OpenEuroLLM / OpenEuroLLM/Taskboard

Hybrid Model Tests

Open
#363 1 comment 0 reactions 1 assignee View on GitHub

@kpoeppel is already working on this.

Since Aug 4, 2026.

T4.4 - model training
Dominant language
No language data
Stars
3
Forks
0
PR merge metrics
No merged PRs in 30d

Description

Goal
Testing Hybrid model architectures for scaling in comparison to FullAttn

Description
Test Mamba2, GDN, SWA, mLSTM hybrids compared to FullAttn on multilingual scaling best (bs, lr) points

Deliverable scope
Provide a comparison on different scales showing if we should switch to a hybrid arch.
Preliminary data:

Image

At the moment the speed is: SWA > mLSTM, FullAttn, > Mamba2 > GDN. In terms of GPU-hrs based best performance Mamba2 still seems to work best.

Dependencies
none

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.