File size: 10,763 Bytes
d91766b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
import json
import math
import os
os.environ['CURL_CA_BUNDLE'] = ''
os.environ['REQUESTS_CA_BUNDLE'] = ''
os.environ["HF_ENDPOINT"] = "https://hf-mirror.com"
import random

import re
import numpy as np
import torch
import torch.nn.functional as F
from datasets import load_dataset
from tqdm import tqdm
from transformers import AutoModel, AutoTokenizer
import sys
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from eval.parsers import Parser, is_equiv


def add_gumbel_noise(logits, temperature):
    if temperature == 0:
        return logits
    logits = logits.to(torch.float64)
    noise = torch.rand_like(logits, dtype=torch.float64)
    gumbel_noise = (-torch.log(noise)) ** temperature
    return logits.exp() / gumbel_noise


def get_num_transfer_tokens(mask_index, steps):
    mask_num = mask_index.sum(dim=1, keepdim=True)
    base = mask_num // steps
    remainder = mask_num % steps
    num_transfer_tokens = torch.zeros(mask_num.size(0), steps, device=mask_index.device, dtype=torch.int64) + base
    for i in range(mask_num.size(0)):
        num_transfer_tokens[i, :remainder[i]] += 1
    return num_transfer_tokens


@torch.no_grad()
def generate_trajectory(
    model,
    prompt,
    attention_mask=None,
    steps=16,
    gen_length=128,
    block_length=128,
    temperature=0.0,
    cfg_scale=0.0,
    remasking="low_confidence",
    mask_id=126336,
    logits_eos_inf=False,
    confidence_eos_eot_inf=False,
):
    x = torch.full((prompt.shape[0], prompt.shape[1] + gen_length), mask_id, dtype=torch.long).to(model.device)
    x[:, :prompt.shape[1]] = prompt.clone()

    if attention_mask is not None:
        attention_mask = torch.cat(
            [attention_mask, torch.ones((prompt.shape[0], gen_length), dtype=attention_mask.dtype, device=model.device)],
            dim=-1,
        )

    prompt_index = x != mask_id
    assert gen_length % block_length == 0
    num_blocks = gen_length // block_length
    assert steps % num_blocks == 0
    steps = steps // num_blocks

    trajectory = {}
    step_idx = 0
    trajectory[f"step{step_idx}"] = x[:, prompt.shape[1]:].clone()
    step_idx += 1

    for num_block in range(num_blocks):
        block_mask_index = (
            x[
                :,
                prompt.shape[1] + num_block * block_length : prompt.shape[1] + (num_block + 1) * block_length,
            ]
            == mask_id
        )
        num_transfer_tokens = get_num_transfer_tokens(block_mask_index, steps)
        for i in range(steps):
            mask_index = x == mask_id
            if cfg_scale > 0.0:
                un_x = x.clone()
                un_x[prompt_index] = mask_id
                x_ = torch.cat([x, un_x], dim=0)
                if attention_mask is not None:
                    attention_mask_ = torch.cat([attention_mask, attention_mask], dim=0)
                logits = model(x_, attention_mask=attention_mask_).logits
                logits, un_logits = torch.chunk(logits, 2, dim=0)
                logits = un_logits + (cfg_scale + 1) * (logits - un_logits)
            else:
                logits = model(x, attention_mask=attention_mask).logits

            if logits_eos_inf:
                logits[:, :, 126081] = -torch.inf

            logits_with_noise = add_gumbel_noise(logits, temperature=temperature)
            x0 = torch.argmax(logits_with_noise, dim=-1)

            if confidence_eos_eot_inf:
                logits_with_noise[:, :, 126081] = logits[:, :, 126348] = -torch.inf

            if remasking == "low_confidence":
                p = F.softmax(logits, dim=-1)
                x0_p = torch.squeeze(torch.gather(p, dim=-1, index=torch.unsqueeze(x0, -1)), -1)
            elif remasking == "random":
                x0_p = torch.rand((x0.shape[0], x0.shape[1]), device=x0.device)
            else:
                raise NotImplementedError(remasking)

            x0_p[:, prompt.shape[1] + (num_block + 1) * block_length :] = -np.inf
            x0 = torch.where(mask_index, x0, x)
            confidence = torch.where(mask_index, x0_p, -np.inf)

            transfer_index = torch.zeros_like(x0, dtype=torch.bool, device=x0.device)
            for j in range(confidence.shape[0]):
                k = int(num_transfer_tokens[j, i].item())
                if k > 0:
                    _, select_index = torch.topk(confidence[j], k=k)
                    transfer_index[j, select_index] = True
            x[transfer_index] = x0[transfer_index]
            trajectory[f"step{step_idx}"] = x[:, prompt.shape[1]:].clone()
            step_idx += 1

    return x, trajectory


def build_prompt(tokenizer, question):
    messages = [{"role": "user", "content": question}]
    return tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)


def extract_last_number(text):
    nums = re.findall(r"-?\d{1,3}(?:,\d{3})+(?:\.\d+)?|-?\d+(?:\.\d+)?", text)
    return nums[-1] if nums else None


def extract_pred(text):
    pred = Parser.extract_answer_boxed(text)
    if pred is None:
        pred = Parser.extract_answer_gsm8k(text)
    if pred is None:
        return extract_last_number(text)
    pred_num = extract_last_number(pred)
    return pred_num if pred_num is not None else pred


def extract_example(example, question_key, answer_key):
    if question_key and answer_key:
        return example[question_key], example[answer_key], "custom"
    if "question" in example and "answer" in example:
        return example["question"], example["answer"], "gsm8k"
    if "problem" in example and "solution" in example:
        return example["problem"], example["solution"], "math"
    raise ValueError("Cannot infer question/answer fields, please set --question_key and --answer_key.")


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--model_name", type=str, default="Model Path")
    parser.add_argument("--output_path", type=str, required=True)
    parser.add_argument("--max_new_tokens", type=int, default=256)
    parser.add_argument("--block_length", type=int, default=32)
    parser.add_argument("--steps", type=int, default=256)
    parser.add_argument("--num_samples", type=int, default=1)
    parser.add_argument("--temperature", type=float, default=0.7)
    parser.add_argument("--seed", type=int, default=42)
    parser.add_argument("--limit", type=int, default=0)
    parser.add_argument("--dataset_name", type=str, default="gsm8k")
    parser.add_argument("--dataset_config", type=str, default=None)
    parser.add_argument("--dataset_split", type=str, default="train")
    parser.add_argument("--question_key", type=str, default="")
    parser.add_argument("--answer_key", type=str, default="")
    args = parser.parse_args()

    random.seed(args.seed)
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)

    device = "cuda" if torch.cuda.is_available() else "cpu"
    model = AutoModel.from_pretrained(args.model_name, trust_remote_code=True, torch_dtype=torch.bfloat16).to(device).eval()
    tokenizer = AutoTokenizer.from_pretrained(args.model_name, trust_remote_code=True)
    if tokenizer.padding_side != "left":
        tokenizer.padding_side = "left"
    assert tokenizer.pad_token_id != 126336

    if args.dataset_config is not None:
        ds = load_dataset(args.dataset_name, args.dataset_config, split=args.dataset_split)
    else:
        ds = load_dataset(args.dataset_name, split=args.dataset_split)


    os.makedirs(os.path.dirname(args.output_path), exist_ok=True)
    out_f = open(args.output_path, "w", encoding="utf-8")

    total = len(ds) if args.limit <= 0 else min(args.limit, len(ds))
    solved = 0
    for idx in tqdm(range(total), desc=f"dataset={args.dataset_name}"):
        example = ds[idx]
        raw_q, raw_a, kind = extract_example(example, args.question_key, args.answer_key)
        question = f"Question: {raw_q}\nAnswer:"
        answer = example.get("answer", None)
        if answer is None:
            answer = example.get("solution", None)
        
        if kind == "gsm8k":
            answer_gt = Parser.extract_answer_gsm8k(raw_a)
        elif kind == "math":
            answer_gt = Parser.extract_answer_boxed(raw_a)
        else:
            answer_gt = Parser.extract_answer_boxed(raw_a)

        question = (
            f"Question: {raw_q}\n"
            f"Reference Answer: {raw_a}\n"
            f"After understanding the reference answer, please try to solve this problem "
            f"using your own approach below and output a detailed solution process:"
        )

        prompt = build_prompt(tokenizer, question)
        encoded = tokenizer([prompt], add_special_tokens=False, padding=True, return_tensors="pt")
        input_ids = encoded["input_ids"].to(device)
        attention_mask = encoded["attention_mask"].to(device)
        counted = False

        for sample_id in range(args.num_samples):
            steps = args.steps
            gen_length = args.max_new_tokens
            block_length = args.block_length

            final_x, trajectory = generate_trajectory(
                model,
                input_ids,
                attention_mask=attention_mask,
                steps=steps,
                gen_length=gen_length,
                block_length=block_length,
                temperature=args.temperature,
                cfg_scale=0.0,
                remasking="low_confidence",
            )

            output_text = tokenizer.batch_decode(final_x[:, input_ids.shape[1] :], skip_special_tokens=True)[0]
            pred = extract_pred(output_text)
            print(f"Pred: {pred}")
            if is_equiv(pred, answer_gt):
                if not counted:
                    solved += 1
                    counted = True
                traj_dict = {k: v.squeeze(0).tolist() for k, v in trajectory.items()}
                record = {
                    "dataset_name": args.dataset_name,
                    "index": idx,
                    "sample_id": sample_id,
                    "prompt": prompt,
                    "question": question,
                    "answer_gt": answer_gt,
                    "pred": pred,
                    "prompt_len": int(input_ids.shape[1]),
                    "trajectory": traj_dict,
                    "output_text": output_text,
                    "answer": answer,
                }
                out_f.write(json.dumps(record, ensure_ascii=False) + "\n")
                out_f.flush()

    print(f"Accuracy_any_success: {solved}/{total} = {solved/total*100:.2f}%")
    out_f.write(json.dumps({"summary": "any_success", "solved": solved, "total": total, "accuracy": solved/total}, ensure_ascii=False) + "\n")
    out_f.close()


if __name__ == "__main__":
    main()