Spaces:
Sleeping
Sleeping
Static to streamlit
Browse files
Home.py
ADDED
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from PIL import Image
|
3 |
+
|
4 |
+
st.set_page_config(layout="wide")
|
5 |
+
|
6 |
+
col1, col2, col3 = st.columns(3)
|
7 |
+
with col2:
|
8 |
+
st.markdown("""<div style='background-color: #0f5132; padding: 10px; border-radius: 5px; color: white;'><center>Une version en français est disponible sur mon <a href="https://lbourdois.github.io/blog/ssm/">blog</a></center></div>""", unsafe_allow_html=True)
|
9 |
+
st.markdown(""" """)
|
10 |
+
st.markdown(""" """)
|
11 |
+
|
12 |
+
st.markdown("""
|
13 |
+
October 7, 2021, while wondering whether [AK](https://hf.co/akhaliq) was a bot or a human, I saw one of his [tweets](https://twitter.com/_akhaliq/status/1445931206030282756). A link to a publication on [open-review.net](https://openreview.net/forum?id=uYLFoz1vlAC) accompanied by the following image:
|
14 |
+
""")
|
15 |
+
|
16 |
+
st.markdown("""
|
17 |
+
<div style="display: flex; justify-content: center;">
|
18 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/613b0a62a14099d5afed7830/CActxqIyAyMCa23qujg4L.png" alt="Centered Image">
|
19 |
+
</div>
|
20 |
+
""", unsafe_allow_html=True)
|
21 |
+
st.markdown(""" """)
|
22 |
+
|
23 |
+
st.markdown("""
|
24 |
+
Intrigued by the results announced, I decided to read about this S3 model, which would be renamed less than a month later to [S4](https://twitter.com/_albertgu/status/1456031299194470407) ([link](https://github.com/lbourdois/blog/blob/master/assets/efficiently_modeling_long_sequences_s3.pdf) of the version from when it was still called S3 for those interested).
|
25 |
+
This brilliant article impressed me. At the time, I was convinced that State Space Models (SSM) were going to be a revolution, replacing transformers in the coming months. Two years later, I'm forced to admit that I was completely wrong, given the tsunami of LLMs making the news in NLP.
|
26 |
+
Nevertheless, on Monday December 4, 2023, the announcement of Mamba by [Albert Gu](https://twitter.com/_albertgu/status/1731727672286294400) and [Tri Dao](https://twitter.com/tri_dao/status/1731728602230890895) revived their interest. This phenomenon was accentuated 4 days later with the announcement of [StripedHyena](https://twitter.com/togethercompute/status/1733213267185762411) by Together AI.
|
27 |
+
A good opportunity for me to write a few words about the developments in SSM over the last two years.
|
28 |
+
|
29 |
+
I plan to write three articles first, where the aim is to illustrate the basics of SSM with S4 before doing a literature review of the evolution of SSM since that first paper:
|
30 |
+
- [Introduction to SSM and S4](https://huggingface.co/blog/lbourdois/get-on-the-ssm-train)
|
31 |
+
- [SSM's history in 2022](https://huggingface.co/blog/lbourdois/ssm-2022)
|
32 |
+
- [SSM's history in 2023](WIP) (WIP)
|
33 |
+
""")
|