llama.cpp Release b9670 Fixes NVFP4 Edge Cases and LORA Dequantization

Core summary

One sentence to understand this update

llama.cpp has released version b9670, addressing NVFP4 edge cases in llama-graph and implementing necessary post-GEMM MUL for LORA dequantization and bias addition.

Impact & opportunity

What this could mean

Builders using llama.cpp will benefit from improved stability and accuracy, especially when working with NVFP4 and LORA models, enabling more reliable local deployments.