Spaces:
Sleeping
Sleeping
Varun Wadhwa
commited on
Logs
Browse files
app.py
CHANGED
@@ -122,6 +122,8 @@ def evaluate_model(model, dataloader, device):
|
|
122 |
model.eval() # Set model to evaluation mode
|
123 |
all_preds = []
|
124 |
all_labels = []
|
|
|
|
|
125 |
|
126 |
# Disable gradient calculations
|
127 |
with torch.no_grad():
|
@@ -149,12 +151,18 @@ def evaluate_model(model, dataloader, device):
|
|
149 |
all_preds.extend(valid_preds.tolist())
|
150 |
all_labels.extend(valid_labels.tolist())
|
151 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
# Calculate evaluation metrics
|
153 |
print("evaluate_model sizes")
|
154 |
print(len(all_preds))
|
155 |
print(len(all_labels))
|
156 |
-
print(id2label[all_preds[0]])
|
157 |
-
print(id2label[all_labels[0]])
|
158 |
all_preds = np.asarray(all_preds, dtype=np.float32)
|
159 |
all_labels = np.asarray(all_labels, dtype=np.float32)
|
160 |
accuracy = accuracy_score(all_labels, all_preds)
|
|
|
122 |
model.eval() # Set model to evaluation mode
|
123 |
all_preds = []
|
124 |
all_labels = []
|
125 |
+
sample_count = 0
|
126 |
+
num_samples=100
|
127 |
|
128 |
# Disable gradient calculations
|
129 |
with torch.no_grad():
|
|
|
151 |
all_preds.extend(valid_preds.tolist())
|
152 |
all_labels.extend(valid_labels.tolist())
|
153 |
|
154 |
+
if sample_count < num_samples:
|
155 |
+
print(f"Sample {sample_count + 1}:")
|
156 |
+
print(f"Tokens: {tokenizer.convert_ids_to_tokens(input_ids[i])}")
|
157 |
+
print(f"True Labels: {[id2label[label] for label in valid_labels]}")
|
158 |
+
print(f"Predicted Labels: {[id2label[pred] for pred in valid_preds]}")
|
159 |
+
print("-" * 50)
|
160 |
+
sample_count += 1
|
161 |
+
|
162 |
# Calculate evaluation metrics
|
163 |
print("evaluate_model sizes")
|
164 |
print(len(all_preds))
|
165 |
print(len(all_labels))
|
|
|
|
|
166 |
all_preds = np.asarray(all_preds, dtype=np.float32)
|
167 |
all_labels = np.asarray(all_labels, dtype=np.float32)
|
168 |
accuracy = accuracy_score(all_labels, all_preds)
|