Update app.py
Browse files
app.py
CHANGED
@@ -147,18 +147,17 @@ with gr.Blocks(theme=gr.themes.Default(), css=css) as demo:
|
|
147 |
global_state = gr.State([])
|
148 |
with gr.Row():
|
149 |
with gr.Column(scale=5):
|
150 |
-
gr.Markdown(''
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
''', line_breaks=True)
|
162 |
with gr.Column(scale=1):
|
163 |
gr.Markdown('<span style="font-size:180px;">π€</span>')
|
164 |
|
|
|
147 |
global_state = gr.State([])
|
148 |
with gr.Row():
|
149 |
with gr.Column(scale=5):
|
150 |
+
gr.Markdown('# π Self-Interpreting Models')
|
151 |
+
with gr.Accordion(label='πΎ **This space is a simple introduction to the emerging trend of models interpreting their _own hidden states_ in free form natural language**!! πΎ'):
|
152 |
+
gr.Markdown(
|
153 |
+
'''This idea was explored in the paper **Patchscopes** ([Ghandeharioun et al., 2024](https://arxiv.org/abs/2401.06102)) and was later investigated further in **SelfIE** ([Chen et al., 2024](https://arxiv.org/abs/2403.10949)).
|
154 |
+
An honorary mention of **Speaking Probes** ([Dar, 2023](https://towardsdatascience.com/speaking-probes-self-interpreting-models-7a3dc6cb33d6) -- my own work!! π₯³) which was less mature but had the same idea in mind.
|
155 |
+
We will follow the SelfIE implementation in this space for concreteness. Patchscopes are so general that they encompass many other interpretation techniques too!!!
|
156 |
+
|
157 |
+
πΎ **The idea is really simple: models are able to understand their own hidden states by nature!** πΎ
|
158 |
+
If I give a model a prompt of the form ``User: [X] Assistant: Sure'll I'll repeat your message`` and replace ``[X]`` *during computation* with the hidden state we want to understand,
|
159 |
+
we hope to get back a summary of the information that exists inside the hidden state, because it is encoded in a latent space the model uses itself!! How cool is that! π―π―π―
|
160 |
+
''', line_breaks=True)
|
|
|
161 |
with gr.Column(scale=1):
|
162 |
gr.Markdown('<span style="font-size:180px;">π€</span>')
|
163 |
|