Skip to content

[FEA]: Update heuristics for core retriever pipeline using the vl embedder #1774

@randerzander

Description

@randerzander

Is this a new feature, an improvement, or a change to existing functionality?

New Feature

How would you describe the priority of this feature request

Currently preventing usage

Please provide a clear description of problem this feature solves

The nemo_retriever pipeline has builtin heuristics for how many actors and what GPU fractions should be used.

The heuristic logic should be updated to better handle A10G and other low vram SKUs such that the core retriever pipeline works out of the box without requiring manual actor/GPU fraction specification.

Describe the feature, and optionally a solution or implementation and any alternatives

N/A

Additional context

No response

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions