Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -237,3 +237,61 @@ with ui.navset_card_tab(id="tab"):
|
|
237 |
)
|
238 |
if fig:
|
239 |
return fig
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
237 |
)
|
238 |
if fig:
|
239 |
return fig
|
240 |
+
with ui.nav_panel("Logits View"):
|
241 |
+
ui.panel_title("Logits et all")
|
242 |
+
with ui.card():
|
243 |
+
ui.input_selectize(
|
244 |
+
"model_bigness",
|
245 |
+
"Select Model size:",
|
246 |
+
["14", "31", "70", "160", "410"],
|
247 |
+
multiple=True,
|
248 |
+
selected=["70", "160"],
|
249 |
+
)
|
250 |
+
ui.input_selectize(
|
251 |
+
"loss_loss_loss",
|
252 |
+
"Select Loss Type:",
|
253 |
+
["compliment", "cross_entropy", "headless", "2d_representation_GaussianPlusCE", "2d_representation_MSEPlusCE"],
|
254 |
+
multiple=True,
|
255 |
+
selected=["cross_entropy"],
|
256 |
+
)
|
257 |
+
ui.input_selectize(
|
258 |
+
"logits_select",
|
259 |
+
"Select logits:",
|
260 |
+
["1", "2", "3", "4", "5", "6", "7", "8"],
|
261 |
+
multiple=True,
|
262 |
+
selected=["6"],
|
263 |
+
)
|
264 |
+
|
265 |
+
def plot_logits_representation(model_bigness, loss_type, logits):
|
266 |
+
num_rows = 2 # Number of rows in the subplot grid
|
267 |
+
num_cols = len(logits) # Number of columns based on the number of selected logits
|
268 |
+
fig, axs = plt.subplots(num_rows, num_cols, figsize=(20, 10))
|
269 |
+
axs = axs.flatten() # Flatten axs to handle 1D indexing
|
270 |
+
|
271 |
+
for size in model_bigness:
|
272 |
+
for loss in loss_type:
|
273 |
+
file_name = f"virus_pythia_{size}_1024_{loss}_logit_cumsums.npy"
|
274 |
+
if os.path.exists(file_name):
|
275 |
+
data = np.load(file_name, allow_pickle=True).item()
|
276 |
+
for k, logit in enumerate(logits):
|
277 |
+
logit_index = int(logit) - 1
|
278 |
+
axs[k].plot(data['lm_logits_y_cumsum'][0, :, logit_index], label=f'Generated_{loss}_{size}')
|
279 |
+
axs[k].plot(data['shift_labels_y_cumsum'][0, :, logit_index], label=f'Expected_{loss}_{size}')
|
280 |
+
axs[k].set_title(f'Logit: {logit}')
|
281 |
+
axs[k].legend()
|
282 |
+
else:
|
283 |
+
print(f"File not found: {file_name}")
|
284 |
+
|
285 |
+
for k in range(len(logits), num_cols):
|
286 |
+
fig.delaxes(axs[k]) # Remove any extra subplots if fewer logits are selected
|
287 |
+
|
288 |
+
plt.tight_layout()
|
289 |
+
return fig
|
290 |
+
|
291 |
+
@render.plot()
|
292 |
+
def plot_logits_representation_ui():
|
293 |
+
fig = plot_logits_representation(
|
294 |
+
input.model_bigness(), input.loss_loss_loss(), input.logits_select()
|
295 |
+
)
|
296 |
+
if fig:
|
297 |
+
return fig
|