In a Training Loop 🔄

Asankhaya Sharma

codelion

http://asankhaya.github.io/

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

upvoted a collection 1 day ago

Nano LLMs

updated a collection 1 day ago

Nano LLMs

updated a collection 1 day ago

Nano LLMs

View all activity

Organizations

upvoted a collection 1 day ago

Nano LLMs

Collection

Really small LLMs pre-trained on data efficient 1 B tokens • 3 items • Updated 1 day ago • 1

upvoted an article 1 day ago

Article

Scaling Pedagogical Pretraining: From Optimal Mixing to 10 Billion Tokens

1 day ago

•

upvoted a collection 5 days ago

🤏 Smol-Data

Collection

Tried and tested mixes for strong pretraining. Inspired by https://huggingface.co/blog/codelion/optimal-dataset-mixing • 14 items • Updated 5 days ago • 11

upvoted a paper about 1 month ago

PaperBanana: Automating Academic Illustration for AI Scientists

Paper • 2601.23265 • Published Jan 30 • 213

upvoted an article about 1 month ago

Article

Reverse Engineering a $500M Mystery: From HashHop to Memory-Augmented Language Models

Jan 23

•

upvoted an article 2 months ago

Article

The Optimal Architecture for Small Language Models

Dec 26, 2025

•

118

upvoted a paper 3 months ago

Universal Reasoning Model

Paper • 2512.14693 • Published Dec 16, 2025 • 43

upvoted an article 3 months ago

Article

Ellora: Enhancing LLMs with LoRA - Standardized Recipes for Capability Enhancement

Dec 3, 2025

•

upvoted a paper 3 months ago

Budget-Aware Tool-Use Enables Effective Agent Scaling

Paper • 2511.17006 • Published Nov 21, 2025 • 33

upvoted an article 4 months ago

Article

The 1 Billion Token Challenge: Finding the Perfect Pre-training Mix

Nov 3, 2025

•

upvoted an article 5 months ago

Article

Python Is All You Need? Introducing Dria-Agent-α

Jan 10, 2025

•

upvoted 2 collections 5 months ago

Sutra Pedagogical Datasets

Collection

High-quality synthetic educational datasets designed for LLM pretraining with structured pedagogical content across 9 knowledge domains. • 6 items • Updated 1 day ago • 1

Dhara Foundational Models

Collection

Diffusion Language Models combining deep narrow networks, Canon layers (depthwise causal convolutions), and WSD (Warmup-Stable-Decay) training. • 2 items • Updated 1 day ago • 3

upvoted a paper 5 months ago

Less is More: Recursive Reasoning with Tiny Networks

Paper • 2510.04871 • Published Oct 6, 2025 • 509

upvoted an article 5 months ago

Article

mem-agent: Equipping LLM Agents with Memory Using RL

Oct 9, 2025

•

upvoted a collection 6 months ago

Mem-Agent

Collection

Small sized agents from Dria trained on interacting with an obsidian-like memory system using python tools. Trained on Qwen3-4B-Thinking-2507. • 4 items • Updated Sep 5, 2025 • 5

upvoted a paper 6 months ago

BeyondWeb: Lessons from Scaling Synthetic Data for Trillion-scale Pretraining

Paper • 2508.10975 • Published Aug 14, 2025 • 60

upvoted an article 6 months ago

Article

mem-agent: Persistent, Human Readable Memory Agent Trained with Online RL

Sep 11, 2025

•

upvoted a collection 7 months ago

Nemotron-Pre-Training-Datasets

Collection

Large scale pre-training datasets used in the Nemotron family of models. • 12 items • Updated 1 day ago • 103

upvoted an article 7 months ago

Article

Building Enterprise-Ready Text Classifiers in Minutes with Adaptive Learning

Aug 9, 2025

•

Asankhaya Sharma

AI & ML interests

Recent Activity

Organizations

codelion's activity

Scaling Pedagogical Pretraining: From Optimal Mixing to 10 Billion Tokens

Reverse Engineering a $500M Mystery: From HashHop to Memory-Augmented Language Models

The Optimal Architecture for Small Language Models

Ellora: Enhancing LLMs with LoRA - Standardized Recipes for Capability Enhancement

The 1 Billion Token Challenge: Finding the Perfect Pre-training Mix

Python Is All You Need? Introducing Dria-Agent-α

mem-agent: Equipping LLM Agents with Memory Using RL

mem-agent: Persistent, Human Readable Memory Agent Trained with Online RL

Building Enterprise-Ready Text Classifiers in Minutes with Adaptive Learning