Repository navigation
Expand file tree
/
Copy pathcompare_method.py
More file actions
329 lines (277 loc) · 14.6 KB
/
Copy pathcompare_method.py
File metadata and controls
329 lines (277 loc) · 14.6 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
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
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
import json
import copy
import gc
import random
import torch
import numpy as np
import numpy as np
from FlagEmbedding import FlagModel
from scipy.optimize import linear_sum_assignment
from utils.utils import *
from utils.rapidapi import RapidAPICall
from models.gpt import GPTModel
from prompts.compare import system_prompt, user_prompt
from utils.logger import Logger
class CompareFCBase:
def __init__(self, args, logger) -> None:
self.embedding = FlagModel('BAAI/bge-large-en-v1.5',
query_instruction_for_retrieval="Represent this sentence for searching relevant passages:",
use_fp16=True)
with open("utils/tool_info.json", 'r') as f:
tool_info = json.load(f)
tool_info = tool_info['booking-com15']
self.api_call = RapidAPICall(tool="booking-com15", tool_info=tool_info)
self.model = GPTModel("gpt-4o-2024-05-13")
self.logger = logger
self.error_message = []
self.exact_match_dict = load_json("utils/exact_match_values.json")
self.free_function_list = ["Location_to_Lat_Long", "Search_Hotel_Destination", "Search_Attraction_Location",
"Search_Car_Location", "Search_Flight_Location", "Taxi_Search_Location"]
def format_check(self, func_call, functions):
name_to_func = {func['name']: func for func in functions}
if func_call['name'] not in name_to_func:
return {"error": f"Function {func_call['name']} is not defined in the function list."}
used_func = name_to_func[func_call['name']]
required_params = used_func['parameters']['required']
if not set(required_params).issubset(set(func_call['arguments'].keys())):
# 找到在required_params中，但不在func_call['arguments'].keys()中的参数
missing_param = set(required_params) - set(func_call['arguments'].keys())
return {"error": f"Function {used_func['name']} requires parameters {required_params}, but {list(func_call['arguments'].keys())} do not provide {missing_param}"}
if not set(func_call['arguments'].keys()).issubset(set(used_func['parameters']['properties'].keys())):
missing_param = set(func_call['arguments'].keys()) - set(used_func['parameters']['properties'].keys())
return {"error":f"Function {used_func['name']} does not have parameters {missing_param}" }
# 参数类型验证
for param_name, param_value in func_call['arguments'].items():
if used_func['parameters']['properties'][param_name]['type'] == "string":
if not isinstance(param_value, str):
return {"error": f"Parameter {param_name} of function {used_func['name']} should be a string, but {type(param_value)} is provided."}
elif used_func['parameters']['properties'][param_name]['type'] == "number":
if not isinstance(param_value, int) and not isinstance(param_value, float):
return {"error": f"Parameter {param_name} of function {used_func['name']} should be a number, but {type(param_value)} is provided."}
elif used_func['parameters']['properties'][param_name]['type'] == "boolean":
if not isinstance(param_value, bool):
return {"error": f"Parameter {param_name} of function {used_func['name']} should be a boolean, but {type(param_value)} is provided."}
elif used_func['parameters']['properties'][param_name]['type'] == "array":
if not isinstance(param_value, list):
return {"error": f"Parameter {param_name} of function {used_func['name']} should be an array, but {type(param_value)} is provided."}
return True
def add_free_function(self, convs):
# free function is optional
self.free_functions = {}
for i, turn in enumerate(convs):
if "function_call" not in turn:
continue
for j, func_call in enumerate(turn['function_call']):
if func_call['name'] in self.free_function_list:
if json.dumps(func_call) not in self.free_functions:
self.free_functions[json.dumps(func_call)] = {
"called": False,
"obs": convs[i+1]['content'][j]
}
def rule_based(self, predict, golden):
"""
Rule-based Match.
"""
if predict['name'] != golden['name']:
return False
if sorted(predict['arguments'].keys()) != sorted(golden['arguments'].keys()):
return False
for k, v in predict['arguments'].items():
if k == "categories_filter":
pred_filter = [s.strip() for s in v.split(",")]
golden_filter = [s.strip() for s in golden['arguments'][k].split(",")]
if set(golden_filter) == set(pred_filter):
continue
if v != golden['arguments'][k]:
return False
return True
def response_based(self, predict, golden):
try:
resp_1 = self.api_call._call(predict)
if resp_1 == {}:
self.error_message = f"API call failed for {predict}."
return False
if isinstance(resp_1, dict):
if "status" in resp_1 and resp_1["status"] == False:
self.error_response = resp_1
else:
self.error_response = resp_1
resp_2 = self.api_call._call(golden)
except:
return False
if resp_1 is None or resp_2 is None:
return False
return resp_1 == resp_2
def similarity_based(self, predict, golden):
embedding_1 = self.embedding.encode([json.dumps(predict, ensure_ascii=False)])
embedding_2 = self.embedding.encode([json.dumps(golden, ensure_ascii=False)])
similarity = embedding_1 @ embedding_2.T
del embedding_1, embedding_2
torch.cuda.empty_cache()
gc.collect()
self.logger.debug(f"Similarity-based comparison output: {similarity[0][0]}")
return similarity[0][0] > 0.98
def llm_based(self, functions, history, predict, golden):
kwargs = {
"functions": json.dumps(functions, ensure_ascii=False),
"history": json.dumps(history, ensure_ascii=False),
"function_call_1": json.dumps(predict, ensure_ascii=False),
"function_call_2": json.dumps(golden, ensure_ascii=False),
}
output = self.model(system_prompt, user_prompt, **kwargs)
decode_output = decode_json(output)
if decode_output:
self.logger.debug(f"LLM-based comparison output: {decode_output}")
return decode_output['is_equal']
else:
return None
class CompareFC(CompareFCBase):
def __init__(self, args, logger) -> None:
super().__init__(args, logger)
def value_checker(self, pred_call, golden_call):
if pred_call['name'] != golden_call['name']:
return False, {"error_type": "func_error", "content": "Do not call the correct function."}
param_list = self.exact_match_dict[pred_call['name']]
for k, v in golden_call['arguments'].items():
if k in param_list:
if k not in pred_call['arguments']:
return False, {"error_type": "param_missing", "content": f"Missing parameter {k} in prediction."}
elif k != "categories_filter" and pred_call['arguments'][k] != v:
return False, {"error_type": "value_error", "content": f"Parameter {k} value is not correct in prediction."}
elif k == "categories_filter":
golden_filter = [s.strip() for s in v.split(",")]
pred_filter = [s.strip() for s in pred_call['arguments'][k].split(",")]
if not set(golden_filter) == set(pred_filter):
return False, {"error_type": "value_error", "content": f"Parameter {k} value is not correct in prediction."}
return True, ""
def remove_called_fc(self, golden, golden_obs):
pop_index = []
for singel_golden in golden:
if json.dumps(singel_golden) in self.free_functions and self.free_functions[json.dumps(singel_golden)]['called'] == True:
pop_index.append(golden.index(singel_golden))
for index in sorted(pop_index, reverse=True):
golden.pop(index)
golden_obs.pop(index)
return golden, golden_obs
def get_error_message(self, pred_call, golden_call):
# value error
for k, v in golden_call['arguments'].items():
if k not in pred_call['arguments']:
return {"error_type": "param_missing", "content": f"Missing parameter {k} in prediction."}
if v != pred_call['arguments'][k]:
return {"error_type": "value_error", "content": f"Parameter {k} value do not equal to golden."}
# hallucination
for k, v in pred_call['arguments'].items():
if k not in golden_call['arguments']:
return {'error_type': "param_hallucination", "content": f"Parameter {k} is hallucinated."}
def mapping_call(self, predict, golden, golden_obs):
def sort_arguments(call_list):
for value in call_list:
sorted_arguments = {k: value['arguments'][k] for k in sorted(value['arguments'])}
value['arguments'] = sorted_arguments
sort_arguments(predict)
sort_arguments(golden)
# exact match
exact_matches = []
remaining_predict = []
remaining_predict_index = {}
remaining_golden = []
remaining_golden_index = {}
matched_indices = set()
for p_index, p_value in enumerate(predict):
match_found = False
for g_index, g_value in enumerate(golden):
if g_index in matched_indices:
continue
if p_value == g_value:
exact_matches.append({
"idx": p_index,
"pred_call": p_value,
"golden_call": g_value,
"golden_obs": golden_obs[g_index]
})
matched_indices.add(g_index)
match_found = True
break
elif json.dumps(p_value) in self.free_functions:
exact_matches.append({
"idx": p_index,
"pred_call": p_value,
"golden_call": p_value,
"golden_obs": self.free_functions[json.dumps(p_value)]['obs']
})
match_found = True
self.free_functions[json.dumps(p_value)]['called'] = True
break
if not match_found:
remaining_predict.append(p_value)
remaining_predict_index[len(remaining_predict) - 1] = p_index
for g_index, g_value in enumerate(golden):
if g_index not in matched_indices:
remaining_golden.append(g_value)
remaining_golden_index[len(remaining_golden) - 1] = g_index
if remaining_predict == [] or remaining_golden == []:
return exact_matches
# embedding match
pred_embed = self.embedding.encode([json.dumps(value, ensure_ascii=False) for value in remaining_predict])
gold_embed = self.embedding.encode([json.dumps(value, ensure_ascii=False) for value in remaining_golden])
matrix = pred_embed @ gold_embed.T
del pred_embed, gold_embed
torch.cuda.empty_cache()
gc.collect()
row_ind, col_ind = linear_sum_assignment(-matrix)
embedding_matches = []
for i, j in zip(row_ind, col_ind):
embedding_matches.append({
"idx": remaining_predict_index[i],
"pred_call": remaining_predict[i],
"golden_call": remaining_golden[j],
"golden_obs": golden_obs[remaining_golden_index[j]]
})
matching = exact_matches + embedding_matches
return matching
def compare_single_call(self, functions, history, pred_call, golden_call):
self.logger.info(f"Start compare_single_call: \n{pred_call}\n{golden_call}")
# rule-based
if self.rule_based(pred_call, golden_call):
self.logger.info(f"Rule-based compare success.")
return True, None
is_valid, error_message = self.value_checker(pred_call, golden_call)
if not is_valid:
self.logger.info(f"{error_message}")
return False, error_message
# Response-based
if self.response_based(pred_call, golden_call):
self.logger.info(f"Response-based compare success.")
return True, None
# LLM-based
if self.llm_based(functions, history, pred_call, golden_call):
self.logger.info(f"LLM-based compare success.")
return True, None
self.logger.info(f"All compare method failed.")
return False, None
def compare_turn_prediction(self, functions, history, predict, golden, golden_obs):
self.error_message = []
golden, golden_obs = self.remove_called_fc(golden, golden_obs)
if len(golden) == 0:
raise NotImplementedError()
match_list = self.mapping_call(predict, golden, golden_obs)
format_error = {}
success_map, success_matched = {}, []
for match_item in match_list:
# format error check
message = self.format_check(match_item['pred_call'], functions)
if message == True:
is_match, single_message = self.compare_single_call(functions, history, match_item['pred_call'], match_item['golden_call'])
if is_match:
success_map[match_item['idx']] = match_item['golden_obs']
success_matched.append(match_item['golden_call'])
else:
if single_message:
self.error_message.append(single_message)
else:
self.error_message.append(self.get_error_message(match_item['pred_call'], match_item['golden_call']))
else:
format_error[match_item['idx']] = message
self.logger.info(f"Success matched: {success_matched}")
return self.error_message, success_map, success_matched, format_error