exolabs/nemotron-3-nano-30b-a3b-mlx-8bit-dequant-bf16

TEXT GENERATIONConcurrent Unit Cost:2Model Size:30BQuant:FP8Context Size:32kPublished:Jun 4, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The exolabs/nemotron-3-nano-30b-a3b-mlx-8bit-dequant-bf16 model is a 30 billion parameter language model, derived from the NVIDIA Nemotron-3-Nano-30B-A3B architecture. This specific version is a dequantized bfloat16 export, originating from an 8-bit quantized MLX checkpoint. It is designed for vLLM validation and represents a conversion rather than the original upstream bfloat16 model. Its primary utility lies in providing a bfloat16 representation for environments requiring this specific data type after an 8-bit quantization process.

Loading preview...

Model Overview

The exolabs/nemotron-3-nano-30b-a3b-mlx-8bit-dequant-bf16 is a 30 billion parameter language model. It is a dequantized bfloat16 export, specifically created for vLLM validation purposes. This model is derived from an 8-bit quantized MLX checkpoint, which itself originated from the nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 model.

Key Characteristics

  • Parameter Count: 30 billion parameters.
  • Data Type: Exported in bfloat16 (Brain Floating Point 16-bit) format.
  • Origin: This is a dequantized version of an 8-bit quantized MLX checkpoint, not the original upstream bfloat16 model.
  • Purpose: Primarily intended for validation within vLLM environments.
  • Architecture: Based on the NVIDIA Nemotron-3-Nano-30B-A3B architecture.

Use Cases

This model is particularly suited for:

  • vLLM Validation: Its main purpose is to facilitate validation processes within vLLM frameworks.
  • Research and Development: Useful for researchers and developers working with dequantized models or exploring the performance characteristics of bfloat16 exports from quantized versions.
  • Specific Deployment Scenarios: Applicable in environments where a bfloat16 representation of a previously 8-bit quantized model is required.