Skip to content

Feature/gemma 3n e4b support#1

Merged
michalharakal merged 4 commits intodevelopfrom
feature/gemma-3n-e4b-support
Mar 9, 2026
Merged

Feature/gemma 3n e4b support#1
michalharakal merged 4 commits intodevelopfrom
feature/gemma-3n-e4b-support

Conversation

@michalharakal
Copy link
Copy Markdown
Contributor

No description provided.

michalharakal and others added 4 commits March 8, 2026 23:03
Implement the three key E4B architectural components:
- AltUp (Alternating Updates): parallel hidden states with per-layer
  predict/correct and global init/finalize projections
- Gaussian top-k activation sparsity for FFN layers
- Extended weight loading for both GGUF and SafeTensors formats

Also adds Python scripts (uv) for model download and GGUF inspection,
extends config/metadata for E4B fields, and updates the runtime forward
pass to support AltUp when present (E2B path unchanged).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@michalharakal michalharakal merged commit 50c7d95 into develop Mar 9, 2026
2 checks passed
@michalharakal michalharakal deleted the feature/gemma-3n-e4b-support branch March 9, 2026 11:19
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant