Spaces:
Sleeping
Sleeping
Upload README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# What is M E S ?
|
2 |
M E S (short for MAMBA ENCODER SWARM) is a novel architecture that comprises of MAMBA's structured state space, configured to implement a multiple encoder swarm that are dynamically, sparsely routed to spread the heavy QxKxV matrix multiplication computional intensity across multiple MAMBA encoders (between 5 to 1000) and with the output sparsely aggregated with a MAMBA decoder, thereby bypassing the high cost of inference without sacrificing on the response generation quality.
|
3 |
|
|
|
1 |
+
title: Mamba Encoder Swarm
|
2 |
+
emoji: 🐍
|
3 |
+
colorFrom: orange
|
4 |
+
colorTo: yellow
|
5 |
+
sdk: gradio
|
6 |
+
sdk_version: "4.0.0"
|
7 |
+
app_file: app.py
|
8 |
+
pinned: false
|
9 |
+
license: mit
|
10 |
+
|
11 |
+
|
12 |
# What is M E S ?
|
13 |
M E S (short for MAMBA ENCODER SWARM) is a novel architecture that comprises of MAMBA's structured state space, configured to implement a multiple encoder swarm that are dynamically, sparsely routed to spread the heavy QxKxV matrix multiplication computional intensity across multiple MAMBA encoders (between 5 to 1000) and with the output sparsely aggregated with a MAMBA decoder, thereby bypassing the high cost of inference without sacrificing on the response generation quality.
|
14 |
|