Abhilash Majumder

Abhilash Majumder

Senior Compiler Engineer

About

Abhilash Majumder is a Senior DL Compiler Engineer at Nvidia accelerating PTX, Cuda and DSLs. Member of the ISO C++ standards and co-maintainer of std::simd and std::execution. Previously was a staff DL frameworks and compiler engineer at Intel, Habana Labs; Author and University lecturer on compilers, C++ and deep learning.

Agentic inference is predominantly at an inflection point where gradual reasoning and ”understanding” capabilities of LLM are getting refined for generic intelligence. Thought this advancement,
world models which includes mimicing generic intelligence or super-intelligence on a set of inference
tasks, have fallen short on transferable reasoning and understanding. The chain of thought process
, preceded by human feedback driven reinforcement learning and large vector databases for contextual embeddings are currently insufficient to meet the bar of ”human level intelligence”. This
hands-on session is a deep dive in to the brain of a reasoning module of an LLM - their generating
capabilities, thought process, alignment and coherence of thoughts and meta learning - to have
transferable neural understanding of models.

Read More →