NVIDIA open-sources Bolt E4B, and demonstrates it running simultaneously with a vision model on a 8GB device
BIBIBI
AT A GLANCE
NVIDIA says Bolt E4B can run simultaneously with a vision model on a device with 8GB of memory.
Article
NVIDIA open-sources Bolt E4B：8GB memory device to run vision and conversational models simultaneously
Beating AI News: NVIDIA open-sources a small AI Nemotron 3.5 Bolt E4B 4GB, specifically for local voice assistants, game characters, and robots. It is based on continued training of Google Gemma 4 E4B and enhances continuous conversation, instruction understanding, and tool-calling capabilities. The model has approximately 8000000000 total parameters, but many parameters only need to be looked up rather than subjected to complex computations, with a core computational scale of approximately 4000000000 parameters.
NVIDIA also demonstrated the desktop robot Reachy Mini. Cosmos 3 Edge identifies the surrounding environment, while Bolt handles conversation and calls preset programs to control the robot's movements. The two models run simultaneously on a Jetson Orin Nano with only 8GB of memory.
Original link https://m.theblockbeats.info/flash/373572
Key points
01
NVIDIA has open-sourced Nemotron 3.5 Bolt E4B 4GB。
02
The model is designed for local voice assistants, game characters, and robots.
03
The model is based on continued training of Google Gemma 4 E4B and enhances continuous conversation, instruction understanding, and tool-calling capabilities.
04
The model has approximately 8000000000, with a core computational scale of approximately 4000000000 parameters.
05
In the demonstration, Cosmos 3 Edge identifies the environment, while Bolt handles conversation and calls preset programs to control Reachy Mini.
06
The two models run simultaneously on a 8GB-memory Jetson Orin Nano.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This is a small model designed for local devices. The demonstration shows that the vision model and conversational model can work together on the same device with 8GB of memory.
Why it matters to readers
If the actual performance is stable, the lower memory requirements may facilitate deployment of local voice assistants, robots, and game characters.
You can think of it as a small device that simultaneously handles environmental perception, conversation, and action execution.
Risks and unknowns
The original text does not provide links to NVIDIA's official announcement, code, or model weights.
The original text does not provide model license, benchmark, or actual memory-usage data.
The original text does not specify the compatible-device scope or subsequent adoption of the demonstration.
Related Developments
Loading event timeline…
Related concepts
edge AI
This term is not in the glossary yet. Browse related concepts in the glossary.
