# timesfm | Google Research | Knowledge Base | Instagit

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

GitHub Stars: 12.7k

Repository: https://github.com/google-research/timesfm

---

## Articles

### [When Should `infer_is_positive` Be Set to False in TimesFM?](/google-research/timesfm/when-to-set-infer-is-positive-false-timesfm)

Set infer_is_positive=False in TimesFM for time series with legitimate negative values like temperature, financial returns, or signed sensor data.

- Tags: best-practices
- Published: 2026-04-02

### [Understanding the `force_flip_invariance` Parameter in TimesFM](/google-research/timesfm/timesfm-force-flip-invariance-parameter)

Learn about the force flip invariance parameter in TimesFM. Discover how it uses dual input processing to ensure accurate affine transformations for reliable time series forecasting.

- Tags: deep-dive
- Published: 2026-04-02

### [How to Ensure Quantile Monotonicity with TimesFM: Non-Crossing Forecasts Explained](/google-research/timesfm/ensure-quantile-monotonicity-timesfm)

Ensure quantile monotonicity with TimesFM forecasts by enabling fix_quantile_crossing=True. Learn how this post-processing feature prevents non-crossing forecasts for PyTorch and JAX.

- Tags: how-to-guide
- Published: 2026-04-02

### [TimesFM max_context and max_horizon Constraints: Validation Rules and Code Examples](/google-research/timesfm/timesfm-context-horizon-constraints)

Understand TimesFM max_context and max_horizon constraints. Learn validation rules and see code examples for setting these parameters correctly.

- Tags: deep-dive
- Published: 2026-04-02

### [How to Use TimesFM with JAX/Flax: A Complete Inference Guide](/google-research/timesfm/how-to-use-timesfm-jax-flax)

Learn how to use TimesFM with JAX/Flax! This guide covers inference with JAX and Flax models, supporting JIT compilation and multi-device parallelism for efficient time series forecasting.

- Tags: how-to-guide
- Published: 2026-04-02

### [What Is the Maximum Forecast Horizon Supported by TimesFM?](/google-research/timesfm/timesfm-maximum-forecast-horizon)

Discover TimesFM's maximum forecast horizon. Learn how to extend the default 128 time steps for your forecasting needs with Google's TimesFM model. Get the best results.

- Tags: faq
- Published: 2026-04-02

### [How to Estimate RAM Usage for TimesFM: Memory Planning Guide](/google-research/timesfm/estimate-timesfm-ram-usage)

Estimate TimesFM RAM usage with the built-in system check. Calculate memory needs for model weights, input dimensions, and batch size. Plan your memory effectively.

- Tags: how-to-guide
- Published: 2026-04-02

### [TimesFM Memory Requirements: Hardware Specs for TimesFM 1.0 and 2.0](/google-research/timesfm/timesfm-1-0-2-0-memory-requirements)

Discover the memory requirements for TimesFM 1.0 and 2.0. Learn the RAM and VRAM specs needed for optimal performance with TimesFM models. Understand hardware needs for inference.

- Tags: performance
- Published: 2026-04-02

### [How to Load a Locally Saved TimesFM Model: PyTorch and JAX Guide](/google-research/timesfm/load-local-timesfm-model)

Learn to load a locally saved TimesFM model using PyTorch and JAX. Follow our guide to instantiate TimesFmCheckpoint and TimesFm for seamless model integration.

- Tags: how-to-guide
- Published: 2026-04-02

### [Transformer Configuration Details for TimesFM 2.5: Complete Architecture Guide](/google-research/timesfm/timesfm-2-5-transformer-configuration)

Explore the TimesFM 2.5 transformer configuration. Discover its 20-layer architecture, 1280D hidden states, 16 attention heads, RoPE, and bias-free FFNs within the TransformerConfig.

- Tags: architecture
- Published: 2026-04-02

### [What Is REVIN Normalization in TimesFM? A Complete Technical Guide](/google-research/timesfm/what-is-revin-normalization-timesfm)

Explore REVIN normalization in TimesFM, a key technique for advanced time-series forecasting. Learn how this per-instance preprocessing enhances accuracy and restores original scales for reliable predictions. Understand the tec...

- Tags: deep-dive
- Published: 2026-04-02

### [Can TimesFM Handle Time Series with Different Scales? Scale-Equivariance Explained](/google-research/timesfm/timesfm-handle-different-scales)

TimesFM natively handles time series with different scales via scale-equivariance, offering optional normalization for extreme ranges.

- Tags: deep-dive
- Published: 2026-04-02

### [How to Interpret the Quantile Forecast Output from TimesFM](/google-research/timesfm/interpret-timesfm-quantile-forecast)

Learn to interpret Quantile Forecast output from Google Research's TimesFM. Understand mean predictions and quantile levels for your time series analysis.

- Tags: how-to-guide
- Published: 2026-04-02

### [TimesFM XReg Modes: How to Combine Covariates with Foundation Model Forecasts](/google-research/timesfm/timesfm-xreg-modes)

Explore TimesFM XReg modes Combine covariates with foundation model forecasts Learn how xreg timesfm and timesfm xreg modes optimize forecasting accuracy

- Tags: deep-dive
- Published: 2026-04-02

### [How to Use TimesFM with Exogenous Variables (Covariates): A Complete Guide](/google-research/timesfm/timesfm-forecast-with-covariates)

Learn to use TimesFM with exogenous variables. This guide details using the forecast_with_covariates method for dynamic and static covariates, enhancing your forecasts.

- Tags: how-to-guide
- Published: 2026-04-02

### [TimesFM Output Format: Understanding NumPy Arrays and DataFrame Forecasts](/google-research/timesfm/timesfm-forecast-output-format)

Explore the TimesFM output format. Learn how forecasts are returned as NumPy arrays and DataFrame objects, detailing point and quantile forecasts for your time series analysis.

- Tags: api-reference
- Published: 2026-04-02

### [How to Perform Zero-Shot Forecasting with TimesFM: A Complete Implementation Guide](/google-research/timesfm/zero-shot-forecasting-timesfm)

Learn how to perform zero-shot forecasting with TimesFM using our implementation guide. Configure hyperparameters, load checkpoints, and forecast raw time-series data easily with this powerful tool.

- Tags: how-to-guide
- Published: 2026-04-02

### [What Is the Maximum Context Length for TimesFM 2.5? (16,384 Time‑Steps)](/google-research/timesfm/timesfm-2-5-max-context-length)

Discover the maximum context length for TimesFM 2.5, supporting 16,384 time-steps. Learn how this parameter enhances the model's capabilities in the google-research/timesfm repository.

- Tags: getting-started
- Published: 2026-04-02

### [How TimesFM Handles NaN Values in Time Series: The Complete Preprocessing Pipeline](/google-research/timesfm/timesfm-handle-nan-values)

TimesFM preprocesses time series by handling NaN values through infinite conversion, stripping NaNs, linear interpolation, and normalization. Learn its complete pipeline for dense, finite data.

- Tags: how-to-guide
- Published: 2026-04-02

### [Recommended ForecastConfig for TimesFM: Production-Ready Settings Explained](/google-research/timesfm/recommended-timesfm-forecastconfig)

Discover the recommended ForecastConfig for TimesFM production settings. Learn how these parameters optimize memory, speed, and accuracy for your forecasting needs.

- Tags: best-practices
- Published: 2026-04-02

### [How to Compile the TimesFM Model for Forecasting: PyTorch and JAX Guide](/google-research/timesfm/how-to-compile-timesfm-model)

Compile the TimesFM model for forecasting with PyTorch and JAX. Load weights, configure lengths, and use model.compile() for high-performance batched inference.

- Tags: how-to-guide
- Published: 2026-04-02

### [System Requirements for TimesFM 2.5: Hardware Tiers and Memory Configuration](/google-research/timesfm/timesfm-2-5-system-requirements)

Explore TimesFM 2.5 system requirements across four hardware tiers. Understand memory needs from CPU-only to GPU setups for efficient model deployment.

- Tags: system-requirements
- Published: 2026-04-02

### [How to Load the TimesFM 2.5 Model with PyTorch: A Complete Guide](/google-research/timesfm/how-to-load-timesfm-2-5-pytorch)

Learn to load the TimesFM 2.5 model with PyTorch. Import TimesFM_2p5_200M_torch, use from_pretrained to download the checkpoint, and compile for efficient inference.

- Tags: how-to-guide
- Published: 2026-04-02