facebookresearch/fairseq

Is there any plans to support torch.export?

Open

#5,519 opened on Jul 5, 2024

View on GitHub
 (0 comments) (0 reactions) (0 assignees)Python (29,107 stars) (6,224 forks)batch import
enhancementhelp wantedneeds triage

Description

Hello.

Is there any plans to support torch.export? It would be awesome for developers who wants compilation of fairseq for backends or their deep learning accelerators.

The code would be like below.

from fairseq.models import FairseqEncoderDecoderModel
import torch

translator= FairseqEncoderDecoderModel.from_pretrained(
    # ..
)
example_inputs = translator.dummy_inputs

with torch.no_grad():
    ep = torch.export.export(translator_model, example_inputs)

Contributor guide

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.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
over 1 week
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
unclear
Prerequisites
PythonPyTorchfairseq model internalstorch.export API
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20
Research direction
Investigate whether fairseq currently has any export mechanism (e.g., ONNX, TorchScript) and evaluate compatibility with torch.export. Review the FairseqEncoderDecoderModel class in fairseq/models to understand the model architecture and dummy inputs method. Check the PyTorch torch.export documentation for constraints on dynamic shapes and control flow. Search the fairseq repository for any open PRs or discussions related to export or compilation. Provide a concrete proposal for integrating torch.export, noting potential challenges with fairseq's sequence to sequence models.