File size: 6,839 Bytes
1904ee8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import shutil
from dataclasses import dataclass, field
from typing import Optional

import torch
from accelerate import Accelerator
from datasets import Dataset, DatasetDict, DatasetInfo, load_dataset
from torch.utils.data import DataLoader
from tqdm import tqdm
from transformers import (
    AutoModelForSequenceClassification,
    AutoTokenizer,
    BitsAndBytesConfig,
    HfArgumentParser,
)


shutil.disk_usage = lambda x: shutil._ntuple_diskusage(1, 1, 1)


@dataclass
class ScriptArguments:
    output_dir: Optional[str] = field(
        default="/home/toolkit/huggingface/openai_summarize_comparison_pseudolabel",
        metadata={"help": "output folder"},
    )
    model_name: Optional[str] = field(default="EleutherAI/pythia-6.9b-deduped", metadata={"help": "the model name"})
    # tokenizer_name: Optional[str] = field(default=None, metadata={"help": "the tokenizer name"})
    dataset_name: Optional[str] = field(
        default="mnoukhov/openai_summarize_comparisons_tldrprompt", metadata={"help": "the dataset name"}
    )
    train_split: Optional[str] = field(default="train[:20]", metadata={"help": "the dataset name"})
    eval_split: Optional[str] = field(default="test[:20]", metadata={"help": "the dataset name"})
    load_in_8bit: Optional[bool] = field(default=False, metadata={"help": "load the model in 8 bits precision"})
    load_in_4bit: Optional[bool] = field(default=False, metadata={"help": "load the model in 4 bits precision"})
    better_transformer: Optional[bool] = field(default=False)
    flash_attention: Optional[bool] = field(default=False)
    batch_size: Optional[int] = field(default=4)
    bf16: Optional[bool] = field(default=False)
    fp16: Optional[bool] = field(default=False)
    fp16_model: Optional[bool] = field(default=False)
    seq_length: Optional[int] = field(default=560, metadata={"help": "Input sequence length"})


def create_and_prepare_model(args):
    if args.load_in_8bit and args.load_in_4bit:
        raise ValueError("You can't load the model in 8 bits and 4 bits at the same time")
    elif args.load_in_8bit or args.load_in_4bit:
        quantization_config = BitsAndBytesConfig(load_in_8bit=args.load_in_8bit, load_in_4bit=args.load_in_4bit)
        device_map = {"": Accelerator().local_process_index}
    else:
        device_map = None
        quantization_config = None

    if args.bf16:
        torch_dtype = torch.bfloat16
    elif args.fp16_model:
        torch_dtype = torch.float16
    else:
        torch_dtype = None

    model = AutoModelForSequenceClassification.from_pretrained(
        args.model_name,
        quantization_config=quantization_config,
        device_map=device_map,
        num_labels=1,
        torch_dtype=torch_dtype,
    )

    if args.better_transformer:
        model.to_bettertransformer()

    tokenizer = AutoTokenizer.from_pretrained(script_args.model_name)
    if getattr(tokenizer, "pad_token", None) is None:
        tokenizer.pad_token = tokenizer.eos_token

    if getattr(model.config, "pad_token_id", None) is None:
        model.config.pad_token_id = model.config.eos_token_id

    return model, tokenizer


def preprocess_function(examples):
    str_chosen = []
    str_rejected = []

    for prompt, chosen, rejected in zip(examples["prompt"], examples["chosen"], examples["rejected"]):
        str_chosen.append(prompt + " " + chosen)
        str_rejected.append(prompt + " " + rejected)

    tokenized_chosen = tokenizer(
        str_chosen, padding=True, truncation=True, max_length=script_args.seq_length, return_tensors="pt"
    )
    tokenized_rejected = tokenizer(
        str_rejected, padding=True, truncation=True, max_length=script_args.seq_length, return_tensors="pt"
    )

    return {
        "input_ids_chosen": tokenized_chosen["input_ids"],
        "attention_mask_chosen": tokenized_chosen["attention_mask"],
        "input_ids_rejected": tokenized_rejected["input_ids"],
        "attention_mask_rejected": tokenized_rejected["attention_mask"],
    }


parser = HfArgumentParser(ScriptArguments)
script_args = parser.parse_args_into_dataclasses()[0]

model, tokenizer = create_and_prepare_model(script_args)
accelerator = Accelerator()

data_splits = [split for split in [script_args.train_split, script_args.eval_split] if split is not None]
relabel_dataset = DatasetDict()

for split in data_splits:
    dataset = load_dataset(script_args.dataset_name, split=split)

    dataloader = DataLoader(dataset, batch_size=script_args.batch_size)

    model, dataloader = accelerator.prepare(model, dataloader)

    model.eval()

    output_dataset = {"prompt": [], "chosen": [], "rejected": []}

    for examples in tqdm(dataloader):
        inputs = preprocess_function(examples)
        with torch.no_grad():
            # if script_args.flash_attention:
            #     with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=False):
            #         output = model(
            #             batch["input_ids"],
            #             attention_mask=batch["attention_mask"],
            #         )

            rewards_chosen = model(
                input_ids=inputs["input_ids_chosen"].to(accelerator.device),
                attention_mask=inputs["attention_mask_chosen"].to(accelerator.device),
            )[0]
            rewards_rejected = model(
                input_ids=inputs["input_ids_rejected"].to(accelerator.device),
                attention_mask=inputs["attention_mask_rejected"].to(accelerator.device),
            )[0]

            pseudolabels = torch.sign(rewards_chosen - rewards_rejected)

            pseudolabels = accelerator.gather(pseudolabels).cpu().numpy()

            if accelerator.is_local_main_process:
                for prompt, init_chosen, init_rejected, label in zip(
                    examples["prompt"], examples["chosen"], examples["rejected"], pseudolabels
                ):
                    output_dataset["prompt"].append(prompt)
                    if label >= 0:
                        output_dataset["chosen"].append(init_chosen)
                        output_dataset["rejected"].append(init_rejected)
                    else:
                        output_dataset["chosen"].append(init_rejected)
                        output_dataset["rejected"].append(init_chosen)

    if accelerator.is_local_main_process:
        ds_info = DatasetInfo(f"{script_args.dataset_name} relabelled with {script_args.model_name}")
        if not split.isalnum():
            split = "".join(c for c in split if c.isalpha())
        relabel_dataset[split] = Dataset.from_dict(output_dataset, split=split, info=ds_info)

if accelerator.is_local_main_process:
    relabel_dataset.save_to_disk(script_args.output_dir)
    relabel_dataset.push_to_hub(os.path.basename(script_args.output_dir))