Балаганский Никита Николаевич commited on
Commit
df20599
1 Parent(s): 5d91754

fix nested expander

Browse files
Files changed (1) hide show
  1. app.py +4 -4
app.py CHANGED
@@ -154,10 +154,10 @@ def main():
154
  figure = go.Figure(data=[scatter, scatter_tip, scatter_tip_lines], layout=layout)
155
  figure.update_layout(paper_bgcolor="#FFFFFF", plot_bgcolor='#FFFFFF', showlegend=False)
156
  st.plotly_chart(figure, use_container_width=True)
157
- with st.expander("What is it?"):
158
- st.write("""
159
- Text generation with an external classifier requires a huge amount of computation. Therefore text generating with CAIF could be slow. To overcome this issue, we can apply reweighting not for every step. Our hypothesis is that we can run reweighting only if entropy of the next token is above certain threshold. This strategy will reduce the amont of computation. Note that if entropy threshold is too high, we don't get desired attribute in generated text
160
- """)
161
  fp16 = st.checkbox("FP16", value=True)
162
  st.session_state["generated_text"] = None
163
  if "sst" in cls_model_name:
 
154
  figure = go.Figure(data=[scatter, scatter_tip, scatter_tip_lines], layout=layout)
155
  figure.update_layout(paper_bgcolor="#FFFFFF", plot_bgcolor='#FFFFFF', showlegend=False)
156
  st.plotly_chart(figure, use_container_width=True)
157
+ st.subheader("What is it?")
158
+ st.write("""
159
+ Text generation with an external classifier requires a huge amount of computation. Therefore text generating with CAIF could be slow. To overcome this issue, we can apply reweighting not for every step. Our hypothesis is that we can run reweighting only if entropy of the next token is above certain threshold. This strategy will reduce the amont of computation. Note that if entropy threshold is too high, we don't get desired attribute in generated text
160
+ """)
161
  fp16 = st.checkbox("FP16", value=True)
162
  st.session_state["generated_text"] = None
163
  if "sst" in cls_model_name: