Calculates mean and standard deviation of arr across axis 1. It should exclude values where pad is 1. Args: inputs: A JAX array of shape [b,

Calculates mean and standard deviation of `inputs` across axis 1. It excludes values where `padding` is 1. Args: inputs: A PyTorch tensor of

Merges adapter weights with the original model weights. Args: model (TimesFm): The model to update. adapter_train_state (Tra

mse loss.

Postprocess output of stacked transformer.

Postprocess output of stacked transformer.

Prepare sliding window samples from the time series.

Preprocess input for stacked transformer.

Preprocess input for stacked transformer.

Calculates quantile loss. Args: pred: B x T actual: B x T quantile: quantile at which loss is computed. Returns: per

Calculates quantile loss. Args: pred: Predicted values actual: Actual values quantile: Quantile at which l

Output is of shape [B, N, P, Q].

Output is of shape [B, N, P, Q].

Configure model for distributed training.

Shifts rows of seq based on the first 0 in each row of the mask.

Shifts rows of seq based on the first 0 in each row of the mask. Args: mask: mask tensor of shape [B, N] seq: seq tensor of shape [B, N, P]

_smape loss.

Train for one epoch in a distributed setting. Args: train_loader: DataLoader for training data. optimizer: Optimizer

Uses continuous quantile head.

Perform validation. Args: val_loader: DataLoader for validation data. Returns: Average validation loss.

Build a plausible future with 3 injected anomalies. Injected months: 3, 8, 11 (0-indexed within the 12-month horizon). Returns (future_values

Computes and returns causal mask. Args: input_t: A torch.Tensor of shape [B, T, D]. Returns: An attention_mask torch.Tensor of shape

Check if a dataset will fit in available memory. Args: num_series: Number of time series in the dataset context_length: Length of

Check available disk space for model download.

Check GPU availability and VRAM.

Check Python version >= 3.10.

Check if system has enough RAM.

Clean up the distributed environment.

Attempts to compile the model for fast decoding. See configs.ForecastConfig for more details on the supported flags. Args: forecast_co

Converts binary paddings to a logit mask ready to add to attention matrix. Args: paddings: binary torch.Tensor of shape [B, T], with 1 denoti

Creates target vector and covariate matrices for in context regression. Here we use model fitting language to refer to the context as 'train' and

Creates target vector and covariate matrices for in context regression. Here we use model fitting language to refer to the context as 'train' and

Create a single frame of the animation with fixed axes.

Create a sample DataFrame with time series data. Args: start_date (datetime): Start date of the time series. end_date (datet

2x2 figure -- ALL panels share x-axis = weeks 0-35. (0,0) Sales by store -- context solid, horizon dashed (0,1) Store A: actual vs basel

Auto-regressive decoding without caching. Args: inputs: input time-series and paddings. Time-series shape B x C, padding shape shap

Linear detrend + Z-score anomaly detection on context period. Returns ------- records : list of dicts, one per month trend_line :

Classify each forecast month by which PI band it falls outside. CRITICAL = outside 80% PI (q10-q90) WARNING = outside 60% PI (q20-q80) but i

Eval pipeline.

Forecasts the time series. This is a naive implementation for debugging purposes. No forecasting flags are used here. Forecasting quality can

Forecast all series and return results dict.

Returns the frequency map for the given frequency string.

Generate synthetic retail sales data with covariate components stored separately. Returns a dict with: stores: {store_id: {sales, confi

Extracts adapter parameters from the given model parameters for saving the checkpoint. Args: params (dict): The full model parameter

Get all time covariates.

Get forecasts.

Performs linear interpolation to fill NaN values in a 1D numpy array. Args: arr: The 1D numpy array containing NaN values. Retu

Loads a TimesFM model from a checkpoint.

Load CSV and identify time series columns. Returns: (dataframe, value_column_names, date_column_name_or_none)

Loads a checkpoint and compiles the decoder.

Load and compile the TimesFM model.

Download and prepare the retail sales dataset. This is the same dataset used in the Chronos-2 quickstart notebook and in @kashif's TimesFM 2.

Log metrics to W&B if on the main process. Args: metrics: Dictionary of metrics to log. step: Current training step or ep

Merges 2 masks. logscale mask is expected but 0/1 mask is also fine. Args: a: torch.Tensor of shape [1|B, 1, 1|T, S]. b: torch.Tenso

Example of finetuning TimesFM using multiple GPUs with optimized spawn.

Prepare training and validation datasets from time series data. Args: series: Input time series data context_length: Number

Print a human-readable report to stdout.

Recommend per_core_batch_size based on available resources.

Run the system preflight check and return the report.

Initialize the distributed environment.

Basic example of finetuning TimesFM on stock data.

Removes contiguous NaN values from the beginning of a NumPy array. Args: arr: The input NumPy array. Returns: A new NumPy array with

Generator for validation/test data.

Generator for training data.