vLLM Adds Audio Frame Support for Nemotron-3.5-Transcribe
November 1, 2026
vLLM now packs valid audio tower frames for MMEncoderAttention to support Nemotron-3.5-Transcribe. This enables efficient multimodal audio processing within the inference engine.
HOW THIS AFFECTS YOU
●
builderYou can now run Nemotron-3.5-Transcribe for multimodal audio tasks using the vLLM runtime.