[GitHub Trending] google-research/timesfm
8 relevance
Score Breakdown
technical depth 9
novelty 8
actionability 7
community 8
strategic 7
personal 8
Scored daily by a customisable AI persona to surface the most relevant engineering leadership news.
Google's time-series foundation model is novel and strategic for data engineering.
Summary
Google Research released TimesFM 2.5, a 200M-parameter time-series foundation model (down from 500M) supporting up to 16k context length and 1k horizon quantile forecasts via an optional 30M head. The decoder-only model, published at ICML 2024, now includes Flax inference, covariate support via XReg, LoRA fine-tuning through HuggingFace Transformers + PEFT, and an agent skill (SKILL.md) for programmatic orchestration. Available on PyPI with torch or Flax backends, it integrates with BigQuery ML, Google Sheets, and Vertex Model Garden.
Author
google-research