Datasets:

kshitijthakkar
/

large-moe-inference-benchmark

Modalities:

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

README.md exists but content is empty.

Downloads last month: 10

Collection including kshitijthakkar/large-moe-inference-benchmark

Large MoE Architecture Search (1B-2B)

Systematic search for 1B-2B MoE models. Best: bs=1, ctx=2048 achieves 0.32 loss. Top-8 routing beats top-2. • 20 items • Updated Feb 9