-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathMaster Control Program.py
230 lines (181 loc) · 8.67 KB
/
Master Control Program.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
import pandas
import os
import torch
import sklearn
import transformers
from pandas import DataFrame
import datasets
from transformers import TrainingArguments
import peft
# Sentencepiece, protobuf
max_seq_length = 8192
dtype = None
load_in_4bit = True
Dataset = datasets.Dataset
dataset = pandas.read_csv("./Biomarker_dataset_2025_3_12.csv")
#dataset.drop(axis="index")
dataset.drop("RowId", axis="columns")
dataset.drop("NCTId", axis="columns")
dataset.drop("Current", axis="columns")
dataset.drop("Link", axis="columns")
dataset.drop("Field", axis="columns")
dataset = dataset.dropna(subset=['Annotation'])
dataset_Y = dataset.iloc[:,5]
dataset_Y = dataset_Y.reset_index(drop=True)
dataset_X = dataset.iloc[:,-2:]
temp_list = []
for x in range(dataset_X.shape[0]):
temp_list.append(f"<|begin_of_text|>[INST] You are a helpful assistant specializing in medical text annotation. Your task is to analyze outcome measures and descriptions from clinical trials and predict the most medically significant biomarkers. Please list biomarkers in a comma-separated format and do not include any additional information except for the biomarkers.\n" +
"The biomarkers should be labelled in a format as follows - SUPERGROUP_name. For example: CSF_p-tau, BLOOD_Aβ40, BLOOD_p-tau, UD_HbA1c, UD_IGF1, UD_T3, UD_IL-6, UD_TNF-α, URINE_F2-isoP, FDG-PET, fMRI, BLOOD_proteome, CSF_proteome. This list is not exhaustive, nor do all possibilities follow the exact convention listed. [/INST]\n" +
f"Outcome Measure:\n{dataset_X['Outcome Measure']}\nOutcome Description:\n{dataset_X['Outcome Description']}\n")
#print(dataset_Y.index)
temp_list_2 = []
for x in range(dataset_Y.shape[0]):
temp_list_2.append(temp_list[x] + dataset_Y[x] + '<|end_of_text|>')
dataset_X = pandas.DataFrame(temp_list)
dataset_Y = pandas.DataFrame(temp_list_2)
df_concat = pandas.concat([dataset_X, dataset_Y], axis=1)
#print(dataset_X)
tokenizer = transformers.AutoTokenizer.from_pretrained('./Model')
tokenizer.pad_token = tokenizer.eos_token
#def tokenize_function_input(text):
# prompt = f"### Instruction: Please annotate the following clinical trial data using this example format: CSF_p-tau, CSF_t-tau, BLOOD_Aβ40, BLOOD_Aβ42, BLOOD_p-tau\n### Input: {dataset_X['Outcome Measure']}\n{dataset_X['Outcome Description']}"
# return tokenizer(prompt, truncation=True, padding="max_length", max_length=512, return_tensors="pt")
#def tokenize_function_output(text):
# prompt = dataset_Y['Annotation']
#tokenized_dataset = df_concat.apply(tokenize_function, axis=1)
#print(dataset_X)
tokenized_dataset_X = dataset_X.map(lambda x: tokenizer(x, padding='max_length', truncation=True, max_length=512))
tokenized_dataset_Y = dataset_Y.map(lambda x: tokenizer(x, padding='max_length', truncation=True, max_length=512))
TVal1 = []
TVal2 = []
TVal3 = []
for x in range(tokenized_dataset_X[0].shape[0]):
TVal1.append(tokenized_dataset_X[0][x]['input_ids'])
TVal2.append(tokenized_dataset_X[0][x]['attention_mask'])
TVal3.append(tokenized_dataset_Y[0][x]['input_ids'])
#TVal1 = tokenized_dataset_X[0][0].ids
#TVal2 = tokenized_dataset_X[0][0]['attention_mask']
#TVal3 = tokenized_dataset_Y['Annotation'][:]['input_ids']
#print(tokenized_dataset_Y)
#print(dataset_Y)
#tokenized_dataset_X_unpacked = [
# {"input_ids": entry["input_ids"], "attention_mask": entry["attention_mask"]}
# for entry in tokenized_dataset_X
#]
#tokenized_dataset_Y_unpacked = [
# {"input_ids": entry["input_ids"], "attention_mask": entry["attention_mask"]}
# for entry in tokenized_dataset_Y
#]
#tokenized_dataset_X_unpacked = datasets.Dataset.from_pandas(DataFrame(tokenized_dataset_X_unpacked))
#tokenized_dataset_Y_unpacked = datasets.Dataset.from_pandas(DataFrame(tokenized_dataset_Y_unpacked))
#labels = [
# {"labels": entry_Y["input_ids"] + [-100] * (512 - len(entry_Y["input_ids"]))}
# for entry_Y in tokenized_dataset_Y_unpacked
#]
#tokenized_dataset_merged = {
# "input_ids": [entry_X["input_ids"] for entry_X in tokenized_dataset_X_unpacked],
# "attention_mask": [entry_X["attention_mask"] for entry_X in tokenized_dataset_X_unpacked],
# "labels": [label["labels"] for label in labels]
#}
TVal3 = [
[token if token != 128001 else -100 for token in seq]
for seq in TVal3
]
merged_dataset = {
'input_ids': TVal1,
'attention_mask': TVal2,
'labels': TVal3
}
#print(TVal3)
#print(tokenizer.pad_token_id)
#print(merged_dataset)
#print(merged_dataset)
#for x in range(len(tokenized_dataset_merged)):
# print(tokenized_dataset_merged["labels"])
#tokenized_dataset_adjusted = []
#for x in range(len(tokenized_dataset_merged["input_ids"])):
# tokenized_dataset_adjusted.append({
# "input_ids": torch.tensor(tokenized_dataset_merged["input_ids"][x], dtype=torch.long),
# "attention_mask": torch.tensor(tokenized_dataset_merged["attention_mask"][x], dtype=torch.long),
# "labels": torch.tensor(tokenized_dataset_merged["labels"][x], dtype=torch.long),
# })
#for x in range(len(tokenized_dataset_adjusted)):
# print(len(tokenized_dataset_adjusted[x]["input_ids"]))
#print(len(tokenized_dataset_adjusted[0]["input_ids"])) # Should match len(labels)
#print(len(tokenized_dataset_adjusted[0]["labels"]))
#print(tokenized_dataset_merged)
#print(len(tokenized_dataset_adjusted[0]))
#print(len(tokenized_dataset_merged["input_ids"]), type(tokenized_dataset_merged["input_ids"]), len(tokenized_dataset_merged["attention_mask"]), type(tokenized_dataset_merged["attention_mask"]), len(tokenized_dataset_merged["labels"]), type(tokenized_dataset_merged["labels"]))
#print(len(tokenized_dataset_merged["input_ids"][0]))
tokenized_dataset_adjusted = Dataset.from_dict(merged_dataset)
split_dataset = tokenized_dataset_adjusted.train_test_split(test_size=0.2, seed=3000)
#train_loader = torch.utils.data.DataLoader(split_dataset['train'], batch_size=16, shuffle=True, collate_fn=data_collator)
#test_loader = torch.utils.data.DataLoader(split_dataset['test'], batch_size=16, collate_fn=data_collator)
X_train = split_dataset['train']
X_test = split_dataset['test']
#X_train, X_test = sklearn.model_selection.train_test_split(tokenized_dataset_merged, train_size=0.80, test_size=0.20, random_state = 3000)
#X_train = datasets.Dataset.from_pandas(DataFrame(X_train))
#Y_train = datasets.Dataset.from_pandas(DataFrame(Y_train))
#input_length_train = len(X_train["input_ids"])
#labels_train = [-100] * input_length_train + Y_train["input_ids"]
#X_test = datasets.Dataset.from_pandas(DataFrame(X_test))
#Y_test = datasets.Dataset.from_pandas(DataFrame(Y_test))
#input_length_test = len(X_test["input_ids"])
#labels_test = [-100] * input_length_test + Y_test["input_ids"]
#print(X_train)
#torch.distributed.init_process_group(backend="nccl", rank=int(os.environ["SLURM_PROCID"]), world_size=7, init_method="env://")
#model = torch.nn.DataParallel(model)
#model = model.cuda()
#print(model.forward)
#model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[int(os.environ["SLURM_PROCID"])])
#model, _, _, _ = deepspeed.initialize(model=model, config="./ds_config.json")
#device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
#model.to(device)
config = peft.LoraConfig(
task_type=peft.TaskType.CAUSAL_LM,
inference_mode=False,
r=64,
lora_alpha=64,
target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
lora_dropout=0.1,
bias="none",
modules_to_save=["lm_head"],
)
model = transformers.AutoModelForCausalLM.from_pretrained('./Model',local_files_only=True)
#for name, param in model.named_parameters():
# print(name, param.size())
#for name, module in model.named_modules():
# print(name)
model = peft.get_peft_model(model, config)
model.print_trainable_parameters()
data_collator = transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False)
training_args = TrainingArguments(
output_dir="./Model_Output",
per_device_train_batch_size = 2,
per_device_eval_batch_size = 2,
gradient_checkpointing=True,
gradient_accumulation_steps=2,
weight_decay=0.0001,
learning_rate = 0.0003,
num_train_epochs = 10,
eval_strategy="steps",
eval_steps=15,
save_strategy="steps",
save_steps=60,
save_total_limit=2,
load_best_model_at_end=True,
logging_dir='./logs',
deepspeed="./ds_config.json",
#remove_unused_columns=False,
do_train=True,
do_eval=True,
lr_scheduler_type="cosine",
warmup_steps=15
)
# lr_scheduler_type="cosine_with_restarts"
trainer = transformers.trainer.Trainer(model, args=training_args, train_dataset=X_train, eval_dataset=X_test, data_collator=data_collator)
trainer.train()
trainer.save_model("./Model_Output/Final")
tokenizer.save_pretrained("./Model_Output/Final")
torch.distributed.destroy_process_group()