facebookresearch/xformers

Provide an easy way to debug mem-eff issues

Open

#632 opened on Jan 10, 2023

View on GitHub
聽(0 comments)聽(0 reactions)聽(0 assignees)Python聽(6,903 stars)聽(473 forks)batch import
good first issue

Description

馃殌 Feature

We should be able to: (1) Understand which op is used for FW/BW (via additional logging enabled with an env variable?) (2) Possibly disable some ops entirely with env variable (like CUDA_VISIBLE_DEVICES but something like XFORMERS_MEMEFF_FW_OPERATORS=...)

Contributor guide

Tech stack
pythonpytorch
Domain
performancemachine 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.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
PythonPyTorchEnvironment variables
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
35
Research direction
Review the xformers codebase, particularly the modules related to memory efficient operations (e.g., xformers/ops/). Examine existing environment variable usage patterns. Consider adding logging to track which ops are used, and a mechanism to disable specific ops via an env variable, similar to CUDA VISIBLE DEVICES. The implementation should be in the core module.
Provide an easy way to debug mem-eff issues 路 facebookresearch/xformers#632 | Good First Issue