Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1004,7 +1004,8 @@ with ui.navset_card_tab(id="tab"):
|
|
1004 |
multiple=True,
|
1005 |
selected=["compliment", "cross_entropy", "headless"]
|
1006 |
)
|
1007 |
-
|
|
|
1008 |
# interplot each column to be same number of points
|
1009 |
x = np.linspace(0, 1, 1000)
|
1010 |
loss_rates = []
|
@@ -1022,9 +1023,15 @@ with ui.navset_card_tab(id="tab"):
|
|
1022 |
labels.append(str(param_type) + '_' + loss_type + '_' + model_type)
|
1023 |
|
1024 |
fig, ax = plt.subplots()
|
1025 |
-
|
1026 |
|
1027 |
for i, loss_rate in enumerate(loss_rates):
|
|
|
|
|
|
|
|
|
|
|
|
|
1028 |
ax.plot(x, loss_rate, label=labels[i])
|
1029 |
|
1030 |
ax.legend()
|
@@ -1034,12 +1041,18 @@ with ui.navset_card_tab(id="tab"):
|
|
1034 |
return fig
|
1035 |
|
1036 |
import matplotlib as mpl
|
1037 |
-
@render.
|
1038 |
def plot_model_scaling():
|
1039 |
fig = None
|
1040 |
df = pd.read_csv('training_data_5.csv')
|
1041 |
mpl.rcParams.update(mpl.rcParamsDefault)
|
1042 |
-
fig = plot_loss_rates_model(df, input.param_type(),input.loss_type(),input.model_type())
|
|
|
|
|
|
|
|
|
|
|
|
|
1043 |
return fig
|
1044 |
with ui.nav_panel("Scaling Laws"):
|
1045 |
ui.page_opts(fillable=True)
|
|
|
1004 |
multiple=True,
|
1005 |
selected=["compliment", "cross_entropy", "headless"]
|
1006 |
)
|
1007 |
+
ui.input_slider("x_filter", "x_filter", 0, 1, 0.01)
|
1008 |
+
def plot_loss_rates_model(df, param_types, loss_types, model_types, x_filter):
|
1009 |
# interplot each column to be same number of points
|
1010 |
x = np.linspace(0, 1, 1000)
|
1011 |
loss_rates = []
|
|
|
1023 |
labels.append(str(param_type) + '_' + loss_type + '_' + model_type)
|
1024 |
|
1025 |
fig, ax = plt.subplots()
|
1026 |
+
# print(loss_rates)
|
1027 |
|
1028 |
for i, loss_rate in enumerate(loss_rates):
|
1029 |
+
df_madmad = pd.DataFrame({'x':x, 'loss':loss_rate})
|
1030 |
+
|
1031 |
+
df_madmad = df_madmad.sort_values(by='x')
|
1032 |
+
df_madmad = df_madmad[df_madmad['x']>x_filter]
|
1033 |
+
x = df_madmad['x'].to_list()
|
1034 |
+
loss_rate = df_madmad['loss_rate'].to_list()
|
1035 |
ax.plot(x, loss_rate, label=labels[i])
|
1036 |
|
1037 |
ax.legend()
|
|
|
1041 |
return fig
|
1042 |
|
1043 |
import matplotlib as mpl
|
1044 |
+
@render.image
|
1045 |
def plot_model_scaling():
|
1046 |
fig = None
|
1047 |
df = pd.read_csv('training_data_5.csv')
|
1048 |
mpl.rcParams.update(mpl.rcParamsDefault)
|
1049 |
+
fig = plot_loss_rates_model(df, input.param_type(),input.loss_type(),input.model_type(),input.x_filter() )
|
1050 |
+
|
1051 |
+
import tempfile
|
1052 |
+
fd, path = tempfile.mkstemp(suffix = '.svg')
|
1053 |
+
if fig:
|
1054 |
+
fig.savefig(path)
|
1055 |
+
return {"src": str(path), "width": "600px", "format":"svg"}
|
1056 |
return fig
|
1057 |
with ui.nav_panel("Scaling Laws"):
|
1058 |
ui.page_opts(fillable=True)
|