Recurrent Gemma

-> Go to Benchmark

Model overview

Recurrent Gemma architecture developed by Google is a hybrid model architecture.

Recurrent Gemma leverage both attention and state space model.

At the moment (16/08/2024), Recurrent Gemma have 1 major version.

The major version is RecurrentGemma.

Model details

Huggingface model

Support quantization

RecurrentGemma models support quantization.

Quantization Technique	KV cache	Support	Updated
FP16	FP16	Yes	Yes
FP8	FP16	No	No
FP8	FP8	No	No
int8_weight_only	FP16	No	No
int4_weight_only	FP16	No	No
w4a16_awq	FP16	No	No
w4a8_awq	FP16	No	No

Model versions

Model version	Model	Context size
RecurrentGemma	recurrentgemma-2b	4,096
RecurrentGemma	recurrentgemma-9b	4,096