Also check out our Examples to see jiant in action.
If you don't know what to read, why not read our In-Depth Introduction to Jiant?
Contents:
These are quick tutorials that demonstrate jiant usage.
- Quick Start Guide — Using the "Simple" CLI: A simple
jianttraining run in bash, using the "Simple" CLI - Quick Start Guide — Using the "Main" CLI: A simple
jianttraining run in bash, using the "Main" CLI
The "Simple" API provides a single command-line script for training and evaluating models on tasks, while the "Main" API offers more flexibilty by breaking the workflow down into discrete steps (downloading the model, tokenization & caching, writing a fully specific run-configuration, and finally running the experiment). Both interfaces use the same models and task implementations uner the hood.
These are general guides to jiant's design and components. Refer to these if you have questions about parts of jiant:
- In-Depth Introduction to Jiant: Learn about
jiantin greater detail
These are guides to running common NLP benchmarks using jiant:
- GLUE Benchmark: Generate GLUE Benchmark submissions
- SuperGLUE Benchmark: Generate SuperGLUE Benchmark submissions
- XTREME: End-to-end guide for training and generating submission for the XTREME bernchmark
These are more specific guides about running experiments in jiant:
- My Experiment and Me: More info about a
jianttraining/eval run - Tips for Large-scale Experiments
These are notes on the tasks supported in jiant:
- List of supported tasks in
jiant - Task-specific notes: Learn about quirks/caveats about specific tasks
- Adding Tasks: Guide on adding a task to
jiant