--- title: "Latest reads" description: "The engineering internet, summarised so you can actually read it." --- # Latest reads > The engineering internet, summarised so you can actually read it. ## Articles ### [Supabase Beta November 2021: Launch Week Recap](https://yomu.fyi/post/supabase-beta-november-2021-launch-week-recap.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Ant Wilson - Published: Dec 15, 2021 Supabase celebrated its third Launch Week and one year in beta by rolling out several major infrastructure and tooling updates. The company open-sourced its Dashboard for self-hosting and integrated Logflare to provide searchable database and API logs directly in the interface. For database querying and APIs, Supabase introduced a Postgres extension that resolves each GraphQL request with a single SQL statement to minimize network IO overhead, while upgrading its default hosted database to PostgreSQL 14 and deploying PostgREST 9.0. Security capabilities were expanded by enabling Row Level Security policies across the Realtime API to restrict streams and subscriptions on a per-user basis. Additional releases during the week included faster media file delivery, an open-sourced launch methodology, and a free course combining Next.js, Stripe, and Supabase. ### [Perceiver IO: a scalable, fully-attentional model that works on any modality](https://yomu.fyi/post/perceiver-io-a-scalable-fully-attentional-model-that-works-on-any-moda.md) - Company: huggingface.co - Author: Niels Rogge - Published: Dec 15, 2021 Standard Transformer architectures scale poorly in compute and memory because pairwise dot-product self-attention depends quadratically on input size. Perceiver IO addresses this constraint by computing self-attention across a small set of latent variables rather than directly on high-dimensional inputs. Inputs and outputs interact with the model via cross-attention operations, decoupling compute and memory costs from input and output dimensions. Integrated into Hugging Face Transformers via the PerceiverModel class, the architecture supports diverse data types using optional preprocessors, decoders, and postprocessors. Experiments demonstrate competitive performance across text, multimodal video classification, 3D point cloud classification on ModelNet40, and StarCraft II reinforcement learning in AlphaStar. ### [Training CodeParrot 🦜 from Scratch](https://yomu.fyi/post/training-codeparrot-from-scratch.md) - Company: huggingface.co - Author: Leandro von Werra - Published: Dec 8, 2021 Training code generation models from scratch requires careful data filtering, tokenization, and compute management. To build CodeParrot, a 1.5-billion-parameter Python generation model, engineers extracted 20 million Python files from BigQuery and cleaned the resulting 180 GB corpus down to 50 GB after discovering extreme duplication. A custom GPT-2 tokenizer was trained over streamed samples, and the GPT-2 large architecture was initialized with layer-scaled, full-precision attention mechanisms. The training pipeline leveraged Hugging Face Accelerate alongside gradient checkpointing and a streaming iterable dataset designed to yield fixed-length concatenated token sequences. CodeParrot was trained on roughly 25 to 30 billion tokens and evaluated on coding tasks using the pass@k metric. ### [Five more things](https://yomu.fyi/post/five-more-things.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Ant Wilson - Published: Dec 3, 2021 Supabase announced a series of platform updates, technical releases, and community initiatives to conclude its Launch Week. The team open-sourced pg\_graphql, an in-progress PostgreSQL extension that keeps schema generation, query parsing, and resolvers natively on the database server. Supabase also enabled a global CDN spanning over 250 cities, delivering edge caching with a default one-hour cache time for storage assets alongside TLS 1.3, HTTP/3 over QUIC, and Brotli compression. While Supabase Functions is not yet released, two public RFCs were opened to explore architecture options, leaning heavily toward container-based execution. Finally, the update introduced a free SaaS development course with egghead.io and launched the ten-day Holiday Hackdays hackathon. ### [pg\_graphql: A GraphQL extension for PostgreSQL](https://yomu.fyi/post/pg-graphql-a-graphql-extension-for-postgresql.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Oliver Rice - Published: Dec 3, 2021 Supabase has open-sourced pg\_graphql, a native PostgreSQL extension written in C and SQL that enables in-database GraphQL schema reflection and query resolution. Because free-tier virtual machines lacked memory headroom for standalone GraphQL processes like Hasura or Graphile, the team designed an in-engine solution to eliminate extra runtime overhead. The extension parses, validates, and transpiles incoming GraphQL operations into single SQL statements that aggregate data into JSON responses. This architecture avoids N+1 query problems, enforces existing row-level security policies, and maps GraphQL requests directly onto PostgreSQL transactions. The public SQL function can be queried directly or exposed over HTTP via PostgREST remote procedure calls. ### [Kicking off the Holiday Hackdays](https://yomu.fyi/post/kicking-off-the-holiday-hackdays.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Thor Schaeff - Published: Dec 3, 2021 Supabase has launched Holiday Hackdays, a ten-day virtual hackathon running from December 3rd to December 12th. The event invites participants to build open-source projects using Supabase features, following the recent introduction of row-level security for the Realtime API. Teams of one to five developers may build applications using any programming language or framework provided the code and design elements are created during the hackathon. Submissions will be evaluated on criteria such as inventiveness, technical execution, visual polish, and Supabase integration, with prizes dedicated to Realtime API usage and holiday-themed concepts. Members of the Supabase team will also take part by developing an open-source advent calendar showcasing community content. ### [Supabase acquires Logflare](https://yomu.fyi/post/supabase-acquires-logflare.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: Dec 2, 2021 Supabase announced the acquisition of Logflare, a realtime log ingestion and analytics platform. Modern database architectures encompass APIs, authentication, storage, and CDNs, requiring unified observability to diagnose slow queries and trace user requests effectively. Built with Elixir—the same language powering Supabase's Realtime Engine—Logflare automates schema adaptation for streaming inserts and enables developers to construct APIs from SQL queries via Logflare Endpoints. The integration roadmap includes embedding log querying with SQL directly into the Supabase Dashboard, maintaining existing integrations for Vercel and Cloudflare, and transitioning Logflare toward a fully open-source implementation with multiple backend storage engines. Over the coming months, existing Logflare users will be migrated to the Supabase platform under matching pricing tiers. ### [Introducing Snowball Fight ☃️, our first ML-Agents environment](https://yomu.fyi/post/introducing-snowball-fight-our-first-ml-agents-environment.md) - Company: huggingface.co - Author: Thomas Simonini - Published: Dec 2, 2021 Hugging Face has released Snowball Fight 1vs1, its first custom Deep Reinforcement Learning environment created using the Unity Machine Learning Agents Toolkit. The game enables players to shoot snowballs against a trained agent and is hosted on Hugging Face Spaces. This release marks the beginning of an effort to build an ecosystem for Deep Reinforcement Learning researchers and enthusiasts using Unity ML-Agents. The initiative focuses on building open-source environments, hosting models and training environments on the Hugging Face Hub, and deploying interactive demos via Spaces. Future plans include publishing technical tutorials, releasing additional environments, and developing a Snowball Fight 2vs2 version that utilizes the MA-POCA algorithm to train cooperative team behaviors. ### [Realtime Postgres RLS now available on Supabase](https://yomu.fyi/post/realtime-postgres-rls-now-available-on-supabase.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Oliver Rice - Published: Dec 1, 2021 Supabase updated its Realtime server to enforce PostgreSQL Row Level Security (RLS) policies when broadcasting database changes over websockets. Previously, Realtime operated as an opt-in beta feature that sent all replication changes to every client regardless of user authorization. To enforce RLS per subscriber without heavy performance overhead, Supabase introduced WALRUS, a security engine colocated inside PostgreSQL. For each replication change, WALRUS looks up active subscribers, assumes their identities, and evaluates row visibility using prepared statements queried by primary key. This in-database evaluation avoids external network round trips and single-query planning overhead while returning an authorized subscriber list to Realtime. ### [Supabase Studio](https://yomu.fyi/post/supabase-studio.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: Nov 30, 2021 Supabase has released Supabase Studio, making the web dashboard used on its hosted platform available for local development and self-hosted environments. The dashboard originally existed within a single internal monorepo alongside cloud infrastructure, but managing split repositories and code synchronization significantly slowed down production release cycles. To streamline development, the team unified the frontend into a single open-source codebase modeled after Sentry's shared architecture strategy. Built with Next.js, Tailwind CSS, Supabase UI, and MobX, the application works directly with existing deployments across Docker and the Supabase CLI. The open-source release focuses strictly on database management features, including SQL editors, role policies, and extensions, while excluding cloud project administration. ### [Getting Started with Hugging Face Transformers for IPUs with Optimum](https://yomu.fyi/post/getting-started-with-hugging-face-transformers-for-ipus-with-optimum.md) - Company: huggingface.co - Author: Tim Santos, Julien Simon - Published: Nov 30, 2021 Transformer prediction speed can hinder latency-sensitive workloads like conversational systems and search, while optimizing hardware performance requires significant effort. Hugging Face introduced Optimum, an open-source library that reduces Transformer model latency across diverse hardware targets. Through a partnership with Graphcore, BERT is introduced as the initial IPU-optimized model for parallel AI hardware. Developers configure the Poplar SDK environment and PopTorch inside Graphcloud, install optimum-graphcore, and execute the run\_qa.py script with an IPU configuration file. Fine-tuning bert-base-uncased on SQuAD v1.1 using the IPUTrainer class yields an evaluation F1 score of 88.2757 and an exact match score of 80.6623 across 10,784 evaluation samples. ### [Community Day](https://yomu.fyi/post/community-day.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Thor Schaeff - Published: Nov 29, 2021 Supabase kicked off its Launch Week by showcasing integrations, ecosystem updates, and community contributions across its open-source stack. The platform announced that all hosted projects are upgrading to PostgREST 9, introducing inner joins, functions with unnamed parameters, and PostgreSQL 14 compatibility. Through community contributions, Supabase expanded authentication options to include Slack, Spotify, and MessageBird alongside existing providers. Updates to the Python ecosystem include package renaming to supabase, synchronous operation support in postgrest-py, and a rewritten gotrue-py matching JavaScript feature parity. Additionally, Supabase partnered with GitGuardian to automate secret detection and remediation across development lifecycles, protecting sensitive database service role keys. ### [Introducing the Data Measurements Tool: an Interactive Tool for Looking at Datasets](https://yomu.fyi/post/introducing-the-data-measurements-tool-an-interactive-tool-for-looking.md) - Company: huggingface.co - Author: Sasha Luccioni, Yacine Jernite, Margaret Mitchell - Published: Nov 29, 2021 Hugging Face released an open-source Python library and no-code interface named the Data Measurements Tool to calculate metrics for responsible dataset curation. In AI development, scraped data often enters training pipelines without detailed analysis of representation or downstream model influence, partly because querying raw datasets typically requires complex coding. Built using Hugging Face Dataset and Spaces Hubs alongside Streamlit, the tool provides descriptive, distributional, and comparison statistics. It computes missing values, vocabulary sizes, duplicate counts, label distributions, and Zipf's law adherence to identify unnatural language patterns or markup artifacts. Additionally, it offers embedding clusters via a Sentence-Transformer model and normalized pointwise mutual information metrics across gender and sexual orientation identity terms. ### [New in PostgreSQL 14: What every developer should know](https://yomu.fyi/post/new-in-postgresql-14-what-every-developer-should-know.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Gurjeet Singh - Published: Nov 28, 2021 PostgreSQL 14 introduces several enhancements to improve application performance, data manipulation, query troubleshooting, and database security. Developers can traverse nested JSON data structures using subscripting and store noncontiguous values using multirange data types in single operations. Applications relying on libpq can leverage client-side query pipelining to send multiple queries across high-latency networks without waiting for individual responses, while parallel execution now supports foreign data wrappers, parallel sequential scans, and materialized view refreshes. For observability, query IDs are integrated into live execution tracking via pg\_stat\_activity, EXPLAIN VERBOSE outputs, and log files. Security upgrades include predefined read-only roles such as pg\_read\_all\_data alongside SCRAM-SHA-256 enabled as the default authentication mechanism. ### [PostgREST 9](https://yomu.fyi/post/postgrest-9.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Steve Chavez - Published: Nov 27, 2021 PostgREST 9 introduces several core updates to the tool that turns PostgreSQL databases into RESTful APIs. Developers can now perform inner joins when embedding tables using the !inner keyword, allowing top-level table rows to be filtered directly by embedded table attributes across multiple operators. The release also adds support for POST requests to functions containing a single unnamed parameter, which simplifies handling raw JSON payloads sent by external webhooks. To ensure compatibility with PostgreSQL 14, custom authentication functions accessing HTTP context headers and JWT claims must be updated to parse JSON properties from consolidated settings. Additional enhancements included in this version provide support for partitioned tables alongside documentation improvements and bug fixes. ### [How we launch at Supabase](https://yomu.fyi/post/how-we-launch-at-supabase.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Ant Wilson - Published: Nov 26, 2021 Following their participation in YCombinator, Supabase sought a structured method to maintain product momentum and recreate the urgency of an accelerator internally. The team established Launch Week, a quarterly shipping sprint using a fixed-timeline and flexible-scope model where a major feature or announcement releases each day for a week. Each cycle begins with a broad distributed planning session that maps high-level goals and generates ideas, followed by execution and finely scheduled launch days covering coordinated social media, blog posts, and community engagement. Following the launch, team members transition into maintenance mode and downtime to handle support, bug fixes, and open-source contributions. Finally, a structured retrospective evaluates successes, failures, and process relevance before carrying lessons into the next planning meeting. ### [Supabase Launch Week III: Holiday Special](https://yomu.fyi/post/supabase-launch-week-iii-holiday-special.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: Nov 26, 2021 Supabase organized Launch Week III: Holiday Special from November 29 to December 3, 2021, delivering daily product releases, tooling improvements, and ecosystem announcements. The week began with Community Day to showcase external contributions and open source tooling utilized by the team. On Tuesday, Supabase released Supabase Studio, allowing engineers to utilize the official platform management dashboard in self-hosted configurations and local development workflows. Midweek releases included the rollout of Realtime Row Level Security on Wednesday, followed immediately by Thursday's announcement detailing the acquisition of Logflare. The five-day launch series concluded on Friday with a final batch of five distinct feature announcements across the Supabase platform. ### [Using real-world patterns to improve matching in theory and practice](https://yomu.fyi/post/using-real-world-patterns-to-improve-matching-in-theory-and-practice.md) - Company: [Grab](https://yomu.fyi/company/grab.md) - Author: Tenindra Abeywickrama - Published: Nov 22, 2021 Continuous ride-hailing assignment relies on solving the minimum weight bipartite matching problem between passengers and driver-partners. While traditional implementations assume a precalculated cost matrix, computing shortest-path travel times across large road networks dominates total execution time. Researchers introduced an Incremental Kuhn-Munkres algorithm that leverages the spatial locality of optimal matches to compute edge costs on demand. The approach integrates priority queues and lower-bounding techniques with refinement rules to avoid evaluating distant pairs while guaranteeing the same optimal assignment. Evaluated on Singapore road network data and real Grab production workloads, the incremental techniques reduced exact cost calculations and decreased assignment running times by over an order of magnitude. ### [Accelerating PyTorch distributed fine-tuning with Intel technologies](https://yomu.fyi/post/accelerating-pytorch-distributed-fine-tuning-with-intel-technologies.md) - Company: huggingface.co - Author: Julien Simon - Published: Nov 19, 2021 Transfer learning makes CPU-based clusters a viable, cost-effective option for shorter deep learning fine-tuning jobs. To accelerate PyTorch workloads, clusters can leverage Intel Ice Lake architecture featuring AVX-512 and Vector Neural Network Instructions alongside specialized software optimizations. Implementing the Intel Extension for PyTorch and the oneAPI Collective Communications Library (oneCCL) addresses communication bottlenecks across distributed nodes. In demonstrated tests fine-tuning transformer models on GLUE tasks across Amazon EC2 c6i.16xlarge instances, two-node jobs delivered roughly a 1.7x speedup, while four-node setups achieved a 3x speedup. These performance libraries and multi-node configurations maintain consistent scaling across both smaller datasets like MRPC and larger benchmarks like QQP. ### [Fine-Tune XLSR-Wav2Vec2 for low-resource ASR with 🤗 Transformers](https://yomu.fyi/post/fine-tune-xlsr-wav2vec2-for-low-resource-asr-with-transformers.md) - Company: huggingface.co - Author: Patrick von Platen - Published: Nov 15, 2021 Fine-tuning the multilingual XLS-R speech model adapts pretrained cross-lingual audio representations to automatic speech recognition tasks with limited labeled data. The demonstration configures a Wav2Vec2-XLS-R-300M checkpoint using Connectionist Temporal Classification on roughly four hours of validated Turkish audio from Common Voice. Building the pipeline requires pairing a Wav2Vec2FeatureExtractor for audio signals with a custom Wav2Vec2CTCTokenizer derived from dataset transcriptions. Training with Hugging Face Trainer over 3,200 steps decreases the validation word error rate from 0.7000 down to 0.3195. While the resulting transcription demonstrates recognizable phonetic alignment, output quality can be further improved by extending training schedules, refining preprocessing, and adding language model decoding. [Newer posts](https://yomu.fyi/page/61.md) · [Older posts](https://yomu.fyi/page/63.md)