spacy-loggers: Logging utilities for spaCy

Starting with spaCy v3.2, alternate loggers are moved into a separate package so that they can be added and updated independently from the core spaCy library.

spacy-loggers currently provides loggers for:

• Weights & Biases
• MLflow
• ClearML
• PyTorch
• CuPy

spacy-loggers also provides additional utility loggers to facilitate interoperation between individual loggers.

If you'd like to add a new logger or logging option, please submit a PR to this repo!

Setup and installation

spacy-loggers should be installed automatically with spaCy v3.2+, so you usually don't need to install it separately. You can install it with pip or from the conda channel conda-forge:

pip install spacy-loggers

conda install -c conda-forge spacy-loggers

Loggers

WandbLogger

Installation

This logger requires wandb to be installed and configured:

pip install wandb
wandb login

Usage

spacy.WandbLogger.v5 is a logger that sends the results of each training step to the dashboard of the Weights & Biases tool. To use this logger, Weights & Biases should be installed, and you should be logged in. The logger will send the full config file to W&B, as well as various system information such as memory utilization, network traffic, disk IO, GPU statistics, etc. This will also include information such as your hostname and operating system, as well as the location of your Python executable.

spacy.WandbLogger.v4 and below automatically call the default console logger. However, starting with spacy.WandbLogger.v5, console logging must be activated through the use of the ChainLogger. This allows the user to configure the console logger's parameters according to their preferences.

Note that by default, the full (interpolated) training config is sent over to the W&B dashboard. If you prefer to exclude certain information such as path names, you can list those fields in "dot notation" in the remove_config_values parameter. These fields will then be removed from the config before uploading, but will otherwise remain in the config file stored on your local system.

Example config

[training.logger]
@loggers = "spacy.WandbLogger.v5"
project_name = "monitor_spacy_training"
remove_config_values = ["paths.train", "paths.dev", "corpora.train.path", "corpora.dev.path"]
log_dataset_dir = "corpus"
model_log_interval = 1000

MLflowLogger

Installation

This logger requires mlflow to be installed and configured:

pip install mlflow

Usage

spacy.MLflowLogger.v2 is a logger that tracks the results of each training step using the MLflow tool. To use this logger, MLflow should be installed. At the beginning of each model training operation, the logger will initialize a new MLflow run and set it as the active run under which metrics and parameters wil be logged. The logger will then log the entire config file as parameters of the active run. After each training step, the following actions are performed:

• The final score is logged under the metric score.
• Individual component scores are logged under their default names.
• Loss values of different components are logged with the loss_ prefix.
• If the final score is higher than the previous best score (for the current run), the model artifact is additionally uploaded to MLflow. This action is only performed if the output_path argument is provided during the training pipeline initialization phase.

By default, the tracking API writes data into files in a local ./mlruns directory.

spacy.MLflowLogger.v1 and below automatically call the default console logger. However, starting with spacy.MLflowLogger.v2, console logging must be activated through the use of the ChainLogger. This allows the user to configure the console logger's parameters according to their preferences.

Note that by default, the full (interpolated) training config is sent over to MLflow. If you prefer to exclude certain information such as path names, you can list those fields in "dot notation" in the remove_config_values parameter. These fields will then be removed from the config before uploading, but will otherwise remain in the config file stored on your local system.

Example config

[training.logger]
@loggers = "spacy.MLflowLogger.v2"
experiment_id = "1"
run_name = "with_fast_alignments"
nested = False
remove_config_values = ["paths.train", "paths.dev", "corpora.train.path", "corpora.dev.path"]

ClearMLLogger

Installation

This logger requires clearml to be installed and configured:

pip install clearml
clearml-init

Usage

spacy.ClearMLLogger.v2 is a logger that tracks the results of each training step using the ClearML tool. To use this logger, ClearML should be installed and you should have initialized (using the command above). The logger will send all the gathered information to your ClearML server, either the hosted free tier or the open source self-hosted server. This logger captures the following information, all of which is visible in the ClearML web UI:

• The full spaCy config file contents.
• Code information such as git repository, commit ID and uncommitted changes.
• Full console output.
• Miscellaneous info such as time, python version and hardware information.
• Output scalars:
  • The final score is logged under the scalar score.
  • Individual component scores are grouped together on one scalar plot (filterable using the web UI).
  • Loss values of different components are logged with the loss_ prefix.

In addition to the above, the following artifacts can also be optionally captured:

• Best model directory (zipped).
• Latest model directory (zipped).
• Dataset used to train.
  • Versioned using ClearML Data and linked to under Configuration -> User Properties on the web UI.

spacy.ClearMLLogger.v1 and below automatically call the default console logger. However, starting with spacy.ClearMLLogger.v2, console logging must be activated through the use of the ChainLogger. This allows the user to configure the console logger's parameters according to their preferences.

Note that by default, the full (interpolated) training config is sent over to ClearML. If you prefer to exclude certain information such as path names, you can list those fields in "dot notation" in the remove_config_values parameter. These fields will then be removed from the config before uploading, but will otherwise remain in the config file stored on your local system.

Example config

[training.logger]
@loggers = "spacy.ClearMLLogger.v2"
project_name = "Hello ClearML!"
task_name = "My spaCy Task"
model_log_interval = 1000
log_best_dir = training/model-best
log_latest_dir = training/model-last
log_dataset_dir = corpus
remove_config_values = ["paths.train", "paths.dev", "corpora.train.path", "corpora.dev.path"]

PyTorchLogger

Installation

This logger requires torch to be installed:

pip install torch

Usage

spacy.PyTorchLogger.v1 is different from the other loggers above in that it does not act as a bridge between spaCy and an external framework. Instead, it is used to query PyTorch-specific metrics and make them available to other loggers. Therefore, it's primarily intended to be used with ChainLogger.

Whenever a logging checkpoint is reached, it queries statistics from the PyTorch backend and stores them in the dictionary passed to it. Downstream loggers can thereafter lookup the statistics and log them to their preferred framework.

The following PyTorch statistics are currently supported:

• CUDA memory statistics

Example config

[training.logger]
@loggers = "spacy.ChainLogger.v1"
logger1 = {"@loggers": "spacy.PyTorchLogger.v1", "prefix": "pytorch", "device": "0", "cuda_mem_metric": "current"}
# Alternatively, you can use any other logger that provides the `log_custom_stats` parameter.
logger2 = {"@loggers": "spacy.LookupLogger.v1", "patterns": ["pytorch"]}

CupyLogger

Installation

This logger requires cupy to be installed:

pip install cupy

Usage

Similar to PyTorchLogger, spacy.CupyLogger.v1 does not act as a bridge between spaCy and an external framework but rather is used with the ChainLogger to facilitate the flow of metrics to other loggers. The CupyLogger queries statistics from the CuPy backend and stores them in the info dictionary passed to it. Downstream loggers can thereafter lookup the statistics and log them to their preferred framework.

The following CuPy statistics are currently supported:

• CUDA memory pool statistics

Example config

[training.logger]
@loggers = "spacy.ChainLogger.v1"
logger1 = {"@loggers": "spacy.CupyLogger.v1", "prefix": "cupy"}
# Alternatively, you can use any other logger that provides the `log_custom_stats` parameter.
logger2 = {"@loggers": "spacy.LookupLogger.v1", "patterns": ["cupy"]}

Utility Loggers

ChainLogger

Usage

This logger can be used to daisy-chain multiple loggers and execute them in-order. Loggers that are executed earlier in the chain can pass information to those that come later by adding it to the dictionary that is passed to them.

Currently, up to 10 loggers can be chained together.

Example config

[training.logger]
@loggers = "spacy.ChainLogger.v1"
logger1 = {"@loggers": "spacy.PyTorchLogger.v1"}
logger2 = {"@loggers": "spacy.ConsoleLogger.v1", "progress_bar": "true"}

LookupLogger

Usage

This logger can be used to lookup statistics in the info dictionary and print them to stdout. It is primarily intended to be used as a tool when developing new loggers.

Example config

[training.logger]
@loggers = "spacy.ChainLogger.v1"
logger1 = {"@loggers": "spacy.PyTorchLogger.v1", "prefix": "pytorch"}
logger2 = {"@loggers": "spacy.LookupLogger.v1", "patterns": ["^[pP]ytorch"]}

Bug reports and other issues

Please use spaCy's issue tracker to report a bug, or open a new thread on the discussion board for any other issue.