Skip to content

Work About Skills Writing Contact

Have something in mind? Let's build it.

Open to full-stack and AI/ML work — freelance, contract or full-time. Email is the fastest way to reach me.

Based in New Delhi, India. Building for the modern web.

Pages

Work
About
Skills
Writing
Contact

Elsewhere

GitHub
LinkedIn
X
RSS feed

© 2026 Aman Kushwaha

Set in Syne, Instrument Sans & Geist Mono

Boosting MoE Training Throughput with Advanced Fusion Kernels

Learn how advanced fusion kernels dramatically boost MoE training throughput on GPU clusters.

Aman KushwahaJune 16, 20262 min read

MoE
training
fusion kernels
GPU
AI
NVIDIA

Loading article content...

Did this land?

Discussion (0)

No comments yet — be the first to share your thoughts.

Keep reading

Building Deploy Detective — An Approval-Gated Incident Response Agent on TrueForge
Engineering
Adaptive Parallel Reasoning: Scaling LLM Inference with Dynamic Parallelism
Beyond PEFT: Why LoRA May Not Be the Best Fine‑Tuning Choice