Skip to content

llama 3.1 8B OOM with nightly build 02272025 #127

@zpcore

Description

@zpcore

We notice llama 3.1 8B OOM issue in 02272025 PTXLA nightly image. The issue doesn't happen on PTXLA r2.6 release.

Model/exp config: seq_len: 8k, batch: 4 batch/device, minibatch on.

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions