/blog /logs /tags /projects /research /talks /about /contact

hi, i'm shashank shekhar

I do machine learning research and engineering. This website is my place to talk about about ML, startups, interesting projects, writing, and learning. Speaking of which, you can learn more about me here.

Subscribe to get blog posts and a weekly log digest

Latest

On human (mostly mine) and machine learning.

/bloglong-form

Published on
March 17, 2026·25 min read
Energy-Based Models: From Basics to LLMs
llm energy-based-model likelihood score-matching diffusion
A whirlwind tour of energy-based models (EBMs) paradigms, from basics to modern LLM applications. Based on my talk at the Toronto LLM Meetup.
Published on
February 9, 2026·40 min read
Understanding DeepSeek's Multi-Head Latent Attention (MLA)
llm attention transformers deepseek mla kv-cache inference
On bottlenecks in attention, kv caching, long-context decoding, attention variants, and how DeepSeek MLA came to be. Part 1 of the FlashMLA blog series.
Published on
May 23, 2025·35 min read
Data Quality Is All You Need?
llm pretraining midtraining posttraining data-quality synthetic-data dpo
Notes on Microsoft phi-4 data pipeline for pre-training, 'mid-training', supervised fine-tuning and preference optimization

All Blogs →

/logsbyte-sized

Published on
March 25, 2026·7 min read
Stop using torch.cat for your KV cache implementations
llm kv-cache pytorch inference optimization transformers
tl;dr: `torch.cat` is not in-place, instead use pre-allocated buffers

All dev logs →

hi, i'm shashank shekhar

Latest

/bloglong-form

Energy-Based Models: From Basics to LLMs

Understanding DeepSeek's Multi-Head Latent Attention (MLA)

Data Quality Is All You Need?

/logsbyte-sized

Stop using torch.cat for your KV cache implementations