MLOps Shenanigans

Functional Programming Bits in Python

Beyond Map & Filter

Feb 2 • Martynas Šubonis

4

2

1

January 2026

InfiniBand and High-Performance Clusters

Fat-tree topologies, credit-based flow control, RDMA, and SHARP reductions

Jan 6 • Martynas Šubonis

5

2

December 2025

MLOps Shenanigans: Wrapping Up 2025

As 2025 is coming to an end, I thought it made sense to close the year with a short note.

Dec 30, 2025 • Martynas Šubonis

5

2

1

November 2025

Advanced, Overlooked Python Typing

There is a common debate in Python circles: if you want static typing, why choose Python to begin with?

Nov 30, 2025 • Martynas Šubonis

35

3

October 2025

Scheduling ML Workloads on Kubernetes

On Gang Scheduling, Bin Packing, Consolidation, and the Like

Oct 23, 2025 • Martynas Šubonis

2

1

July 2025

Dissecting the Model Context Protocol

What the protocol got right and what it still lacks for production use

Jul 7, 2025 • Martynas Šubonis

7

2

April 2025

Zero Temperature Randomness in LLMs

The randomness of LLM outputs is controlled by a parameter known as "temperature." A higher temperature increases randomness, while a lower temperature…

Apr 29, 2025 • Martynas Šubonis

5

2

January 2025

5 Empirical Laws of Software Engineering

What We Observe In Practice

Jan 27, 2025 • Martynas Šubonis

5

Tensor and Fully Sharded Data Parallelism

How Trillion Parameter Models Are Trained

Jan 19, 2025 • Martynas Šubonis

4

December 2024

Model and Pipeline Parallelism

Scaling Training for Large Models

Dec 31, 2024

2

1

Distributed Data Parallel Training

Scaling Training Across Multiple Devices/Machines

Dec 16, 2024 • Martynas Šubonis

5

1

Latency and System Design

Unpacking Different Latencies for Informed Engineering

Dec 2, 2024 • Martynas Šubonis

3

1