This model is tracked in our catalogue for reference but is not hosted on TSFM.ai's inference runtime. Its task surface (anomaly detection, classification, imputation, and similarity search) does not fit our forecasting-centric API. To run it, use the Hugging Face repository directly.

Granite-TimeSeries-TSPulse-R1

Tracked, not hosted

View on Hugging Face

ibm-research/granite-timeseries-tspulse-r1

~1M params | Apache-2.0 | Not hosted on TSFM.ai

TSPulse is IBM's ultra-compact time-series understanding model for anomaly detection, classification, imputation, and similarity search. The official card reports +20% on TSB-AD, +25% on semantic similarity search, and +50% in zero-shot imputation while keeping the model around 1M parameters. Tracked on TSFM.ai as part of the Granite TSFM family but not currently hosted on our inference runtime: TSPulse is not a forecasting model, and our API surface is forecasting-centric. To run TSPulse for anomaly detection, classification, or imputation, use the Hugging Face repository directly.

Model Classification

Family

Granite TSPulse

Type

time series foundation model

Pretrained time-series model exposed on TSFM.ai for zero-shot or few-shot forecasting workloads.

Resources

HuggingFace Paper

Training Data

IBM's public card emphasizes multi-domain time-series pretraining for anomaly detection, classification, imputation, and similarity search, but does not publish a more detailed source-by-source corpus breakdown on the card.

Recommended For

• Time-series anomaly detection in production monitoring and industrial settings
• Lightweight multi-task classification, imputation, and retrieval on resource-constrained infrastructure

Strengths

• Purpose-built for anomaly detection with state-of-the-art TSB-AD benchmark results
• Ultra-compact (~1M params) enables CPU-friendly deployment

Limitations

• Not designed for long-horizon forecasting — use a forecasting-first model for that
• Smaller model capacity limits complex multivariate reasoning

Capabilities

anomaly-detectionclassificationimputationretrievalhigh-throughput

Specifications

Parameters: ~1M
Architecture: MLP-Mixer with dual-space (time + frequency) masked reconstruction
Context length: n/a
Max output: n/a
Avg latency: n/a
Uptime: n/a
Plan limits: 1,000 rpm free · 1,000,000 rpm with billing
Accelerator: n/a
Regions: n/a
License: Apache-2.0

Pricing

Input / 1M tokens: n/a
Output / 1M tokens: n/a

Performance

Average latency: n/a
Availability: n/a
Plan limits: 1,000 rpm free · 1,000,000 rpm with billing