Memory Leak
Published in

Memory Leak

Memory Leak — #13

VC Astasia Myers’ perspectives on machine learning, cloud infrastructure, developer tools, open source, and security. Sign up here.

🚀 Products


The simplest, fastest repository for training/finetuning medium-sized GPTs. It is a rewrite of minGPT that prioritizes teeth over education.

Why does this matter? We are seeing two camps of ML practitioners arise: 1) those that train and fine-tune models and 2) those that prompt engineer foundational models. NanoGPT supports operators in the first camp and is a lightweight requiring 2 files of 300 lines. It achieved 8.9K Github stars over the past few weeks.

Source: NanoGPT


Promptable helps ML practitioners streamline their GPT-3 prompt engineering workflow with tools like organizing prompts, tracking changes, evaluating prompts, and deploying prompts.

Why does this matter? Promptable helps ML practitioners in the second camp that prompt engineer foundational models. Prompt engineering is a natural language processing (NLP) concept that involves discovering inputs that yield desirable or useful results from a machine learning model. During our research on prompt engineering we’ve heard that prompt management is critical and will endure even as foundational models improve like the forthcoming GPT-4. Teams prompt test multiple ML APIs to see which one is the best fit for their data and make decisions around which service to use based on the performance/cost curve.


Chris Riccomini open sourced Recap, which makes it easy for engineers to build infrastructure and tools that need metadata. Unlike traditional data catalogs, Recap is designed to power software.

Why does this matter? There have been a wave of data catalogs over the years like Alation, Collibra, Amundsen, among others that focus on the end user and data compliance. Alternatively Recap sits between databases and SaaS interfaces like BI and notebooks and end users don’t interact with it. Machines and software use Recap.

📰 Content

Self-Serve Feature Platforms: Architectures and APIs

Chip Huyen, CEO and co-founder of Claypot, discusses a core component of the MLOps stack, feature platforms. The part discusses the evolution of feature platforms, how they differ from model platforms and feature stores. The second part discusses the core challenges of making feature platforms self-serve for data scientists and increase the iteration speed for feature engineering.

Why does this matter? In machine learning, a feature is a measurable property of the object a data scientist is trying to analyze. Feature engineering is the process of converting raw observations into features that are inputs used during training and inference. Feature platforms are critical to handle both feature retrieval and feature computation. Huyen underscores that there are different types of features including batch, real-time, near real-time that require different infrastructure to ingest data and compute features based on their latency requirements.

Source: Self-Serve Feature Platforms: Architectures and APIs

Managing the Cost of Kubernetes

The New Stack piece discusses the challenges Kubernetes costs and visibility. In the Cloud Native Computing Foundation (CNCF) FinOps for Kubernetes survey, 68% of respondents said Kubernetes costs are rising, and half of that group experience an increase of more than 20% per year.

Why does this matter? Unlike traditional data centers, cloud environments are self-service so developers can spin up new resources with limited oversight. Autoscaling of cloud environments can raise bills, particularly if there aren’t programmatic efforts to stop processes. In the current macroenvironment, efficiency is top of mind and we’ve seen the rise of startups tackling cloud cost management like Kubecost and Vantage. These newer solutions are different than past generations as it provides more visibility to the individual developer rather than just the FinOps and cloud engineering teams.

Introduction to Graph Machine Learning

This Hugging Face piece covers the basics of graph machine learning. It describes what graphs are, why they are used, and how best to represent them. It then covers briefly how people learn on graphs, from pre-neural methods (exploring graph features at the same time) to what are commonly called Graph Neural Networks (GNN). Lastly, it peeks into the world of Transformers for graphs.

Why does this matter? Researchers are exploring use cases for GNNs in computer graphics, cybersecurity, genomics and materials science. We have seen a rise in the interest in GNNs over the past six months.

💼 Jobs

⭐️Dragonfly — Developer Advocate (fully remote)

⭐️ Grit — Founding ML Engineer

⭐️ Omni — Senior Data Visualization Engineer



VC Astasia Myers’ perspectives on machine learning, cloud infrastructure, developer tools, open source and security.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Astasia Myers

Founding Enterprise Partner @ Quiet Capital, previously Investor @ Redpoint Ventures and Cisco Investments