Hugging Face's logo
Nick B's picture
1 1

Nick B PRO

drrockso72
·

AI & ML interests

None yet

Recent Activity

replied to codelion's post about 16 hours ago
Scaling Pedagogical Pre-training to 10 Billion Tokens New blog post exploring what happens when you take optimal data mixing insights and scale up the data generation itself. We built Sutra, a multi-stage framework for generating pedagogical pre-training data guided by a knowledge graph of ~2,000 concepts across 9 domains. The pipeline includes structured content generation, six-dimension quality evaluation, diversity management across 20 content styles, and a cleaning stage to prevent collapse. The result is https://huggingface.co/datasets/codelion/sutra-10B, a 10.2 billion token pedagogical dataset with rich metadata (domain, complexity, prerequisites, quality scores) on every entry. We trained https://huggingface.co/codelion/SmolLM2-70M on it for 3 full epochs (30.6B tokens) on a single A10 GPU in ~78 hours. Key finding: perplexity kept improving across epochs, but benchmark gains plateaued fast. At 70M parameters, the model hits a representational ceiling that more data alone can't break through. Full writeup with comparisons against 7 other datasets, detailed benchmark breakdowns, and connections to recent work on synthetic data scaling, curriculum learning, and data mixing laws: https://huggingface.co/blog/codelion/scaling-pedagogical-pretraining-10-billion-tokens All datasets at multiple scales (10M, 100M, 1B, 10B) plus seed concepts and an SFT variant are in the Sutra Pedagogical Datasets collection.
reacted to codelion's post with 🔥 about 16 hours ago
Scaling Pedagogical Pre-training to 10 Billion Tokens New blog post exploring what happens when you take optimal data mixing insights and scale up the data generation itself. We built Sutra, a multi-stage framework for generating pedagogical pre-training data guided by a knowledge graph of ~2,000 concepts across 9 domains. The pipeline includes structured content generation, six-dimension quality evaluation, diversity management across 20 content styles, and a cleaning stage to prevent collapse. The result is https://huggingface.co/datasets/codelion/sutra-10B, a 10.2 billion token pedagogical dataset with rich metadata (domain, complexity, prerequisites, quality scores) on every entry. We trained https://huggingface.co/codelion/SmolLM2-70M on it for 3 full epochs (30.6B tokens) on a single A10 GPU in ~78 hours. Key finding: perplexity kept improving across epochs, but benchmark gains plateaued fast. At 70M parameters, the model hits a representational ceiling that more data alone can't break through. Full writeup with comparisons against 7 other datasets, detailed benchmark breakdowns, and connections to recent work on synthetic data scaling, curriculum learning, and data mixing laws: https://huggingface.co/blog/codelion/scaling-pedagogical-pretraining-10-billion-tokens All datasets at multiple scales (10M, 100M, 1B, 10B) plus seed concepts and an SFT variant are in the Sutra Pedagogical Datasets collection.
upvoted an article 1 day ago
Scaling Pedagogical Pre-training: From Optimal Mixing to 10 Billion Tokens
View all activity

Organizations

None yet