From the archive
Wednesday, 21 October
Folded from 14 sources that day.
Sorted by quality and by how many sources covered it, the same for everyone.
-
How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications
Preview of the VocabX method, which rebuilds a small model's tokenizer around per-token cost instead of frequency. A production Portuguese essay grader on Qwen3-4B cut measured inference cost by 77%.
-
OpenAI models are coming to Google Cloud
OpenAI and Google Cloud will offer frontier OpenAI models and Codex workflows in Gemini's Model Garden, available through Google Cloud Marketplace.
-
Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium
Part 1 of a series on profiling PyTorch inference on AWS Trainium with TorchNeuron. It shows how to configure torch.profiler with the Neuron backend and read key_averages() device times.
-
How we built ChatGPT's novel ad stack for user trust and relevance
OpenAI describes how it built an LLM-based ad system for ChatGPT that reads conversational intent, keeps ads separate from answers, and serves relevant ads at scale.