main
official state-spaces/mamba repository, Mamba-2 paper and checkpoint
STATE SPACES / SSD / 2024

Mamba 2 · 2.7B

TOTAL2.7Bparameters
ACTIVE2.7Bper token
DEPTH64repeated blocks
HIDDEN2,560main trunk
CONTEXTstate-basednative tokens
CHECKPOINT≈5.4GB1 shards · Apache-2.0
INTERACTIVE SYSTEM BLUEPRINT

MODEL MRI

Mamba 2 · 2.7B · Architecture Explorer

100K INPUT + 1K OUTPUT · 562.34T arithmetic FLOPs · click any module for evidence

VERIFIED ADAPTER
INPUT100.0KTEXT
TRANSFORMER STACK64 MAMBA-2 BLOCKScausal convolution + SSD state mixer
RMSNorm Residual
++
OUTPUT1.0KTOKENS
wraps every repeated block
SSD STATE MIXER · STRUCTURED STATE SPACE DUALITYSSD 给出训练并行矩阵形式与推理递归形式之间的统一视角。
Gated Expansion输入投影、短卷积、状态更新与门控输出组成一个 Mamba-2 block。
DEPTH64Mamba-2
HIDDEN2,560checkpoint
STATE128typical
HEAD DIM64SSD
ATTENTION0layers
PARAMS2.7Bdense