Latest in AI

NVFP4 Support Merged in llama.cpp: How to Use 4-bit Blackwell Quantization

Following the merge of native NVFP4 (NVIDIA FP4) support in llama.cpp, users are exploring how to leverage this format on Blackwell GPUs (such as the RTX 50-series). The discussion focuses on converting NVFP4 safetensors (like Gemma 4 QAT) to GGUF format and whether importance matrices (imatrix) are required. This enablement promises significant performance gains for local LLM execution on next-gen hardware.