Sachin Goyal

Distilled Pretraining: A modern lens of Data, In-Context Learning and Test-Time Scaling

Beyond Multi-Token Prediction: Pretraining LLMs with Future Summaries

Mode-conditioning unlocks superior test-time compute scaling

Context-Parametric Inversion: Why Instruction Finetuning May Not Actually Improve Context Reliance

Overtrained Language Models Are Harder to Fine-Tune

Safety Pretraining: Toward the Next Generation of Safe AI

Inference Optimal VLMs Need Fewer Visual Tokens and More Parameters