-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathutils_llm_inference.py
493 lines (392 loc) · 20.4 KB
/
utils_llm_inference.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
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT License.
# import openai
from baseline.llm.promptbench.config import LABEL_TO_ID,LABEL_SET
from tqdm import tqdm
import pdb
import utils
import copy
import torch
import gc
"""
This clss implements the inference of the model (including create the model).
"""
class Inference(object):
def __init__(self,args):
self.error_analysis = False
self.args = copy.copy(args)
self.device = args.device
self.model_name = args.model_name
if self.args.dataset == 'sst':
self.args.dataset = 'sst2'
self.create_model()
def create_model(self):
"""
ChatGPT is a special case, we use the openai api to create the model.
"""
if self.model_name not in ['chatgpt', 'gpt4']:
import torch
import os
"""
Here you can add you own model.
"""
if self.model_name == 'google/flan-t5-large':
from transformers import T5Tokenizer, T5ForConditionalGeneration
self.tokenizer = T5Tokenizer.from_pretrained(self.model_name)
self.model = T5ForConditionalGeneration.from_pretrained(self.model_name).to(self.device)
elif self.model_name == 'EleutherAI/gpt-neox-20b':
from transformers import GPTNeoXForCausalLM, GPTNeoXTokenizerFast
self.tokenizer = GPTNeoXTokenizerFast.from_pretrained(self.model_name, device_map="auto")
self.model = GPTNeoXForCausalLM.from_pretrained(self.model_name, device_map="auto", torch_dtype=torch.float16)
elif 'llama' in self.model_name:
from transformers import AutoModelForCausalLM,AutoTokenizer
self.model = AutoModelForCausalLM.from_pretrained(
self.model_name,
cache_dir = "/home/data_shared/llama/",
torch_dtype=torch.float16,
trust_remote_code=True,
low_cpu_mem_usage=True,
use_cache=False).to(self.device)
self.tokenizer = AutoTokenizer.from_pretrained(
self.model_name,
cache_dir = "/home/data_shared/llama/",
trust_remote_code=True,
use_fast=False)
self.tokenizer.pad_token = self.tokenizer.unk_token
self.tokenizer.padding_side = 'left' #this is important! not right!
elif self.model_name.lower() in ["vicuna-13b", "vicuna-13b-v1.3"]:
from transformers import AutoModelForCausalLM, AutoTokenizer
model_dir = os.path.join(self.args.model_dir, self.model_name)
self.tokenizer = AutoTokenizer.from_pretrained(model_dir, device_map="auto", use_fast=False)
self.model = AutoModelForCausalLM.from_pretrained(model_dir, device_map="auto", torch_dtype=torch.float16)
elif 'vicuna-7b-v1.3' in self.model_name:
from transformers import AutoModelForCausalLM,AutoTokenizer
cache_dir = "/home/data_shared/vicuna/"
self.model = AutoModelForCausalLM.from_pretrained(
self.model_name,
cache_dir = cache_dir,
torch_dtype=torch.float16,
trust_remote_code=True,
low_cpu_mem_usage=True,
use_cache=False
).to(self.device)
self.tokenizer = AutoTokenizer.from_pretrained(
self.model_name,
cache_dir = cache_dir,
trust_remote_code=True,
use_fast=False)
self.tokenizer.pad_token = self.tokenizer.unk_token
# self.tokenizer.padding_side = 'left' #this is important! not right! #TODO, why gcg didn't have this
elif 'guanaco' in self.model_name:
from transformers import AutoModelForCausalLM,AutoTokenizer
cache_dir = "/home/data_shared/guanaco/"
self.model = AutoModelForCausalLM.from_pretrained(
self.model_name,
cache_dir = cache_dir,
torch_dtype=torch.float16,
trust_remote_code=True,
low_cpu_mem_usage=True,
use_cache=False
).to(self.device)
self.tokenizer = AutoTokenizer.from_pretrained(
self.model_name,
cache_dir = cache_dir,
trust_remote_code=True,
use_fast=False)
self.tokenizer.pad_token = self.tokenizer.unk_token
#adapt from
self.tokenizer.eos_token_id = 2
self.tokenizer.unk_token_id = 0
if not self.tokenizer.pad_token:
self.tokenizer.pad_token = self.tokenizer.eos_token
elif self.model_name == "google/flan-ul2":
from transformers import T5ForConditionalGeneration, AutoTokenizer
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
self.model = T5ForConditionalGeneration.from_pretrained(self.model_name, torch_dtype=torch.bfloat16, device_map="auto")
elif self.model_name == "tiiuae/falcon-40b-instruct":
from transformers import AutoTokenizer, AutoModelForCausalLM
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
self.model = AutoModelForCausalLM.from_pretrained(self.model_name, torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto",)
elif self.model_name == "cerebras/Cerebras-GPT-13B":
from transformers import AutoTokenizer, AutoModelForCausalLM
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name, device_map="auto")
self.model = AutoModelForCausalLM.from_pretrained(self.model_name, device_map="auto", torch_dtype=torch.float16)
elif self.model_name == "databricks/dolly-v1-6b":
from transformers import AutoModelForCausalLM, AutoTokenizer
self.tokenizer = AutoTokenizer.from_pretrained("databricks/dolly-v1-6b", device_map="auto", padding_side="left")
self.model = AutoModelForCausalLM.from_pretrained("databricks/dolly-v1-6b", device_map="auto", torch_dtype=torch.float16)
else:
raise NotImplementedError("The model is not implemented!")
def process_input(self, prompt, raw_data):
if self.args.dataset in ["cola", "sst2", "mrpc", "qqp", "mnli", "qnli", "rte", "wnli"]:
return self._process_cls_input(prompt, raw_data)
elif self.args.dataset == "mmlu":
return self._process_qa_input(prompt, raw_data)
elif self.args.dataset == "squad_v2":
return self._process_squad_v2_input(prompt, raw_data)
elif self.args.dataset in ['iwslt', 'un_multi']:
return self._process_trans_input(prompt, raw_data)
elif self.args.dataset == 'math':
return self._process_math_input(prompt, raw_data)
elif self.args.dataset == 'bool_logic':
return self._process_bool_logic_input(prompt, raw_data)
elif self.args.dataset == 'valid_parentheses':
return self._process_valid_parentheses_input(prompt, raw_data)
else:
raise NotImplementedError("The dataset is not implemented!")
def process_pred(self, pred):
if self.args.dataset in ["cola", "sst2", "mrpc", "qqp", "mnli", "qnli", "rte", "wnli"]:
return self._process_cls_pred(pred)
elif self.args.dataset == "mmlu":
return self._process_qa_pred(pred)
elif self.args.dataset == "squad_v2":
return self._process_squad_v2_pred(pred)
elif self.args.dataset in ['iwslt', 'un_multi']:
return self._process_trans_pred(pred)
elif self.args.dataset == 'math':
return self._process_math_pred(pred)
elif self.args.dataset == 'bool_logic':
return self._process_bool_logic_pred(pred)
elif self.args.dataset == 'valid_parentheses':
return self._process_valid_parentheses_pred(pred)
else:
raise NotImplementedError("The dataset is not implemented!")
def __call__(self, SS, debug=False):
preds = []
for S in SS:
if isinstance(S, tuple) or isinstance(S, list):
premise = S[0]
S = S[1]
else:
premise = None
llm_input_processor = utils.llm_input_processor(self.args.dataset,premise,self.args.model_name)
S = llm_input_processor.addprompt(S)
if self.model_name in ["chatgpt", "gpt4"]:
pred = self.predict_by_openai_api(S)
else:
pred= self.predict_by_local_inference(S,debug)
if len(SS)!=1:
pred = pred.detach().cpu()
preds.append(pred)
return preds
def predict_by_openai_api(self, S):
raise NotImplementedError
def predict_by_local_inference(self, S,debug=False):
"return logit with size (batch,vocabularysize)"
logits = self.pred_by_generation(S)
if debug:
print('orig_S:', S)
idx = torch.argmax(logits)
raw_pred = self.tokenizer.decode(idx, skip_special_tokens=True)
print('raw_pred:',raw_pred)
return logits
def token_to_logit(self,input_ids,attention_mask = None):
'''
return logit with size (batch,len,vocabularysize)
'''
if 'llama' in self.model_name or 'vicuna-7b-v1.3' in self.model_name:
pred = self.model(input_ids =input_ids, attention_mask =attention_mask).logits
elif 't5' in self.model_name:
inputs_embeds = self.model.encoder.embed_tokens(input_ids)
pred = self.emb_to_logit(inputs_embeds,attention_mask)
else:
raise NotImplementedError
return pred
def emb_to_logit(self,inputs_embeds,attention_mask=None):
'''
return logit with size (batch,len,vocabularysize)
'''
if 'llama' in self.model_name or 'vicuna-7b-v1.3' in self.model_name:
out = self.model(inputs_embeds = inputs_embeds).logits
elif 't5' in self.model_name:
decoder_input_ids = torch.ones(inputs_embeds.shape[0], 1,dtype=torch.int) \
* self.model.config.decoder_start_token_id
decoder_inputs_embeds = self.model.decoder.embed_tokens(decoder_input_ids.to(self.device))
out = self.model(inputs_embeds=inputs_embeds,
decoder_inputs_embeds=decoder_inputs_embeds,
attention_mask = attention_mask).logits
# TODO check attention mask here
else:
raise NotImplementedError
return out
def pred_by_generation_sentence(self, inputs,generate_len):
'''
return sentence in jailbreak
'''
gen_config = self.model.generation_config
gen_config.max_new_tokens = generate_len # self.args.generate_len
if isinstance(inputs[0], str):
# str to str
input_ids = self.tokenizer(inputs, return_tensors="pt").input_ids.to(self.device)
outputs = self.model.generate(input_ids,
generation_config=gen_config,
pad_token_id=self.tokenizer.pad_token_id)
out = self.tokenizer.decode(outputs[0][len(input_ids[0]):], skip_special_tokens=True)
return out
# else:
# raise NotImplementedError
elif torch.is_tensor(inputs):
# emb to str
outputs = self.model.generate(inputs_embeds = inputs,
generation_config=gen_config,
pad_token_id=self.tokenizer.pad_token_id)
out = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
return out
# else:
# raise NotImplementedError
else:
raise NotImplementedError
def pred_by_generation(self, input_text):
'''
return logit with size (batch,vocabularysize)
'''
out = 'error!'
input_ids = self.tokenizer(input_text, return_tensors="pt").input_ids.to(self.device)
if 't5' in self.model_name or 'ul2' in self.model_name:
inputs_embeds = self.model.encoder.embed_tokens(input_ids)
out = self.emb_to_logit(inputs_embeds)
out = out[0,-1,:]
return out
elif self.model_name == 'EleutherAI/gpt-neox-20b':
outputs = self.model.generate(input_ids,
# do_sample=True,
temperature=0.00001,
# max_length=50,
max_new_tokens=self.args.generate_len,
early_stopping=True,
pad_token_id=self.tokenizer.eos_token_id)
out = self.tokenizer.decode(outputs[0])
elif self.model_name == "facebook/opt-66b":
outputs = self.model.generate(input_ids)
out = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
elif 'llama' in self.model_name or 'vicuna-7b-v1.3' in self.model_name:
out = self.model(input_ids).logits
out = out[0,-1,:]
return out
elif self.model_name in ['databricks/dolly-v1-6b', 'cerebras/Cerebras-GPT-13B']:
outputs = self.model.generate(input_ids,
temperature=0,
max_new_tokens=self.args.generate_len,
pad_token_id=self.tokenizer.eos_token_id,
early_stopping=True)
out = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
elif self.model_name == "tiiuae/falcon-40b-instruct":
outputs = self.model.generate(input_ids,
temperature=0,
max_new_tokens=self.args.generate_len,
pad_token_id=self.tokenizer.eos_token_id,
early_stopping=True)
out = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
return out
def _process_valid_parentheses_input(self, prompt, raw_data):
question, label = raw_data['question'], raw_data['answer']
input_text = prompt + '\n'
if self.args.shot > 0:
input_text += "\n"+self.args.data.get_few_shot_examples(raw_data['task'])
input_text += ("Question: " + question + '\nAnswer: ')
return input_text, label
def _process_bool_logic_input(self, prompt, raw_data):
question, label = raw_data['question'], raw_data['answer']
input_text = prompt + '\n'
if self.args.shot > 0:
input_text += "\n"+self.args.data.get_few_shot_examples(raw_data['task'])
input_text += ("Question: " + question + '\nAnswer: ')
return input_text, label
def _process_math_input(self, prompt, raw_data):
from config import MATH_QUESTION_TYPES
question_type, question, label = MATH_QUESTION_TYPES[raw_data['task']], raw_data['question'], raw_data['answer']
input_text = prompt.format(question_type) + '\n'
if self.args.shot > 0:
input_text += "\n"+self.args.data.get_few_shot_examples(raw_data['task'])
input_text += ("Question: " + question + '\nAnswer: ')
return input_text, label
def _process_trans_input(self, prompt, raw_data):
from config import LANGUAGES
source, target, task = raw_data['source'], raw_data['target'], raw_data['task']
src_lang, des_lang = task.split('-')
input_text = prompt.format(LANGUAGES[src_lang], LANGUAGES[des_lang]) + '\n'
if self.args.shot > 0:
input_text += "\n"+self.args.data.get_few_shot_examples(task)
input_text += (source + '\nAnswer: ')
return input_text, target
def _process_squad_v2_input(self, prompt, raw_data):
id, content = raw_data["id"], raw_data["content"]
input_text = prompt
if self.args.shot > 0:
input_text += "\n"+self.args.data.get_few_shot_examples(self.args.dataset)
input_text += (content + "Answer: ")
return input_text, id
def _process_qa_input(self, prompt, raw_data):
task, content = raw_data["task"], raw_data["content"]
label = raw_data["label"]
input_text = prompt.format(task) + "\n"
if self.args.shot > 0:
input_text += "\n"+self.args.data.get_few_shot_examples(task.replace(" ", "_"))
input_text += content + "\nAnswer: "
return input_text, label
def _process_cls_input(self, prompt, raw_data):
content = raw_data["content"]
label = raw_data["label"]
input_text = prompt
if self.args.shot > 0:
few_shot_examples = self.args.data.get_few_shot_examples(self.args.dataset)
input_text += "\n"+few_shot_examples
if self.args.dataset == "sst2" or self.args.dataset == "cola":
input_text += "Sentence: "
input_text += (content + ' Answer: ')
return input_text, label
def _process_bool_logic_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
return pred
def _process_valid_parentheses_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
return pred
def _process_math_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
return pred
def _process_trans_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
return pred
def _process_squad_v2_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
return pred
def _process_cls_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
if self.args.dataset!= 'sst2':
pred = pred.split(" ")[-1]
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
if pred in LABEL_SET[self.args.dataset]:
pred = LABEL_TO_ID[self.args.dataset][pred]
else:
print("The predicted label: '{}' is not in label set.".format(pred))
pred = -1
return pred
def _process_qa_pred(self, raw_pred):
pred = raw_pred.lower()
pred = pred.replace("<pad>", "")
pred = pred.replace("</s>", "")
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
pred = pred.split(" ")[-1]
pred = pred.strip(",._\"\'-+=!?()&^%$#@:\\|\{\}[]<>/`\n\t\r\v\f ")
if pred not in LABEL_SET[self.args.dataset]:
print("The predicted label: '{}' is not in label set.".format(pred))
pred = 'no_answer'
return pred