facebookresearch/fairseq

Add tokens to extend embeddings during fine-tuning

Open

#2,632 opened on Sep 18, 2020

View on GitHub
 (13 comments) (0 reactions) (0 assignees)Python (6,224 forks)batch import
enhancementhelp wantedneeds triage

Repository metrics

Stars
 (29,107 stars)
PR merge metrics
 (No merged PRs in 30d)

Description

🚀 Feature Request

Like add_tokens() in huggingface/transformers (link), this will allow fine-tuning add new embeddings of tokens, which will not be sperated.

Motivation

Some NLG tasks may have non-terminals, which are expected not to be separated by subwords.

Contributor guide

Research direction
Study the Hugging Face 'add tokens' implementation in transformers library, then identify where tokenization and embedding are handled in fairseq (e.g., in Dictionary and TransformerModel). Propose a method to add new tokens and extend the embedding matrix accordingly.
Tech stack
pythonpytorch
Domain
machine learning
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PythonPyTorchTokenizers
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40