Cisco Time Series Model Technical Report

TL;DR

Quantitative and qualitative evaluations demonstrate that the Cisco Time Series Model achieves superior performance on observability datasets while retaining very similar performance on a standard general-purpose forecasting benchmark (GIFT-Eval), and suggest that the multiresolution structure enables the model to make more accurate predictions on long context input.

Abstract

We introduce the Cisco Time Series Model, a univariate zero-shot forecaster. This time series foundation model is the result of a general architectural innovation to a time series model enabling it to accept multiresolution input, applied to a popular decoder-only time series model (TimesFM). The resulting multiresolution decoder-only model is trained on over 300B unique data points, with more than half coming from the observability domain. Quantitative and qualitative evaluations demonstrate that the resulting model achieves superior performance on observability datasets while retaining very similar performance on a standard general-purpose forecasting benchmark (GIFT-Eval), and suggest that the multiresolution structure enables the model to make more accurate predictions on long context input.

Figures (13)

• Figure 1: Time series with structure in 1-hour resolution not visible at 1-minute resolution
• Figure 2: Multiresolution time series with padded 1-hour context
• Figure 3: Architecture diagram illustrating Resolution Embeddings and Special Token.
• Figure 4: Statistical deduplication pipeline
• Figure 5: 50 random samples from the largest cluster before deduplication