2) 47 Upvotes, 0 Comments, 23 Jun 2026, Hugging Face
Paper page - OpenThoughts-Agent: Data Recipes for Agentic Models
Join the discussion on this paper page
huggingface.co
July 19, 2026 at 12:04 AM
nanoatp.bsky.social
(1/2) 127 Upvotes, 0 Comments, 23 Jun 2026, alphaXiv
OpenThoughts-Agent: Data Recipes for Agentic Models
This research systematically investigates and establishes an open data curation pipeline for training broadly capable agentic language models, identifying effective strategies for supervised...
www.alphaxiv.org
July 19, 2026 at 12:04 AM
nanoatp.bsky.social
Links: abs, pdf
Search: Bluesky, Twitter, Reddit, Hacker News, Hugging Face, alphaXiv
OpenThoughts-Agent: Data Recipes for Agentic Models
Agentic language models dramatically expand the applications of AI yet little is publicly known about how to curate training data for broadly capable agents. Existing open efforts such as SWE-Smith, S...
arxiv.org
July 19, 2026 at 12:04 AM
nanoatp.bsky.social
エージェント型言語モデルはAIの応用範囲を飛躍的に広げているが、幅広い能力を持つエージェントのための学習データをどのように選定すべきかについては、公に知られていることはほとんどない。

SWE-Smith、SERA、Nemotron-Terminal といった既存のオープンな取り組みは、通常、単一のベンチマークを対象としているため、多様なエージェントタスクに汎化できるモデルをどのように学習させるかという課題は未解決のままである。

OpenThoughts-Agent（OT-Agent）プロジェクトは、エージェントモデルの学習に向けた完全にオープンなデータキュレーションパイプラインを提供...
エージェント型言語モデルはAIの応用範囲を飛躍的に広げているが、幅広い能力を持つエージェントのための学習データをどのように選定すべきかについては、公に知られていることはほとんどない。 SWE-Smith、SERA、Nemotron-Terminal といった既存のオープンな取り組みは、通常、単一のベンチマークを対象としているため、多様なエージェントタスクに汎化できるモデルをどのように学習させるかという課題は未解決のままである。 OpenThoughts-Agent（OT-Agent）プロジェクトは、エージェントモデルの学習に向けた完全にオープンなデータキュレーションパイプラインを提供することで、この課題に取り組んでいます。 パイプラインの各段階を体系的に調査するため、100件以上の制御されたアブレーション実験を実施し、タスクソースと多様性の重要性に関する知見を得た。 次に、パイプラインから10万例のトレーニングセットを構築し、このデータセットを用いてQwen3-32Bの微調整を行った。その結果、7つのエージェント型ベンチマークにおいて平均精度44.8%を達成し、既存のオープンデータを用いたエージェント型モデルの中で最も優れたモデル （Nemotron-Terminal-32B、40.9%）を3.9パーセントポイント上回る結果となった。 さらに、当社のトレーニングデータは強力なスケーリング特性を示しており、計算リソースを制御した比較において、あらゆるトレーニングセットの規模で、他の公開データセットを上回る性能を発揮しています。 openthoughts.ai では、エージェント型モデルの学習に関する今後のオープンリサーチを支援するため、トレーニングセット、データパイプライン、実験データ、およびモデルを公開しています。
July 19, 2026 at 12:04 AM
nanoatp.bsky.social
(2/2) 47 Upvotes, 0 Comments, 23 Jun 2026, Hugging Face
Paper page - OpenThoughts-Agent: Data Recipes for Agentic Models
Join the discussion on this paper page
huggingface.co
July 18, 2026 at 12:04 AM
nanoatp.bsky.social
(1/2) 127 Upvotes, 0 Comments, 23 Jun 2026, alphaXiv
OpenThoughts-Agent: Data Recipes for Agentic Models
This research systematically investigates and establishes an open data curation pipeline for training broadly capable agentic language models, identifying effective strategies for supervised...
www.alphaxiv.org
July 18, 2026 at 12:04 AM