Description
NVIDIA's 2-bit 74B vision MoE that fits in 24GB
Stats
1 Download
Capabilities
Minimum system memory
Tags
Description
NVIDIA's 2-bit 74B vision MoE that fits in 24GB
Stats
1 Download
Capabilities
Minimum system memory
Tags
Last updated
Updated 3 hours agobyREADME
NVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB
Vision-language reasoning model from NVIDIA, compressed from Nemotron 3.5 Super VL for local use on RTX GPUs. Contains 74B total parameters with only 9B active at a time, using a hybrid Mamba-2, Mixture-of-Experts and Attention architecture with a C-RADIO v4-H vision encoder. NVIDIA provides it in 2-bit (IQ2) form, so the full model fits in about 24 GB.
Accepts image and text input. Features a reasoning toggle to enable or disable intermediate reasoning traces, and an optional medium reasoning effort mode. Includes native agentic capabilities for tool use and structured output.
Supports a context length of 256K tokens.
Custom Fields
Special features defined by the model author
Enable Thinking
: boolean
(default=true)
Controls whether the model will think before replying
Medium Effort
: boolean
(default=false)
Controls whether the model uses medium reasoning effort
Truncate Thinking History
: boolean
(default=true)
Controls whether thinking history will be truncated to save context space
Parameters
Custom configuration options included with this model
Sources
The underlying model files this model uses
Based on