Skip to content

Qwen3-235B-A22B OOMs with sufficient amount of VRAM #2886

@krammnic

Description

@krammnic

We're encountering an out-of-memory (OOM) error on a 2TB machine when attempting to load Qwen3-235B-A22B.

The issue seems to be that torchtune loads the full, unsharded checkpoint with self._checkpoint_client.load_base_checkpoint(). For a bf16 model of this size, the complete weights (~3.7TB) exceed the available memory. Loading the model in shards would likely resolve this.

Has anyone else encountered this?

Originally posted by @leng-yue in #2867 (comment)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions