Skip to content

[AutoDeploy] Detach sharding pattern matcher from pattern executor #5916

@greg-kwasniewski1

Description

@greg-kwasniewski1

Sharding detection (TP, EP, BMM) should create a "sharding config", instead of directly modifying the compute graph, they should parse the graph read-only.

The config should be then processed by a separate pattern executor that apply graph transformations provided.

This way, the autosharding logic could be bypassed/overwritten by manually specifying the sharding config.

Metadata

Metadata

Labels

AutoDeploy<NV> AutoDeploy Backend

Type

Projects

Status

Done

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions