Thank you ๐ซก !
Asankhaya Sharma
codelion
AI & ML interests
Creator of OptiLLM, OpenEvolve, Adaptive Classifier, and Ellora. Pioneering a new category in AI infrastructure: inference-time compute for LLMs.
Recent Activity
updated
a collection
about 14 hours ago
Sutra Pedagogical Datasets liked
a dataset about 14 hours ago
codelion/sutra-improved-100M updated
a dataset about 14 hours ago
codelion/sutra-improved-100M Organizations
replied to their post 5 days ago
reacted to their post with ๐ค๐๐๐ฅ 8 days ago
Post
3086
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 codelion/sutra-10B, a 10.2 billion token pedagogical dataset with rich metadata (domain, complexity, prerequisites, quality scores) on every entry.
We trained 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.
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 codelion/sutra-10B, a 10.2 billion token pedagogical dataset with rich metadata (domain, complexity, prerequisites, quality scores) on every entry.
We trained 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.
-
- 2 replies
posted an
update 8 days ago
Post
3086
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 codelion/sutra-10B, a 10.2 billion token pedagogical dataset with rich metadata (domain, complexity, prerequisites, quality scores) on every entry.
We trained 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.
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 codelion/sutra-10B, a 10.2 billion token pedagogical dataset with rich metadata (domain, complexity, prerequisites, quality scores) on every entry.
We trained 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.
-
-
- 2 replies
reacted to their post with โค๏ธ๐๐๐ฅ about 2 months ago
Post
3230
Reverse Engineering a $500M Mystery: From HashHop to Memory-Augmented Language Models
I wrote a deep dive into how Magic AI's 100M token context window might work, starting from their HashHop benchmark and building up to MALM - a Memory-Augmented Language Model.
Key insight: treating each key as a single token enables perfect retrieval at unlimited context lengths.
The article covers:
- How HashHop works and why its perfect accuracy is suspicious
- Building a tokenized solver that achieves 100% accuracy
- Scaling to MALM for real code search tasks
- Why this approach could handle 100M+ tokens
Read the full article: https://huggingface.co/blog/codelion/reverse-engineering-magic-hashhop
Try the model: codelion/malm-165m
Code: https://github.com/codelion/hash-hop
I wrote a deep dive into how Magic AI's 100M token context window might work, starting from their HashHop benchmark and building up to MALM - a Memory-Augmented Language Model.
Key insight: treating each key as a single token enables perfect retrieval at unlimited context lengths.
The article covers:
- How HashHop works and why its perfect accuracy is suspicious
- Building a tokenized solver that achieves 100% accuracy
- Scaling to MALM for real code search tasks
- Why this approach could handle 100M+ tokens
Read the full article: https://huggingface.co/blog/codelion/reverse-engineering-magic-hashhop
Try the model: codelion/malm-165m
Code: https://github.com/codelion/hash-hop
- 1 reply
posted an
update about 2 months ago
Post
3230
Reverse Engineering a $500M Mystery: From HashHop to Memory-Augmented Language Models
I wrote a deep dive into how Magic AI's 100M token context window might work, starting from their HashHop benchmark and building up to MALM - a Memory-Augmented Language Model.
Key insight: treating each key as a single token enables perfect retrieval at unlimited context lengths.
The article covers:
- How HashHop works and why its perfect accuracy is suspicious
- Building a tokenized solver that achieves 100% accuracy
- Scaling to MALM for real code search tasks
- Why this approach could handle 100M+ tokens
Read the full article: https://huggingface.co/blog/codelion/reverse-engineering-magic-hashhop
Try the model: codelion/malm-165m
Code: https://github.com/codelion/hash-hop
I wrote a deep dive into how Magic AI's 100M token context window might work, starting from their HashHop benchmark and building up to MALM - a Memory-Augmented Language Model.
Key insight: treating each key as a single token enables perfect retrieval at unlimited context lengths.
The article covers:
- How HashHop works and why its perfect accuracy is suspicious
- Building a tokenized solver that achieves 100% accuracy
- Scaling to MALM for real code search tasks
- Why this approach could handle 100M+ tokens
Read the full article: https://huggingface.co/blog/codelion/reverse-engineering-magic-hashhop
Try the model: codelion/malm-165m
Code: https://github.com/codelion/hash-hop
-
- 1 reply
reacted to their post with ๐๐ค๐๐ฅ 3 months ago
Post
6137
Introducing Dhara-70M: A diffusion language model that achieves 3.8x higher throughput than autoregressive models!
Key findings from our research on optimal architectures for small language models:
โ Depth beats width: 32 layers outperforms 12 layers at the same parameter count
โ Best-in-class factuality: 47.5% on TruthfulQA
โ 10x training efficiency using WSD (Warmup-Stable-Decay) conversion
โ Canon layers add only 0.13% parameters but improve reasoning
We trained on 1B tokens using the optimal 50-30-20 dataset mix (PDFs + filtered web + educational content), then converted to diffusion with just 100M additional tokens.
Blog: https://huggingface.co/blog/codelion/optimal-model-architecture
Model: codelion/dhara-70m
Key findings from our research on optimal architectures for small language models:
โ Depth beats width: 32 layers outperforms 12 layers at the same parameter count
โ Best-in-class factuality: 47.5% on TruthfulQA
โ 10x training efficiency using WSD (Warmup-Stable-Decay) conversion
โ Canon layers add only 0.13% parameters but improve reasoning
We trained on 1B tokens using the optimal 50-30-20 dataset mix (PDFs + filtered web + educational content), then converted to diffusion with just 100M additional tokens.
Blog: https://huggingface.co/blog/codelion/optimal-model-architecture
Model: codelion/dhara-70m
- 1 reply
posted an
update 3 months ago
Post
6137
Introducing Dhara-70M: A diffusion language model that achieves 3.8x higher throughput than autoregressive models!
Key findings from our research on optimal architectures for small language models:
โ Depth beats width: 32 layers outperforms 12 layers at the same parameter count
โ Best-in-class factuality: 47.5% on TruthfulQA
โ 10x training efficiency using WSD (Warmup-Stable-Decay) conversion
โ Canon layers add only 0.13% parameters but improve reasoning
We trained on 1B tokens using the optimal 50-30-20 dataset mix (PDFs + filtered web + educational content), then converted to diffusion with just 100M additional tokens.
Blog: https://huggingface.co/blog/codelion/optimal-model-architecture
Model: codelion/dhara-70m
Key findings from our research on optimal architectures for small language models:
โ Depth beats width: 32 layers outperforms 12 layers at the same parameter count
โ Best-in-class factuality: 47.5% on TruthfulQA
โ 10x training efficiency using WSD (Warmup-Stable-Decay) conversion
โ Canon layers add only 0.13% parameters but improve reasoning
We trained on 1B tokens using the optimal 50-30-20 dataset mix (PDFs + filtered web + educational content), then converted to diffusion with just 100M additional tokens.
Blog: https://huggingface.co/blog/codelion/optimal-model-architecture
Model: codelion/dhara-70m
-
- 1 reply
reacted to their post with ๐๐ค๐๐ฅ 3 months ago
Post
2416
Introducing PTS Visualizer - an interactive tool for exploring how language models reason!
Visualize pivotal tokens, thought anchors, and reasoning circuits. See which tokens and sentences significantly impact success probability, explore embedding clusters, and trace reasoning step-by-step.
Try it: codelion/pts-visualizer
Explore PTS datasets:
- Qwen3-0.6B: codelion/Qwen3-0.6B-pts
- DeepSeek-R1: codelion/DeepSeek-R1-Distill-Qwen-1.5B-pts
Or upload your own JSONL files!
GitHub: https://github.com/codelion/pts
Visualize pivotal tokens, thought anchors, and reasoning circuits. See which tokens and sentences significantly impact success probability, explore embedding clusters, and trace reasoning step-by-step.
Try it: codelion/pts-visualizer
Explore PTS datasets:
- Qwen3-0.6B: codelion/Qwen3-0.6B-pts
- DeepSeek-R1: codelion/DeepSeek-R1-Distill-Qwen-1.5B-pts
Or upload your own JSONL files!
GitHub: https://github.com/codelion/pts