🚨 NVIDIA JUST DROPPED A 0.6B CPU-ONLY SPEECH MODEL

@whitewhoadie

32.6K followers · usually 18.7K views a post

Views

100.5K

Likes

4.1K

Comments

60

Shares

564

SoundTikTok sound · original sound - whitewhoadie

Promoted product

Nemotron-3.5-ASR

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Why it worked

The video leverages the novelty of a major tech company like NVIDIA releasing a powerful, yet lightweight and accessible, AI model that can run on consumer hardware. The clear demonstration of performance and the direct link to the model's repository make it highly valuable for developers and AI enthusiasts.

Summary

The video announces NVIDIA's new 0.6 billion parameter, CPU-only speech recognition model called Nemotron-3.5-ASR. It highlights the model's capabilities, including real-time streaming, support for over 40 languages, and its ability to run entirely on CPU.

Structure

1Introduction of NVIDIA's new speech model
2Demonstration of CPU vs GPU performance
3Introduction of Nemotron-3.5-ASR
4Highlighting key features: parameter count, language support, real-time streaming

Original caption

🚨 NVIDIA JUST DROPPED A 0.6B CPU-ONLY SPEECH RECOGNITION MODEL Nemotron-3.5-ASR delivers real-time streaming ASR across 40+ languages — and it runs entirely on CPU. Ships with: • Only 0.6B parameters • Real-time streaming output • 2.5x faster than official Nemo runtime (same accuracy) • Fully offline capable • Easy integration into local agent pipelines Another strong small model for on-device/local AI stacks. 👉 https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b Who’s running local speech recognition today? Drop a 🔥

More from @whitewhoadie

Product placement

NVIDIA NeMo (PyTorch) ACTS - it is a speech recognition model that is being demonstrated and compared. Removing it would change the video's content. NVIDIA ACTS - it is the company that dropped the speech model. Removing it would change the video's content. Nemotron-3.5-ASR ACTS - it is the speech recognition model being discussed and demonstrated. Removing it would change the video's content. LocalAI ACTS - it is a team/project that is associated with the model and its integration. Removing it would change the video's content.

Call to action

https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b

On-screen text

🚨 NVIDIA JUST DROPPED A 0.6B CPU-ONLY SPEECH MODEL parakeet.cpp (q8_0) vs NVIDIA CPU, no GPU identical output Well, I don't wish to see it 24 ms NVIDIA NeMo (PyTorch) CPU, no GPU Well, I don't wish to 82 ms WHAT'S UP, GUYS? https://huggingface.co/ nvidia/nemotron-3.5-asr- streaming-0.6b SO NVIDIA DROPPED Well, I don't wish to see it any more, observed Phoebe, turning away her eyes. It is certainly very like the old portrait. 138 ms 57% NVIDIA NeMo (PyTorch) CPU, no GPU Well, I don't wish to 192 ms 88% A SPEECH RECOGNITION https://huggingface.co/ nvidia/nemotron-3.5-asr- streaming-0.6b MODEL WITH ONLY 241 ms 31x realtime fastest 100% NVIDIA NeMo (PyTorch) CPU, no GPU Well, I don't wish to see it any more, observed Pho away her eyes. It is certainly 303 ms https://huggingface.co/ nvidia/nemotron-3.5-asr- streaming-0.6b 0.6 BILLION PARAMETERS. OKAY? IT'S CALLED https://huggingface.co/ nvidia/nemotron-3.5-asr- streaming-0.6b NEMOTRON 3.5 ASR https://huggingface.co/ nvidia/nemotron-3.5-asr- streaming-0.6b IT SUPPORTS OVER SAME MODEL - SAME CPU - IDENTICAL OUTPUT parakeet.cpp (q8_0) NVIDIA NeMo (PyTorch) 60B 3x faster on the same CPU, no GPU github.com/mudler/parakeet.cpp 241 ms 12x https://huggingface.co/ nvidia/nemotron-3.5-asr- streaming-0.6b 40 LANGUAGES, LOCALAI from the LocalAI team github.com/mudler/parakeet.cpp huggingface.co/mudler/parakeet.cpp AND IT HAS REAL TIME STREAMING OUT.

Transcript

What's up, guys? So NVIDIA dropped a speech recognition model with only 0.6 billion parameters. Okay? It's called Nemotron 3.5 ASR. It supports over 40 languages, and it has real-time streaming out.