Verena is an AI consulting company helping SMBs build and deploy production AI systems, from research to production, without losing either.https://verenalabs.com/https://verenalabs.com/posts/evaluating_muon__moe_in_llms/https://verenalabs.com/posts/evaluating_muon__moe_in_llms/Muon and MoE are becoming standard in frontier LLMs, but neither is plug and play at smaller scale. Part I covers getting Muon to work at 280M parameters and why splitting fused projections is the key.Thu, 26 Mar 2026 12:00:00 GMThttps://verenalabs.com/posts/evaluating-modern-llm-optimizations-on-vit/https://verenalabs.com/posts/evaluating-modern-llm-optimizations-on-vit/This blog recounts my testing on the impact of modern LLM optimizations on ViT-based models, such as SwiGLU, removing bias, and the Muon optimizer.Thu, 19 Feb 2026 00:00:00 GMThttps://verenalabs.com/posts/rec-flow-results-stage-4-5/https://verenalabs.com/posts/rec-flow-results-stage-4-5/Selected Stage 5 rec-flow generations with shared settings and short observations across style categories.Sat, 14 Feb 2026 12:00:00 GMT