File size: 16,844 Bytes
13c5606
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import numpy as np
from math import log
from collections import defaultdict
import math
import os
import typing

import torch
import torch.nn.functional as F
import torchmetrics
import transformers
from transformers import logging
logging.set_verbosity_error()

# TODO: for tc

LOG2 = math.log(2)

# For take the total correlation between x0, x1, and y


class TC():
    def __init__(self):
        self.x0 = []
        self.x1 = torch.tensor([])

    def update(self, x0, x1):
        # note that x0, x1, y are all torch tensors
        assert len(x0) == len(x1), "All must have same length"

        # it will be a [8, 1024]
        if self.x1.device != x1.device:
            self.x1 = self.x1.to(x1.device)
        self.x0 += x0
        self.x1 = torch.cat((self.x1, x1), dim=0)

    def entropy_from_counts(self, counts, base=2):
        """
        Compute the empirical entropy H(X) from a dictionary of counts,
        using either base-2 (bits) or base-e (nats).
        """
        total = sum(counts.values())
        if total == 0:
            return 0.0
        entropy = 0.0
        for c in counts.values():
            if c > 0:
                p = c / total
                # Use math.log(p, base) for bits (base=2) or nats (base=np.e)
                entropy -= p * log(p, base)
        return entropy

    def compute_total_correlation_x1(self, x1_group):
        """
        Given an array x1_group of shape (N, 16, 16), where each row is
        one sample of x1, estimate T(X1) = sum_j H(X1_j) - H(X1)
        by naive frequency counting.

        We treat each of the 256 pixels in the 16x16 image as a separate
        discrete variable. We compute:
           H(X1)   via joint frequencies over all 256 pixels
           H(X1_j) via marginal frequencies of each pixel j
        Then T(X1) = sum_j H(X1_j) - H(X1).
        """
        N = x1_group.shape[0]
        if N < 2:
            # With <2 samples, you cannot really estimate correlation reliably.
            # We return 0.0 by convention or skip it entirely.
            return 0.0

        # Flatten each image into 256-dimensional vector:
        # shape becomes (N, 1024)
        flattened = x1_group.reshape(N, -1)
        d = flattened.shape[1]  # should be 1024

        # ---- Joint distribution (all 256 dims) ----#
        # We'll store each flattened row as a tuple, then count frequencies.
        joint_counts = defaultdict(int)
        for row in flattened:
            key = tuple(row)   # row is length-1024
            joint_counts[key] += 1
        H_joint = self.entropy_from_counts(joint_counts, base=2)

        # ---- Marginal distributions (one pixel at a time) ----#
        # We'll compute an entropy for each pixel dimension j.
        marginal_entropies = []
        for j in range(d):
            counts_j = defaultdict(int)
            for row in flattened:
                val_j = row[j].long().item()
                counts_j[val_j] += 1
            H_j = self.entropy_from_counts(counts_j, base=2)
            marginal_entropies.append(H_j)

        sum_marginals = sum(marginal_entropies)

        # ---- Total correlation ----#
        T_val = sum_marginals - H_joint
        return T_val, sum_marginals, H_joint

    def compute_conditional_total_correlation_x1_given_x0y(self, x0, x1):
        """
        1) Group all samples by unique x0[i] pair.
           - Here x0[i] is a [1024] tensor
        2) Collect the corresponding x1[i] arrays for each group.
        3) For each group, estimate T(X1) by naive frequency counting
           (thus approximating T(X1 | x0=x0_val, y=y_val)).

        Returns a dict:  (x0_bytes) -> estimated total correlation in bits.
        """

        # Safety checks:
        assert len(x0) == len(x1), "All must have same length"
        N = len(x0)

        # Group x1 by unique x0
        groups = defaultdict(list)
        for i in range(N):
            # We need a hashable key for x0[i], which is shape (1028)
            # Convert to bytes, or a tuple if you prefer
            x0_key = x0[i]
            key = x0_key
            groups[key].append(x1[i])

        # Compute total correlation for each group
        results = {}
        marginals = {}
        joints = {}
        for key, x0_x1_tup in groups.items():
            # x1_list is a tensor, shape each (1024)
            # print(len(x1_list))
            # shape = (num_samples_for_this_group, 1024)
            group = torch.stack(x0_x1_tup, dim=0)
            T_val, sum_marginals, H_joint = self.compute_total_correlation_x1(
                group)
            results[key] = T_val
            marginals[key] = sum_marginals
            joints[key] = H_joint

        return results, joints, marginals

    def compute(self):
        """
        Compute the total correlation of x1 given x0 and y.
        """
        tc_results, joints, marginals = self.compute_conditional_total_correlation_x1_given_x0y(
            self.x0, self.x1)

        avg_tc = np.mean(list(tc_results.values()))
        avg_joints = np.mean(list(joints.values()))
        avg_marginals = np.mean(list(marginals.values()))

        return avg_tc, avg_joints, avg_marginals


class NLL(torchmetrics.aggregation.MeanMetric):
    def update(self,
               value: typing.Union[float, torch.Tensor],
               weight: typing.Union[float, torch.Tensor] = 1.0) -> None:
        """Update state with data.

        Args:
          value: Either a float or tensor containing data.
            Additional tensor dimensions will be flattened
          weight: Either a float or tensor containing weights
            for calculating the average. Shape of weight should
            be able to broadcast with the shape of `value`.
            Default to `1.0` corresponding to simple harmonic
            average.
        """
        # broadcast weight to value shape
        if not isinstance(value, torch.Tensor):
            value = torch.as_tensor(value, dtype=self.dtype,
                                    device=self.device)
        if (weight is not None
                and not isinstance(weight, torch.Tensor)):
            weight = torch.as_tensor(weight,
                                     dtype=self.dtype,
                                     device=self.device)
        weight = torch.broadcast_to(weight, value.shape)
        value, weight = self._cast_and_nan_check_input(value,
                                                       weight)

        if value.numel() == 0:
            return
        self.mean_value += value.sum()
        self.weight += weight.sum()


class BPD(NLL):
    def compute(self) -> torch.Tensor:
        """Computes the bits per dimension.

        Returns:
          bpd
        """
        return self.mean_value / self.weight / LOG2


class Perplexity(NLL):
    def compute(self) -> torch.Tensor:
        """Computes the Perplexity.

        Returns:
         Perplexity
        """
        return torch.exp(self.mean_value / self.weight)


class Metrics:
    def __init__(self, gen_ppl_eval_model_name_or_path=None,
                 eval_ppl_batch_size=None) -> None:
        metrics = torchmetrics.MetricCollection({
            'nll': NLL(), 'bpd': BPD(), 'ppl': Perplexity()})
        metrics.set_dtype(torch.float64)
        self.train_nlls = metrics.clone(prefix='train/')
        self.train_aux = BPD()
        self.valid_nlls = metrics.clone(prefix='val/')
        self.valid_aux = BPD()
        self.gen_ppl = Perplexity()
        self.sample_entropy = torchmetrics.aggregation.MeanMetric()
        self.unique_token_count = 0
        self.tc = TC()
        self.eval_ppl_batch_size = eval_ppl_batch_size
        self.gen_ppl_eval_model_name_or_path = gen_ppl_eval_model_name_or_path
        self.tokenizer = transformers.AutoTokenizer.\
            from_pretrained(gen_ppl_eval_model_name_or_path)
        if self.tokenizer.pad_token is None:
            self.tokenizer.pad_token = self.tokenizer.eos_token
            self.tokenizer.pad_token_id = self.tokenizer.eos_token_id

    def to(self, *args, **kwargs):
        self.gen_ppl = self.gen_ppl.to(*args, **kwargs)
        self.sample_entropy = self.sample_entropy.to(*args, **kwargs)
        self.train_nlls = self.train_nlls.to(*args, **kwargs)
        self.train_aux = self.train_aux.to(*args, **kwargs)
        self.valid_nlls = self.valid_nlls.to(*args, **kwargs)
        self.valid_aux = self.valid_aux.to(*args, **kwargs)

    def reset(self):
        self.gen_ppl.reset()
        self.sample_entropy.reset()
        self.train_nlls.reset()
        self.train_aux.reset()
        self.valid_nlls.reset()
        self.valid_aux.reset()

    def update_train(self, nll, aux_loss, num_tokens):
        self.train_nlls.update(nll, num_tokens)
        self.train_aux.update(aux_loss, num_tokens)

    def update_valid(self, nll, aux_loss, num_tokens):
        self.valid_nlls.update(nll, num_tokens)
        self.valid_aux.update(aux_loss, num_tokens)

    @torch.no_grad()
    def _eval_retokenize(self, text_samples, max_length,
                         device):
        """Retokenizes samples for the eval model.

        Args:
            text_samples: List of sentences generated by the model.
        Returns:
            samples: Samples re-tokenized for the eval model
            attn_mask: Attention mask for the eval model
            eval_context_size: Size of the context for the eval model
        """
        if 'llama2' in self.gen_ppl_eval_model_name_or_path.lower():
            tokenizer_kwargs = {
                'text_samples': text_samples,
                'return_tensors': 'pt',
                'return_token_type_ids': False,
                'return_attention_mask': True,
                'truncation': True,
                'padding': True,
                'max_length': max_length,
            }
            eval_context_size = 4096
        elif 'llama3' in self.gen_ppl_eval_model_name_or_path.lower():
            tokenizer_kwargs = {
                'text': text_samples,
                'return_tensors': 'pt',
                'padding': True,
            }
            eval_context_size = 8192
        else:
            tokenizer_kwargs = {
                'return_tensors': 'pt',
                'return_token_type_ids': False,
                'return_attention_mask': True,
                'truncation': True,
                'padding': True,
                'max_length': max_length,
            }
            eval_context_size = 1024
        samples = self.tokenizer(text_samples,
                                 **tokenizer_kwargs)
        attn_mask = samples['attention_mask']
        samples = samples['input_ids']
        if 'llama' not in self.gen_ppl_eval_model_name_or_path.lower():
            attn_mask = attn_mask.to(device)
            samples = samples.to(device)
        return samples, attn_mask, eval_context_size

    @torch.no_grad()
    def record_entropy(self, tokens):
        for sample in tokens:
            _, counts = torch.unique(
                sample, return_counts=True, sorted=False)
            entropy = torch.special.entr(
                counts.float() / counts.sum()).sum().item()
            self.sample_entropy.update(entropy)

    @torch.no_grad()
    def record_unique_tokens(self, samples):
        """Record the count of unique tokens across all samples.
        
        Args:
            samples: torch.Tensor of shape (batch_size, seq_length)
        
        Returns:
            int: Number of unique tokens in the samples
        """
        unique_tokens = torch.unique(samples.flatten())
        self.unique_token_count = len(unique_tokens)
        return self.unique_token_count

    def reset_unique_tokens(self):
        """Reset unique token count."""
        self.unique_token_count = 0

    @torch.no_grad()
    def record_generative_perplexity(
            self,
            text_samples: typing.List[str],
            max_length: int,
            retokenize: bool = True,
            device='cuda') -> None:

        os.environ['TOKENIZERS_PARALLELISM'] = 'false'
        if 'llama' not in self.gen_ppl_eval_model_name_or_path:
            eval_model = transformers.AutoModelForCausalLM.from_pretrained(
                self.gen_ppl_eval_model_name_or_path).eval()
            eval_model = eval_model.to(device)
            # Re-tokenize using eval model's tokenizer
            if retokenize:
                (samples, attn_mask,
                 eval_context_size) = self._eval_retokenize(
                    text_samples, max_length=max_length, device=device)
            else:
                samples = text_samples
                attn_mask = torch.ones(samples.shape).to(device)
                eval_context_size = samples.shape[-1]
            batch_size = min(self.eval_ppl_batch_size,
                             samples.shape[0])
            num_batches = samples.shape[0] // batch_size
            for i in range(num_batches):
                _samples = torch.split(
                    samples[i * batch_size: (i + 1) * batch_size],
                    eval_context_size,
                    dim=-1)
                _attn_mask = torch.split(
                    attn_mask[i * batch_size: (i + 1) * batch_size],
                    eval_context_size,
                    dim=-1)
                for (sample_chunk, attn_mask_chunk) in zip(_samples,
                                                           _attn_mask):
                    logits = eval_model(sample_chunk.to(device),
                                        attention_mask=attn_mask_chunk.to(device))
                    logits = logits[0].transpose(-1, -2)
                    nlls = F.cross_entropy(logits[..., :-1],
                                           sample_chunk[..., 1:],
                                           reduction='none')
                    first_eos = (
                        sample_chunk
                        == self.tokenizer.eos_token_id).cumsum(-1) == 1
                    token_mask = sample_chunk != self.tokenizer.eos_token_id
                    valid_tokens = first_eos[..., 1:] + token_mask[..., 1:]
                    self.gen_ppl.update(nlls * valid_tokens, valid_tokens)
        else:
            eval_model = transformers.AutoModelForCausalLM.from_pretrained(
                self.gen_ppl_eval_model_name_or_path,
                torch_dtype=torch.bfloat16).eval()
            eval_model = eval_model.to(device)
            # Re-tokenize using eval model's tokenizer
            tokenizer_llama = transformers.AutoTokenizer.from_pretrained(
                self.gen_ppl_eval_model_name_or_path)
            if tokenizer_llama.pad_token is None:
                tokenizer_llama.pad_token = tokenizer_llama.eos_token
                tokenizer_llama.pad_token_id = tokenizer_llama.eos_token_id
            tokenizer_gpt = transformers.AutoTokenizer.from_pretrained(
                'gpt2')
            if tokenizer_gpt.pad_token is None:
                tokenizer_gpt.pad_token = tokenizer_gpt.eos_token
                tokenizer_gpt.pad_token_id = tokenizer_gpt.eos_token_id
            num_samples = len(text_samples)
            batch_size = min(16, num_samples)
            num_batches = num_samples // batch_size

            with torch.inference_mode():
                # 1. divide into batches of 16
                # 2. encode each batch
                # 3. eval each batch of 16
                for i in range(num_batches):
                    batch_text_samples = text_samples[i *
                                                      batch_size:(i+1)*batch_size]
                    encoded_inputs = tokenizer_llama(
                        batch_text_samples,
                        return_tensors="pt",
                        padding=True,
                    )

                    input_ids = encoded_inputs['input_ids'].to(device)
                    attention_mask = encoded_inputs['attention_mask'].to(
                        device)

                    labels = input_ids.clone()
                    labels[labels == tokenizer_llama.pad_token_id] = 50000
                    labels = labels.to(device)

                    outputs = eval_model(
                        input_ids=input_ids, attention_mask=attention_mask, labels=labels)

                    llama_logits = outputs.logits

                    logits = llama_logits.transpose(-1, -2)
                    nlls = F.cross_entropy(logits[..., :-1],
                                           labels[..., 1:],
                                           reduction='none')
                    valid_tokens = attention_mask[..., 1:].bool()
                    self.gen_ppl.update(nlls * valid_tokens, valid_tokens)

    @torch.no_grad()
    def record_tc(self, noise_index, sample):
        self.tc.update(noise_index, sample)