NVIDIA-Nemotron-3-Nano-4B-oQ4

This model was quantized using oQ mixed-precision quantization.

Quantization details

• Model type: nemotron_h
• Bits: 4
• Group size: 64
• Format: MLX safetensors

Benchmark

Model	File size	MMLU	JMMLU	HELLASWAG	GSM8K	ARC_CHALLENGE
NVIDIA-Nemotron-3-Nano-4B-oQ3.5	1.86 GB	53.7%	45.3%	66.0%	79.3%	74.0%
NVIDIA-Nemotron-3-Nano-4B-oQ4	2.19 GB	61.0%	51.7%	71.7%	81.7%	81.3%

Detail

Model	Benchmark	Accuracy	Correct	Total	Time(s)
NVIDIA-Nemotron-3-Nano-4B-oQ3.5	MMLU	53.7%	161	300	572.9
NVIDIA-Nemotron-3-Nano-4B-oQ3.5	JMMLU	45.3%	136	300	159.8
NVIDIA-Nemotron-3-Nano-4B-oQ3.5	HELLASWAG	66.0%	198	300	200.4
NVIDIA-Nemotron-3-Nano-4B-oQ3.5	ARC_CHALLENGE	79.3%	238	300	114.9
NVIDIA-Nemotron-3-Nano-4B-oQ3.5	GSM8K	74.0%	222	300	904
NVIDIA-Nemotron-3-Nano-4B-oQ4	MMLU	61.0%	183	300	612.2
NVIDIA-Nemotron-3-Nano-4B-oQ4	JMMLU	51.7%	155	300	162.3
NVIDIA-Nemotron-3-Nano-4B-oQ4	HELLASWAG	71.7%	215	300	210.4
NVIDIA-Nemotron-3-Nano-4B-oQ4	ARC_CHALLENGE	81.7%	245	300	116.1
NVIDIA-Nemotron-3-Nano-4B-oQ4	GSM8K	81.3%	244	300	885.6