Emanuela Boros
commited on
Commit
·
69c8f9c
1
Parent(s):
f68e037
added confidence
Browse files- generic_nel.py +7 -8
generic_nel.py
CHANGED
@@ -111,26 +111,25 @@ class NelPipeline(Pipeline):
|
|
111 |
num_beams=1,
|
112 |
num_return_sequences=1,
|
113 |
max_new_tokens=30,
|
114 |
-
|
115 |
-
|
116 |
)
|
117 |
-
|
118 |
-
|
119 |
# Decode the predictions into readable text
|
120 |
wikipedia_predictions = self.tokenizer.batch_decode(
|
121 |
-
outputs, skip_special_tokens=True
|
122 |
)
|
123 |
# Process the scores for each token
|
124 |
|
125 |
# print(scores)
|
126 |
# print(type(scores), len(scores), type(list(scores)))
|
127 |
-
|
128 |
#
|
129 |
# # scores_tensor = scores.clone().detach()
|
130 |
-
|
131 |
# percentages = (probabilities * 100.0).cpu().numpy().tolist()
|
132 |
|
133 |
-
percentages = [0, 1, 2, 3, 3, 5]
|
134 |
# # Process each score (logits for the generated tokens)
|
135 |
# for i, score in enumerate(scores):
|
136 |
# # Apply softmax to convert logits into probabilities
|
|
|
111 |
num_beams=1,
|
112 |
num_return_sequences=1,
|
113 |
max_new_tokens=30,
|
114 |
+
return_dict_in_generate=True,
|
115 |
+
output_scores=True,
|
116 |
)
|
117 |
+
print(outputs.keys(), "!!!!")
|
118 |
+
token_ids, scores = outputs.sequences, outputs.scores
|
119 |
# Decode the predictions into readable text
|
120 |
wikipedia_predictions = self.tokenizer.batch_decode(
|
121 |
+
outputs.sequences, skip_special_tokens=True
|
122 |
)
|
123 |
# Process the scores for each token
|
124 |
|
125 |
# print(scores)
|
126 |
# print(type(scores), len(scores), type(list(scores)))
|
127 |
+
scores = torch.cat(scores, dim=0)
|
128 |
#
|
129 |
# # scores_tensor = scores.clone().detach()
|
130 |
+
probabilities = torch.exp(scores)
|
131 |
# percentages = (probabilities * 100.0).cpu().numpy().tolist()
|
132 |
|
|
|
133 |
# # Process each score (logits for the generated tokens)
|
134 |
# for i, score in enumerate(scores):
|
135 |
# # Apply softmax to convert logits into probabilities
|