Spaces:
Running
on
Zero
Running
on
Zero
da03
commited on
Commit
•
e5d31e2
1
Parent(s):
313c68d
app.py
CHANGED
@@ -56,90 +56,93 @@ def predict_product(num1, num2):
|
|
56 |
finished_per_model = {model_name: False for model_name in models}
|
57 |
past_key_values_per_model = {model_name: None for model_name in models}
|
58 |
predicted_annotations_per_model = {}
|
59 |
-
|
60 |
-
#
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
if
|
92 |
-
predicted_annotations_per_model[model_name]
|
93 |
-
|
|
|
|
|
|
|
|
|
94 |
|
95 |
-
|
96 |
-
|
97 |
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
if '=' not in predicted_digits_reversed:
|
105 |
-
predicted_annotations = [(predicted_digit, None) for predicted_digit in predicted_digits_reversed]
|
106 |
-
predicted_digits_reversed = []
|
107 |
-
else:
|
108 |
-
equal_sign_position = predicted_digits_reversed.index('=')
|
109 |
-
predicted_annotations = [(predicted_digit, None) for predicted_digit in predicted_digits_reversed[:equal_sign_position+1]]
|
110 |
-
predicted_digits_reversed = predicted_digits_reversed[equal_sign_position+1:]
|
111 |
-
|
112 |
-
for i in range(len(predicted_digits_reversed)):
|
113 |
-
predicted_digit = predicted_digits_reversed[i]
|
114 |
-
if not valid_input:
|
115 |
-
is_correct_digit = None
|
116 |
-
elif i >= len(ground_truth_digits_reversed):
|
117 |
-
if predicted_digit == '0' and is_correct_sofar:
|
118 |
-
is_correct_digit = True
|
119 |
else:
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
else:
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
|
144 |
color_map = {"correct": "green", "wrong": "red"}
|
145 |
|
|
|
56 |
finished_per_model = {model_name: False for model_name in models}
|
57 |
past_key_values_per_model = {model_name: None for model_name in models}
|
58 |
predicted_annotations_per_model = {}
|
59 |
+
try:
|
60 |
+
for step in range(max(MAX_PRODUCT_DIGITS_PER_MODEL.values())): # Set a maximum limit to prevent infinite loops
|
61 |
+
# Ground Truth
|
62 |
+
if not valid_input:
|
63 |
+
ground_truth_annotations = [('Invalid Input!', None)]
|
64 |
+
else:
|
65 |
+
ground_truth_annotations = [(ground_truth_digit, None) for ground_truth_digit in ground_truth_digits_reversed[:step+1]]
|
66 |
+
ground_truth_annotations = ground_truth_annotations[::-1]
|
67 |
+
# Predicted
|
68 |
+
for model_name in models:
|
69 |
+
model = models[model_name]
|
70 |
+
if finished_per_model[model_name]:
|
71 |
+
continue
|
72 |
+
if step >= MAX_PRODUCT_DIGITS_PER_MODEL[model_name]:
|
73 |
+
continue
|
74 |
+
generation_kwargs = {
|
75 |
+
'input_ids': generated_ids_per_model[model_name],
|
76 |
+
'max_new_tokens': 1,
|
77 |
+
'do_sample': False,
|
78 |
+
'past_key_values': past_key_values_per_model[model_name],
|
79 |
+
'return_dict_in_generate': True,
|
80 |
+
'use_cache': True
|
81 |
+
}
|
82 |
+
if step == 0:
|
83 |
+
del generation_kwargs['past_key_values']
|
84 |
+
outputs = model.generate(**generation_kwargs)
|
85 |
+
generated_ids = outputs.sequences
|
86 |
+
next_token_id = generated_ids[0, -1]
|
87 |
+
#print (next_token_id)
|
88 |
+
|
89 |
+
if next_token_id.item() == tokenizer.eos_token_id:
|
90 |
+
finished_per_model[model_name] = True
|
91 |
+
if valid_input:
|
92 |
+
if len([item for item in predicted_annotations_per_model[model_name] if item[1] is not None]) < len(ground_truth_digits_reversed):
|
93 |
+
predicted_annotations_per_model[model_name].insert(0, ('⠀', 'wrong'))
|
94 |
+
continue
|
95 |
+
|
96 |
+
generated_ids_per_model[model_name] = generated_ids
|
97 |
+
past_key_values_per_model[model_name] = outputs.past_key_values
|
98 |
|
99 |
+
output_text = tokenizer.decode(generated_ids[0, input_len:], skip_special_tokens=True)
|
100 |
+
predicted_digits_reversed = output_text.strip().split(' ')
|
101 |
|
102 |
+
predicted_annotations = []
|
103 |
+
is_correct_sofar = True
|
104 |
+
if model_name == 'explicit':
|
105 |
+
if '=' not in predicted_digits_reversed:
|
106 |
+
predicted_annotations = [(predicted_digit, None) for predicted_digit in predicted_digits_reversed]
|
107 |
+
predicted_digits_reversed = []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
else:
|
109 |
+
equal_sign_position = predicted_digits_reversed.index('=')
|
110 |
+
predicted_annotations = [(predicted_digit, None) for predicted_digit in predicted_digits_reversed[:equal_sign_position+1]]
|
111 |
+
predicted_digits_reversed = predicted_digits_reversed[equal_sign_position+1:]
|
112 |
+
|
113 |
+
for i in range(len(predicted_digits_reversed)):
|
114 |
+
predicted_digit = predicted_digits_reversed[i]
|
115 |
+
if not valid_input:
|
116 |
+
is_correct_digit = None
|
117 |
+
elif i >= len(ground_truth_digits_reversed):
|
118 |
+
if predicted_digit == '0' and is_correct_sofar:
|
119 |
+
is_correct_digit = True
|
120 |
+
else:
|
121 |
+
is_correct_digit = False
|
122 |
else:
|
123 |
+
ground_truth_digit = ground_truth_digits_reversed[i]
|
124 |
+
if predicted_digit == ground_truth_digit:
|
125 |
+
is_correct_digit = True
|
126 |
+
else:
|
127 |
+
is_correct_digit = False
|
128 |
+
if not is_correct_digit:
|
129 |
+
is_correct_sofar = False
|
130 |
+
if is_correct_digit is None:
|
131 |
+
predicted_annotations.append((predicted_digit, None))
|
132 |
+
elif is_correct_digit:
|
133 |
+
predicted_annotations.append((predicted_digit, "correct"))
|
134 |
+
else:
|
135 |
+
predicted_annotations.append((predicted_digit, "wrong"))
|
136 |
+
predicted_annotations = predicted_annotations[::-1]
|
137 |
+
predicted_annotations_per_model[model_name] = predicted_annotations
|
138 |
+
|
139 |
+
predicted_annotations_implicit_cot = predicted_annotations_per_model['implicit']
|
140 |
+
predicted_annotations_nocot = predicted_annotations_per_model['no']
|
141 |
+
predicted_annotations_explicit_cot = predicted_annotations_per_model['explicit']
|
142 |
+
|
143 |
+
yield ground_truth_annotations, predicted_annotations_implicit_cot, predicted_annotations_nocot, predicted_annotations_explicit_cot
|
144 |
+
except Exception as e:
|
145 |
+
pass
|
146 |
|
147 |
color_map = {"correct": "green", "wrong": "red"}
|
148 |
|