NVIDIA-Nemotron-3.5-Bolt-E4B-BF16
★ agentic★ chat
Overview
The full-precision Bolt-E4B - NVIDIA's 8B (4B effective) edge-assistant model with chat-template thinking control, a 64K context window, and GRPO-tuned tool use. Official BF16 checkpoint for vLLM and SGLang.
Good for
- agentic
- chat
Recommended hardware
≈ 24 GB VRAM to run this 8B model at the BF16 quantization.
Fits 24 GB+ consumer GPUs (e.g. RTX 3090, RTX 4090)