Datasets:

google
/

gemma3n-slicing-configs

name stringlengths 10 31	# Layers int64 30 35	# Effective Params (B) float64 1.91 3.98	MMLU PT accuracy stringlengths 6 6	FFN Hidden Dims stringlengths 330 385	Layers Skipped stringclasses 1 value
Main model	35	3.98	62.30%	[2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]	null
Config for official E2B Model	30	1.91	50.90%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	[20, 21, 22, 23, 24]
Config for E1.96B (layer-level)	30	1.96	53.40%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	[20, 21, 22, 23, 24]
Config for E2.54B (layer-level)	35	2.54	55.40%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E2.69B (layer-level)	35	2.69	57.70%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E2.98B (layer-level)	35	2.98	59.50%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]	null
Config for E3.18B (layer-level)	35	3.18	61.80%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8]	null
Config for E3.39B (layer-level)	35	3.39	63.00%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8]	null
Config for E3.59B (layer-level)	35	3.59	63.40%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8]	null
Config for E3.79B (layer-level)	35	3.79	63.40%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]	null
Config for E2.49B (block-level)	35	2.49	54.50%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E2.73B (block-level)	35	2.73	57.10%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E2.98B (block-level)	35	2.98	59.50%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E3.24B (block-level)	35	3.24	60.70%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E3.49B (block-level)	35	3.49	61.40%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]	null
Config for E3.79B (block-level)	35	3.74	62.00%	[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]	null

This repository contains configurations to slice Gemma 3n E4B, which is enabled thanks to it being a MatFormer. The E4B model can be sliced into small models, trading off quality and latency/compute requirements. We recommend exploring the [MatFormer Lab](TODO: add link) to getting started with slicing Gemma 3n E4B yourself.

For each configuration, we calculate the MMLU accuracy. Although these are not the only configurations possible, they are optimal configurations identified by calculating the accuracy of the pre-trained model

To learn more about MatFormers, please review the and generate your own submodels with the [MatFormer Lab](TODO: add link).

This chart show’s MMLU performance vs model size of Gemma 3n Mix-n-Match (pretrained) capability.

Some additional resources:

Downloads last month: 8

Size of downloaded dataset files:

6.97 kB

Size of the auto-converted Parquet files:

6.38 kB

Number of rows: