RCreddit.com
Expert expansion with llama.cpp
- Published
- 09/06, 18:27
- Ingested
- 09/06, 23:00
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
A developer has created a custom branch of llama.cpp to support Expert expansion with MOE models, utilizing Glm 5.3 flash. The developer reports that this new version, named "moex-expansion," performs better than their previous DS4 version on metal. They are now seeking feedback from other platforms and with different models to further evaluate its performance and stability.
With the help of Glm 5.3 flash I built a custom branch of llama.cpp in order to support Expert expansion with MOE models, I've tested only on metal and It works better than my DS4 version, i need feedback from other platforms, and different models.