first commit

ATTRIBUTIONS.md

@@ -1,7 +1,7 @@
 # QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training
 
 <h4 align="left">
-  📄 <a href="https://arxiv.org/abs/2609.19513">Paper (arXiv)</a> · 🤗 <a href="https://huggingface.co/datasets/qvac/GenesisIII">Hugging Face Dataset</a> · 💻 <a href="https://github.com/tether-ai-research/qvac-genesis-III">GitHub</a>
+  📄 <a href="https://arxiv.org/abs/2609.19513">Paper (arXiv)</a> · 🤗 <a href="https://huggingface.co/datasets/qvac/GenesisIII">Hugging Face Dataset</a>
 </h4>
 
 ## Key highlights

README.md

@@ -893,11 +893,6 @@ configs:
 ---
 
 # QVAC Genesis III
-
-This repository contains the QVAC Genesis III dataset associated with the
-COLM 2026 paper
-[QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training](https://arxiv.org/abs/2609.19513).
-
 | Resource | Link |
 | -------- | ---- |
 | 📄 Paper (COLM 2026) | [arXiv:2609.19513](https://arxiv.org/abs/2609.19513) |
@@ -905,8 +900,12 @@ COLM 2026 paper
 | 💻 Code | [github.com/tether-ai-research/qvac-genesis-III](https://github.com/tether-ai-research/qvac-genesis-III) |
 | 🤗 Models | [huggingface.co/collections/qvac/genesis-iii](https://huggingface.co/collections/qvac/genesis-iii) |
 
+---
+
 QVAC Genesis III is a large-scale synthetic STEM corpus for efficient language-model pre-training. It contains **159,646,553 educational documents** and approximately **191 billion training tokens** across 19 curriculum-aligned domains, two complementary generation pipelines, and four educational writing styles.
 
+Genesis III was introduced in **“QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training,” accepted at the Conference on Language Modeling [(COLM) 2026](https://arxiv.org/abs/2609.19513)**.
+
 ![QVAC Genesis III synthetic-data generation pipeline, from seed acquisition and multiple-choice question generation through student answering, LLM parsing, Option-Level Reasoning, and Failure Analysis](https://cdn-uploads.huggingface.co/production/uploads/66ad47f5a45133da70d1c40b/Ha8so8hEmzOpkPWHfbMo3.png)
 
 ## 📋 Dataset Details
@@ -945,6 +944,9 @@ Models trained in the paper's controlled from-scratch experiments:
 | 🔍 Failure Analysis | FA data ablation | [qvac-genesis-iii-qwen3-1.7b-fa](https://huggingface.co/qvac/qvac-genesis-iii-qwen3-1.7b-fa) |
 | 🧠 Option-Level | OL data ablation | [qvac-genesis-iii-qwen3-1.7b-ol](https://huggingface.co/qvac/qvac-genesis-iii-qwen3-1.7b-ol) |
 | 🔗 Combined FA+OL | Full Genesis III mix | [qvac-genesis-iii-qwen3-1.7b-combined](https://huggingface.co/qvac/qvac-genesis-iii-qwen3-1.7b-combined) |
+| 📚 Cosmopedia-v2 (4 epochs) | Controlled baseline | [qvac-genesis-iii-qwen3-1.7b-cosmo4epoch](https://huggingface.co/qvac/qvac-genesis-iii-qwen3-1.7b-cosmo4epoch) |
+| 📚 Cosmopedia-v2 (7 epochs) | Controlled baseline | [qvac-genesis-iii-qwen3-1.7b-cosmo7epoch](https://huggingface.co/qvac/qvac-genesis-iii-qwen3-1.7b-cosmo7epoch) |
+
 ## 🗂️ Dataset Configurations
 
 The dataset has one `train` split and 20 configurations: