Skip to content

Command Line Interface

Documentation for the command line interface of SEB.

CLI

run

Runs the Benchmark either on specified models or on all registered models. Can save the benchmark's results, but also displays them in a table similar to the official website.

Examples: To run all models on all languages and tasks:

{bash} seb run

To run a model on all languages and tasks:

seb run -m sentence-transformers/all-MiniLM-L6-v2

To run multiple models: To run a model on all languages and tasks:

seb run -m sentence-transformers/all-MiniLM-L6-v2,sentence-transformers/all-mpnet-base-v2

if you only want to limit it to a subset of languages or tasks you can use the --languages and --tasks flags.

# Running a model on a subset of languages 
seb run sentence-transformers/all-MiniLM-L6-v2 -o results/ -l nb,nn 
# Running a model on a subset of tasks 
seb run sentence-transformers/all-MiniLM-L6-v2 -o results/ -t DKHate,ScaLA
Argument Type Description Default
--models, -m Optional[list[str], NoneType] Model names or paths. If a model is not registrered in SEB it will be loaded using SentenceTransformers. If none are specified the whole benchmark is run. None
--output-path, -o Path Directory to save all results to. None
--languages, -l Optional[list[str], NoneType] What languages subsection to run the benchmark on. If left blank it will run it on all languages. None
--tasks, -t Optional[list[str], NoneType] What tasks should model be run on. Default to all tasks within the specified languages. None
--ignore-cache bool Ignores caches models. Note that SEB ships with an existing cache. You can set the cache_dir using the environmental variable SEB_CACHE_DIR False
--ignore-errors bool Should errors be ignored when running a model on a benchmark task. False
--code, -c Path Code to run before executing benchmark. Useful for adding custom model to registries. None
--logging-level str Logging level for the benchmark. 'INFO'