Models & Labs
NVIDIA Releases Nemotron 3.5 ASR for Multilingual Streaming
Sam Witteveenhigh confidence
Why it matters
- →Nemotron 3.5 ASR enhances multilingual streaming capabilities with advanced features like word boosting and diarization.
- →The model's cache-aware and quantized versions improve efficiency, making it suitable for long-running applications.
- →This release strengthens NVIDIA's position in the ASR market, offering developers more powerful tools for audio processing.
NVIDIA has unveiled Nemotron 3.5 ASR, a new automatic speech recognition model designed for live multilingual streaming. This release includes features such as word boosting and speaker diarization, enhancing its application in diverse audio scenarios. The model is also cache-aware and available in quantized versions, which improve its efficiency and performance. This development underscores NVIDIA's commitment to advancing ASR technology, providing developers with more robust tools for building sophisticated audio processing systems.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
