-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathcod_cli.py
378 lines (323 loc) · 23.2 KB
/
cod_cli.py
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
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
# 支持英文
import os
import platform
import torch
from threading import Thread
from transformers import AutoTokenizer
from transformers import AutoModelForCausalLM
import argparse
from transformers import TextIteratorStreamer
from transformers.generation.utils import GenerationConfig
import json
import re
import numpy as np
from retrieve_utils import (
construct_flatindex_from_embeddings,
convert_index_to_gpu,
encoder
)
import json
from transformers import LlamaForCausalLM
class DiagnosisChatbot():
def __init__(self, model_dir,confidence_threshold = 0.5):
self.max_len = 4096
self.gen_kwargs = {'max_new_tokens': 768, 'do_sample':True, 'top_p':0.7, 'temperature':0.5, 'repetition_penalty':1.1}
model, tokenizer = self.load_model(model_dir)
self.model = model
self.tokenizer = tokenizer
self.sep = tokenizer.convert_ids_to_tokens(self.tokenizer.eos_token_id)
excluded_ids = ['1656164150671335426', '1656164150562283522', '1656164170883686402', '1656164187891589122', '1656164159110275073', '1656164142131732482', '1656164167129784322', '1656164140797943810', '1656164130668699650', '1656164130798723073', '1656164140768583682', '1656164167251419137', '1656164134657482753', '1656164159026388994', '1656164140818915330', '1656164181088428034', '1656164153162752001', '1656164133688598529', '1656164142135926785', '1656164164525121538', '1656164185387589633']
excluded_ids_en = ['1656164174583062529', '1656164139271217154', '1656164172892758018', '1656164182577405953', '1656164182594183170', '1656164134418407426', '1656164172947283969', '1656164131574669313', '1656164190991179777', '1656164183869251585', '1656164151631831041', '1656164134384852994', '1656164160184016898', '1656164190873739265', '1656164184800387073', '1656164174494982146', '1656164147903094785', '1656164160087547906', '1656164168585207809', '1656164134472933377', '1656164183189774338', '1656164174901829635', '1656164152919482370', '1656164179532341249', '1656164141842325505', '1656164182682263553', '1656164151900266497', '1656164192618569730', '1656164136515559426', '1656164161626857473', '1656164154643341314', '1656164188592037890', '1656164156337840130', '1656164193751031810', '1656164185794437122', '1656164179209379842', '1656164164965523457', '1656164179310043137', '1656164179209379843', '1656164171294728193', '1656164188600426498', '1656164171584135170', '1656164184485814273', '1656164142261755906', '1656164193323212801', '1656164162025316354', '1656164152219033602', '1656164134061891586', '1656164193365155842', '1656164190978596865', '1656164179452649474', '1656164193918803971', '1656164136456839169', '1656164169449234433', '1656164192643735555', '1656164136893046785', '1656164193532928002', '1656164167213670402', '1656164171970011137', '1656164157814235138', '1656164152344862722', '1656164141129293825', '1656164160284680194', '1656164182636126210', '1656164174855692290', '1656164185802825730', '1656164165061992450', '1656164131801161730', '1656164152953036802', '1656164186138370050', '1656164185303703553', '1656164182648709122', '1656164183927971841', '1656164182757761025', '1656164152508440577', '1656164190856962050', '1656164145302626305', '1656164182585794561', '1656164192840867841', '1656164134812672001', '1656164171751907329', '1656164189644808194', '1656164171269562370', '1656164142249172994', '1656164190861156355', '1656164172783706114', '1656164171575746562', '1656164190848573443', '1656164167113007106', '1656164152336474115', '1656164134355492865', '1656164157537411074', '1656164185391783938', '1656164134468739074', '1656164151065600002', '1656164189485424641', '1656164168648122370', '1656164153263415298', '1656164185060433922', '1656164141318037506', '1656164157529022466', '1656164189607059457', '1656164142203035650', '1656164190609498113', '1656164193566482433', '1656164154332962818', '1656164144233078785', '1656164131155238913', '1656164190299119618', '1656164178823503873', '1656164167037509633', '1656164190525612033', '1656164185752494081', '1656164182912950274', '1656164144144998401', '1656164192903782402', '1656164193394515970', '1656164184993325058', '1656164157331890177', '1656164162281168897', '1656164184653586433', '1656164192941531138', '1656164145252294657', '1656164141737467906', '1656164164621590531', '1656164137140510721', '1656164144367296514', '1656164183542095874', '1656164141880074242', '1656164152336474114', '1656164184628420609', '1656164179322626049', '1656164169415680001', '1656164161886904321', '1656164142194647041', '1656164169168216066', '1656164164910997506', '1656164184594866178', '1656164141473226753', '1656164145562673154', '1656164157600325633', '1656164174553702401', '1656164142320476162', '1656164159177383938', '1656164128898703361', '1656164187950309377', '1656164152948842497', '1656164131662749697', '1656164192949919745', '1656164157621297153', '1656164159307407362', '1656164165162655746', '1656164173245079553', '1656164147211034626', '1656164171730935809', '1656164152382611458', '1656164174557896705', '1656164145407483907', '1656164144480542721', '1656164173056335874', '1656164147085205505', '1656164134498099202', '1656164159923970050', '1656164144988053505', '1656164184095744001', '1656164141523558402', '1656164145512341505', '1656164141938794498', '1656164171827404801', '1656164168790728706', '1656164151409532930', '1656164174746640385', '1656164152328085506', '1656164178722840577', '1656164162285363201', '1656164187681873922', '1656164141636804610', '1656164147013902337', '1656164174906023937', '1656164175166070786', '1656164192996057090', '1656164172720791553', '1656164193377738753', '1656164166605496321', '1656164174952161282', '1656164131604029442', '1656164142043652098', '1656164184045412354', '1656164190907293698', '1656164159928164353', '1656164141225762817', '1656164162146951169', '1656164168664899585', '1656164161597497346', '1656164141963960321', '1656164190311702529', '1656164193755226113', '1656164173144416257', '1656164190613692417', '1656164183546290178', '1656164185798631425', '1656164172674654209', '1656164141485809665', '1656164147286532098', '1656164182598377474', '1656164136477810689', '1656164141078962177', '1656164157193478145', '1656164164650950657', '1656164171688992770', '1656164172817260546', '1656164185387589634', '1656164159353544707', '1656164174549508098', '1656164182581600257', '1656164182669680641', '1656164174545313793', '1656164134326132738', '1656164188684312579', '1656164134405824514', '1656164151619248129', '1656164172695625730', '1656164141217374209', '1656164152512634882', '1656164141716496386', '1656164193289658369', '1656164137379586050', '1656164164634173442', '1656164169411485697', '1656164134460350465', '1656164172683042818', '1656164134842032130', '1656164174956355586', '1656164178815115266', '1656164174570479617', '1656164142312087553', '1656164157793263617', '1656164152713961473', '1656164171839987713', '1656164172724985857', '1656164188629786626', '1656164193574871041', '1656164142232395777', '1656164134447767554', '1656164192887005186', '1656164174566285314', '1656164157134757890', '1656164131708887042', '1656164172049702913', '1656164171370225666', '1656164193973329922', '1656164185844768770', '1656164156711133186', '1656164193893638146', '1656164179410706434', '1656164133973811201', '1656164172406218754', '1656164172867592193', '1656164152206450691', '1656164152806236161', '1656164157113786369', '1656164178638954498', '1656164185781854210', '1656164134527459330', '1656164145323597826', '1656164147886317570', '1656164171445723138', '1656164144216301569', '1656164188508151811', '1656164179205185537', '1656164144887390210', '1656164185890906114', '1656164182573211650', '1656164175010881538', '1656164165162655747', '1656164141846519809', '1656164154408460289', '1656164182938116099', '1656164132103151617', '1656164144568623105', '1656164190911488002', '1656164182464159745', '1656164164642562050', '1656164142198841345', '1656164182774538241', '1656164172108423170', '1656164167196893185', '1656164131658555393', '1656164172435578882', '1656164179528146946', '1656164159999467523', '1656164161844961283', '1656164167205281794', '1656164145378123778', '1656164171764490242', '1656164188801753090', '1656164174994104322', '1656164190294925314', '1656164165154267138', '1656164153389244419', '1656164189275709442', '1656164172615933953', '1656164185446309890', '1656164170602668033', '1656164157478690817', '1656164192182362113', '1656164147517218819', '1656164147815014403', '1656164179498786817', '1656164193348378625', '1656164134259023873', '1656164168681676802', '1656164134464544770', '1656164157411581953', '1656164169243713537', '1656164171705769985']
# Chinese Retriever
retriever_zh = self.load_retrieval(os.path.join(model_dir,'retriever/zh/encoder') ,
os.path.join(model_dir,'retriever/zh/index') ,
os.path.join(model_dir,'retriever/zh/disease_database_zh.json'),excluded_disease_id=excluded_ids)
# English Retriever
retriever_en = self.load_retrieval(os.path.join(model_dir,'retriever/en/encoder') ,
os.path.join(model_dir,'retriever/en/index') ,
os.path.join(model_dir,'retriever/en/disease_database_en.json'),excluded_disease_id=excluded_ids+excluded_ids_en)
self.retriever_zh = retriever_zh
self.retriever_en = retriever_en
self.confidence_threshold = confidence_threshold
self.min_sym_num = 2
self.history = []
self.sym_info = {'true_syms':[],'false_syms':[]}
def load_retrieval(self, model_path,pembed_dir,data_path,excluded_disease_id = []):
model = encoder(model_path)
model.to('cuda')
model.eval()
passage_embeddings = np.memmap(os.path.join(pembed_dir, "passages.memmap"), dtype=np.float32, mode="r"
).reshape(-1, model.output_embedding_size)
index = construct_flatindex_from_embeddings(passage_embeddings)
# 加载到GPU
# index = convert_index_to_gpu(index, faiss_gpu_index = 0, useFloat16 = False)
disease = []
diseaseid2id = {}
disease2id = {}
with open(data_path) as f:
datas = json.load(f)
for id,da in enumerate(datas):
disease.append(da)
diseaseid2id[da['disease_id']] = id
disease2id[da['disease']] = id
model.index = index
model.disease = disease
model.diseaseid2id = diseaseid2id
model.disease2id = disease2id
exclude_ids = []
for dis_id in excluded_disease_id:
exclude_ids.append(diseaseid2id[dis_id])
model.exclude_ids = exclude_ids
return model
def load_model(self,model_dir):
tokenizer = AutoTokenizer.from_pretrained(model_dir, use_fast=False, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_dir, torch_dtype=torch.bfloat16).cuda()
model = model.eval()
return model, tokenizer
def generate_prompt(self,query, history):
if not history:
return f"<|user|>\n{query}\n<|assistant|>\n"
else:
prompt = ''
for i, (old_query, response) in enumerate(history):
prompt += "<|user|>\n{}\n<|assistant|>\n{}\n".format(old_query, response)
prompt += "<|user|>\n{}\n<|assistant|>\n".format(query)
return prompt
def remove_overlap(self,str1, str2):
for i in range(len(str1), -1, -1):
if str1.endswith(str2[:i]):
return str2[i:]
return str2
def get_candidate_dis(self,candidate_disease,is_en = False):
max_len = 60
pre_str = '- "{}"'
if is_en:
t2 = '\n\n## Based on the information provided, the likely diagnoses include:\n{}\n\n## Diagnostic reasoning:\n'
post_str = ['.\n',', typically characterized by {}.\n']
else:
t2 = '\n\n## 根据现有信息，病人可能患有的疾病为:\n{}\n\n## 诊断推理:\n'
post_str = ['。\n','，该病常见的症状为{}。\n']
dis_str = ''
for dis in candidate_disease:
if dis['common_symptom'] is not None:
dis_str += pre_str.format(dis['disease']) + post_str[1].format(dis['common_symptom'][:max_len])
else:
dis_str += pre_str.format(dis['disease']) + post_str[0]
return t2.format(dis_str.rstrip())
@torch.no_grad()
def model_genrate_streaming(self, prompt):
inputs = self.tokenizer([prompt], add_special_tokens= False, return_tensors="pt").input_ids
inputs = inputs[:, -self.max_len:]
inputs = inputs.to(self.model.device)
streamer = TextIteratorStreamer(self.tokenizer,skip_prompt=True)
generation_kwargs = dict(input_ids=inputs, streamer=streamer, **self.gen_kwargs)
thread = Thread(target=self.model.generate, kwargs=generation_kwargs)
thread.start()
generated_text = ''
for new_text in streamer:
if self.sep in new_text:
new_text = self.remove_overlap(generated_text,new_text[:-len(self.sep)])
for char in new_text:
generated_text += char
yield char
break
for char in new_text:
generated_text += char
yield char
@torch.no_grad()
def model_genrate(self, prompt):
inputs = self.tokenizer([prompt], add_special_tokens= False, return_tensors="pt").input_ids
inputs = inputs[:, -self.max_len:]
inputs = inputs.to(self.model.device)
outputs = self.model.generate(inputs, **self.gen_kwargs)
res = self.tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
return res
def get_sym_num(self):
return len(self.sym_info['true_syms']) + len(self.sym_info['false_syms'])
def chat(self, query):
prompt = self.generate_prompt(query, self.history)
# print(prompt,flush=True)
generated_text = ''
for char in self.model_genrate_streaming(prompt):
generated_text += char
yield char
cur_generate = generated_text
# For diagnosis.
if 'Enter the diagnostic process, analyzing patient symptoms:' in generated_text or 'Analyzing patient symptoms:' in generated_text or '进入诊断流程，分析病人症状信息:' in generated_text or '总结病人症状信息:' in generated_text:
if 'Enter the diagnostic process, analyzing patient symptoms:' in generated_text or 'Analyzing patient symptoms:' in generated_text:
is_en = True
else:
is_en = False
# Extracting symptoms
p_syms = re.findall(r'(?:\n|,)\s*"([^"]*?)\s*"', generated_text)
f_syms = re.findall(r'(?:没有|No)\s*"([^"]*?)\s*"', generated_text)
self.sym_info = {'true_syms':[],'false_syms':[]}
for sym in p_syms:
if sym not in self.sym_info['true_syms']:
self.sym_info['true_syms'].append(sym)
for sym in f_syms:
if sym not in self.sym_info['false_syms']:
self.sym_info['false_syms'].append(sym)
# print(self.sym_info,flush=True)
if is_en:
candi = self.retriever_en.find_top_k(self.sym_info['true_syms'], self.sym_info['false_syms'])
else:
candi = self.retriever_zh.find_top_k(self.sym_info['true_syms'], self.sym_info['false_syms'])
t2_str = self.get_candidate_dis(candi,is_en)
# print(t2_str,end='',flush = True)
yield t2_str
cur_generate += t2_str
generated_text = ''
for char in self.model_genrate_streaming(prompt+cur_generate):
generated_text += char
yield char
# print('\n',end='',flush=True)
# generated_text += '\n'
try:
if is_en:
pattern = r"## Diagnostic confidence:\s*(.*)"
else:
pattern = r"## 诊断置信度:\s*(.*)"
match = re.search(pattern, generated_text, re.DOTALL)
confidence_distribution = json.loads(match.group(1))
except Exception as e:
print('Error')
print(e)
print(confidence_distribution)
return cur_generate + generated_text
cur_generate += generated_text
max_key = max(confidence_distribution, key=confidence_distribution.get)
max_value = confidence_distribution[max_key]
# if max_value > self.confidence_threshold and (self.get_sym_num(self.sym_info) >= self.min_sym_num or max_value > self.confidence_threshold +0.2 ):
if max_value > self.confidence_threshold:
# 诊断
if is_en:
t4_1 = '\n\n## Diagnosis:\n'
else:
t4_1 = '\n\n## 做出诊断:\n'
# print(t4_1,end='',flush = True)
yield t4_1
cur_generate += t4_1
generated_text = ''
for char in self.model_genrate_streaming(prompt+cur_generate):
generated_text += char
yield char
cur_generate += generated_text
self.history = self.history + [(query, cur_generate)]
else:
# Diagnosis
if is_en:
t4_2 = '\n\n\nInadequate for diagnosis. Ask for symptoms:\n'
else:
t4_2 = '\n\n\n为了更好的诊断疾病，我还需要了解您更多信息，请您回答我的问题:\n'
# print(t4_2,end='',flush = True)
yield t4_2
cur_generate += t4_2
generated_text = ''
for char in self.model_genrate_streaming(prompt+cur_generate):
generated_text += char
yield char
cur_generate += generated_text
self.history = self.history + [(query, cur_generate)]
else:
self.history = self.history + [(query, cur_generate)]
def inference(self, query, history = [], candidate_diseases = None):
self.history = history
prompt = self.generate_prompt(query, self.history)
# print(prompt,flush=True)
generated_text = self.model_genrate(prompt)
cur_generate = generated_text
# For diagnosis.
if 'Enter the diagnostic process, analyzing patient symptoms:' in generated_text or 'Analyzing patient symptoms:' in generated_text or '进入诊断流程，分析病人症状信息:' in generated_text or '总结病人症状信息:' in generated_text:
if 'Enter the diagnostic process, analyzing patient symptoms:' in generated_text or 'Analyzing patient symptoms:' in generated_text:
is_en = True
else:
is_en = False
p_syms = re.findall(r'(?:\n|,)\s*"([^"]*?)\s*"', generated_text)
f_syms = re.findall(r'(?:没有|No)\s*"([^"]*?)\s*"', generated_text)
self.sym_info = {'true_syms':[],'false_syms':[]}
for sym in p_syms:
if sym not in self.sym_info['true_syms']:
self.sym_info['true_syms'].append(sym)
for sym in f_syms:
if sym not in self.sym_info['false_syms']:
self.sym_info['false_syms'].append(sym)
if is_en:
if not candidate_diseases:
candi = self.retriever_en.find_top_k(self.sym_info['true_syms'], self.sym_info['false_syms'])
else:
candi = self.retriever_en.find_top_k_with_candis(self.sym_info['true_syms'], self.sym_info['false_syms'],candidate_diseases)
else:
if not candidate_diseases:
candi = self.retriever_zh.find_top_k(self.sym_info['true_syms'], self.sym_info['false_syms'])
else:
candi = self.retriever_zh.find_top_k_with_candis(self.sym_info['true_syms'], self.sym_info['false_syms'],candidate_diseases)
t2_str = self.get_candidate_dis(candi,is_en)
cur_generate += t2_str
generated_text = self.model_genrate(prompt+cur_generate)
try:
if is_en:
pattern = r"## Diagnostic confidence:\s*(.*)"
else:
pattern = r"## 诊断置信度:\s*(.*)"
match = re.search(pattern, generated_text, re.DOTALL)
confidence_distribution = json.loads(match.group(1))
except Exception as e:
print('Error')
print(e)
print(confidence_distribution)
return cur_generate + generated_text
cur_generate += generated_text
max_key = max(confidence_distribution, key=confidence_distribution.get)
max_value = confidence_distribution[max_key]
if max_value > self.confidence_threshold:
# diagnosis
if is_en:
t4_1 = '\n\n## Diagnosis:\n'
else:
t4_1 = '\n\n## 做出诊断:\n'
# print(t4_1,end='',flush = True)
cur_generate += t4_1
generated_text = self.model_genrate(prompt+cur_generate)
cur_generate += generated_text
self.history = self.history + [(query, cur_generate)]
else:
# Diagnosis
if is_en:
t4_2 = '\n\n\nInadequate for diagnosis. Ask for symptoms:\n'
else:
t4_2 = '\n\n\n为了更好的诊断疾病，我还需要了解您更多信息，请您回答我的问题:\n'
# print(t4_2,end='',flush = True)
cur_generate += t4_2
generated_text = self.model_genrate(prompt+cur_generate)
cur_generate += generated_text
self.history = self.history + [(query, cur_generate)]
return cur_generate, self.history, confidence_distribution
# For other chat.
else:
self.history = self.history + [(query, cur_generate)]
return cur_generate, self.history, {}
def main(args):
os_name = platform.system()
clear_command = 'cls' if os_name == 'Windows' else 'clear'
chatbot = DiagnosisChatbot(args.model_dir,confidence_threshold = 0.5)
pre_string = "DiagnosisGPT: Hello! I'm a large language model designed for disease diagnosis. Please tell me about your symptoms. Type 'clear' to reset the dialogue history, or 'stop' to end the session."
print(pre_string)
while True:
query = input("\nUser: ")
if query == "stop":
break
if query == "clear":
chatbot.history = []
os.system(clear_command)
print(pre_string)
continue
print(f"DiagnosisGPT: ", end="", flush=True)
for char in chatbot.chat(query):
print(char,end='',flush = True)
print('')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--model_dir", type=str, default="")
args = parser.parse_args()
main(args)