Toto-2.0-313m
onlineDatadog/Toto-2.0-313m313M params | 512 context | $0.00025 per forecast | Apache-2.0
Toto-2.0-313m is the main accuracy-oriented hosted Toto 2.0 checkpoint for observability telemetry and the recommended default of the family. At 313M parameters it sits high on the Toto 2.0 size ladder, pairing the new quantile interface with enough multivariate capacity to handle production monitoring workloads without jumping to a billion-parameter model.
It is a u-muP-scaled decoder-only transformer with alternating time/variate attention and a quantile output head, producing native probabilistic, multivariate forecasts across correlated metric streams. Toto 2.0 continues Datadog's observability-first pretraining line for sparse, high-dimensional telemetry, and Datadog positions the release as the current recommended zero-shot generation for BOOM-style infrastructure metrics.
On TSFM.ai this is the first Toto 2.0 checkpoint to reach for, and the designated replacement for the legacy Datadog/Toto-Open-Base-1.0. Drop down to Datadog/Toto-2.0-22m or Datadog/Toto-2.0-4m when latency, cost, or quick experimentation outweigh accuracy, and step up to Datadog/Toto-2.0-1B only for the highest-value workloads where richer multivariate telemetry structure justifies the higher inference cost.
Model Classification
Family
Toto
Type
time series foundation model
Pretrained time-series model exposed on TSFM.ai for zero-shot or few-shot forecasting workloads.
Resources
Training Data
Toto 2.0 continues Datadog's observability-first pretraining line for sparse, high-dimensional telemetry. Datadog positions the release as the current recommended zero-shot generation for BOOM-style infrastructure metrics.
Recommended For
- • Infrastructure, observability, and telemetry forecasting
- • Sparse, noisy, high-dimensional operational metrics
Strengths
- • Built around real observability-like workloads rather than only clean academic datasets
- • Strong benchmark fit for BOOM-style evaluation
Limitations
- • More specialized than general-purpose forecasting families
- • May be less intuitive as a default pick for simple low-dimensional business series
- • Fine-tuning and exogenous-variable support are planned upstream for Toto 2.0 but are not available in the current release
Capabilities
Tags
Specifications
- Parameters
- 313M
- Architecture
- u-muP-scaled decoder-only transformer with alternating time/variate attention and quantile output head
- Context length
- 512
- Max context
- 4,096
- Minimum history
- 32
- Recommended history
- 512
- Input step
- 32 points
- Required target series
- 1
- Temperature
- Ignored
- Top P
- Ignored
- Max output
- 2,048
- Avg latency
- n/a
- Uptime
- n/a
- Plan limits
- 1,000 rpm free · 1,000,000 rpm with billing
- Accelerator
- L40S
- Regions
- Virginia, US
- License
- Apache-2.0
Pricing
- Per forecast
- $0.00025
Performance
- Average latency
- n/a
- Availability
- n/a
- Plan limits
- 1,000 rpm free · 1,000,000 rpm with billing