NVIDIA Releases Nemotron-3.5-Bolt-E4B-4GB for Low-Memory Local Voice Assistants

·2026.10.24 01:13

Key point

Post-trained from Gemma-4-E4B-it, the model improves multi-turn tool use and instruction following for embedded and edge interfaces.

NVIDIA released Nemotron-3.5-Bolt-E4B-4GB, a small language model post-trained from Gemma-4-E4B-it.
The model is optimized for low-memory and low-TTFT operation, targeting fast local voice, device assistants, and embedded/edge interfaces.
It improves multi-turn tool use, instruction following, and input grounding while preserving headroom for adjacent model pipelines.
Commercial and non-commercial use is permitted, with a companion DFlash drafter available for speculative decoding.
A full-precision BF16 version (NVIDIA-Nemotron-3.5-Bolt-E4B-BF16) and GGUF variants are available on Hugging Face.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.