File size: 29,418 Bytes
eb1aec4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
import json
import logging
import math
import os
import time

import numpy as np
import torch
import torch.nn.functional as F
from torch.nn.parallel.distributed import DistributedDataParallel

try:
    import wandb
except ImportError:
    wandb = None

from open_clip import get_input_dtype, CLIP, CustomTextCLIP
from open_clip_train.distributed import is_master
from open_clip_train.zero_shot import zero_shot_eval
from open_clip_train.precision import get_autocast

from torchvision.ops import roi_align
from torchvision.transforms import RandomResizedCrop

class AverageMeter(object):
    """Computes and stores the average and current value"""

    def __init__(self):
        self.reset()
        self.total_sum =0

    def reset(self):
        self.val = 0
        self.avg = 0
        self.sum = 0
        self.count = 0

    def update(self, val, n=1):
        self.val = val
        self.sum += val * n
        self.count += n
        self.avg = self.sum / self.count
        self.total_sum += val * n


def postprocess_clip_output(model_out):
    return {
        "image_features": model_out[0],
        "text_features": model_out[1],
        "logit_scale": model_out[2]
    }


def unwrap_model(model):
    if hasattr(model, 'module'):
        return model.module
    else:
        return model


def backward(total_loss, scaler):
    if scaler is not None:
        scaler.scale(total_loss).backward()
    else:
        total_loss.backward()


def filter_batch_data(batch, used_losses, input_dtype, device):
    selected = {}

    selected["global_image"] = batch["global_image"].to(device=device, dtype=input_dtype, non_blocking=True)
    if 'global_itc' in used_losses:
        selected["global_text"] = batch["global_text"].to(device=device, non_blocking=True)
    if 'local_itc' in used_losses:
        selected["local_images"] = batch["local_images"].to(device=device, dtype=input_dtype, non_blocking=True)
        selected["local_texts"] = batch["local_texts"].to(device=device, non_blocking=True)
        selected["boxes"] = batch["boxes"].to(device=device, non_blocking=True)
        if 'local_categories' in batch:
            selected["local_categories"] = batch["local_categories"].to(device=device)
    if 'distill' in used_losses and 'subset_images' in batch:
        selected["subset_images"] = batch["subset_images"].to(device=device, dtype=input_dtype, non_blocking=True)
        selected["subset_boxes"] = batch["subset_boxes"].to(device=device, non_blocking=True)

    return selected


def get_batch_data(batch, used_losses, input_dtype, device):
    images, global_caption = batch

    global_image = torch.stack([item["global"] for item in images], dim=0)  # (B, C, H, W)
    local_imgs = torch.stack([item["locals"] for item in images], dim=0)  # (B, N, C, H_l, W_l)
    bboxes = torch.stack([item["bboxes"] for item in images], dim=0)  # (B, N, 5)

    selected = {}
    selected["global_image"] = global_image.to(device=device, dtype=input_dtype, non_blocking=True)
    selected["global_text"] = global_caption.to(device=device, non_blocking=True)
    if 'distill' in used_losses:
        selected["subset_images"] = local_imgs.to(device=device, dtype=input_dtype, non_blocking=True)
        selected["subset_boxes"] = bboxes.to(device=device, non_blocking=True)

    return selected


def train_one_epoch(model, teacher, method, data, loss, mpcl_loss, epoch, optimizer, scaler, scheduler, args, tb_writer=None):
    device = torch.device(args.device)
    autocast = get_autocast(args.precision, device_type=device.type)
    input_dtype = get_input_dtype(args.precision)

    model.train()
    if teacher: teacher.eval()

    data['train'].set_epoch(epoch)  # set epoch in process safe manner via sampler or shared_epoch
    dataloader = data['train'].dataloader
    num_batches_per_epoch = dataloader.num_batches // args.accum_freq
    sample_digits = math.ceil(math.log(dataloader.num_samples + 1, 10))

    if args.accum_freq > 1:
        accum_images, accum_texts, accum_features = [], [], {}

    losses_m = {}
    batch_time_m = AverageMeter()
    data_time_m = AverageMeter()
    end = time.time()
    for i, batch in enumerate(dataloader):
        i_accum = i // args.accum_freq      # Accumulated gradient updates at batch (i)
        step = num_batches_per_epoch * epoch + i_accum      # Total steps for whole training

        if not args.skip_scheduler:
            scheduler(step)

        data_time_m.update(time.time() - end)
        optimizer.zero_grad()

        if method.startswith('farslip'):
            assert args.accum_freq == 1, "accum freq disabled"

            # ========== Obtain input batches ==========
            if method == 'farslip1':
                batch = get_batch_data(batch, input_dtype=input_dtype, used_losses=args.loss_type, device=device)
            elif method == 'farslip2':
                batch = filter_batch_data(batch, input_dtype=input_dtype, used_losses=args.loss_type, device=device)
            else: raise ValueError(f"Unknown args.method: {args.method}")

            # ========== Obtain VALID local inputs (images, boxes, categories) ==========
            if "local_itc" in args.loss_type:

                bboxes = batch["boxes"]  # (B, num_boxes, 5)
                local_images = batch["local_images"]  # (B, num_boxes, C, H, W)
                local_texts = batch["local_texts"]  # (B, num_boxes, context_length)

                local_texts_list = []
                bboxes_list, local_imgs_list = [], []
                local_categories_list = []  # (B, num_boxes)

                has_local_categories = 'local_categories' in batch
                if has_local_categories: local_categories = batch['local_categories']

                for idx in range(len(bboxes)):
                    bbox = bboxes[idx]
                    local_img = local_images[idx]
                    validity = bbox[:, -1] == 1
                    bboxes_list.append(bbox[validity, :4])
                    local_imgs_list.append(local_img[validity])
                    local_text = local_texts[idx]
                    local_texts_list.append(local_text[validity])

                    if has_local_categories: local_categories_list.append(local_categories[idx][validity])

                local_texts = torch.cat(local_texts_list)  # (valid_objects, context_length)
                batch["local_texts"] = local_texts

                local_images = torch.cat(local_imgs_list)  # (valid_objects, C, H, W)
                batch['local_images'] = local_images

                if has_local_categories: local_categories = torch.cat(local_categories_list)

            if "distill" in args.loss_type and "subset_images" in batch:

                batch['subset_images'] = batch['subset_images'].reshape(-1, *batch['subset_images'].shape[2:])
                subset_bboxes_list = [bbox[:, :4] for bbox in batch["subset_boxes"]]

            # ============= Start training ==============

            def pad_tensor(tensor, total_len):
                valid_len, feat_dim = tensor.shape
                padded = torch.zeros((total_len, feat_dim), device=tensor.device, dtype=tensor.dtype)
                padded[:valid_len] = tensor
                return padded, valid_len, feat_dim

            visual_backbone = model.module.visual if hasattr(model, 'module') else model.visual
            grid_size = visual_backbone.grid_size   

            with autocast():
                features = model(batch=batch, device=device, input_dtype=input_dtype, used_losses=args.loss_type, last_attn_type=args.last_attn_type)
                if teacher and "distill" in args.loss_type:
                    with torch.no_grad():
                        features_t = teacher(batch=batch, device=device, input_dtype=input_dtype, used_losses=args.loss_type, last_attn_type=args.last_attn_type)
                logit_scale = features["logit_scale"]

                def _denormalize_boxes(normed_boxes, x):
                    h, w = x.shape[-2:]
                    denormed_boxes = []
                    for boxes in normed_boxes:
                        new_boxes = boxes.clone()
                        new_boxes[:, [0, 2]] *= w
                        new_boxes[:, [1, 3]] *= h
                        denormed_boxes.append(new_boxes)
                    return denormed_boxes

                def extract_normed_roi_features(global_patches, bboxes_list, grid_size):
                    """
                    Extract ROI features from global_patches, then perform mean pooling and normalize.
                    """
                    B, N, D = global_patches.shape  # (B, N_patches, D)
                    H, W = grid_size

                    patches_2d = global_patches.view(B, H, W, D).permute(0, 3, 1, 2)  # (B, D, H, W)
                    rois = _denormalize_boxes(bboxes_list, patches_2d)  # (B*n, 5)
                    roi_feats = roi_align(patches_2d, rois, output_size=grid_size, spatial_scale=1.0, sampling_ratio=-1,
                                          aligned=True)
                    # roi_global_features = F.normalize(roi_feats, dim=-1)
                    pooled = roi_feats.mean(dim=[2, 3])                         # (B*n, D)
                    normed = F.normalize(pooled, dim=-1)
                    return normed

                # ============== feat from global patches ===============
                if "distill" in args.loss_type:
                    normed_pooled_roi_global_features = extract_normed_roi_features(
                        features["global_patches"], subset_bboxes_list, grid_size)

                losses = {}
                # ============ distill loss ============
                if "distill" in args.loss_type:     # ["roi2cls", "roi2pooled", "combined"]
                    if args.distill_align == 'roi2pooled':      # patch-to-patch alignment
                        if args.distill_type == 'active':
                            local_patches_t = features["subset_patches"]
                        else: local_patches_t = features_t["subset_patches"]

                        local_patches_t = local_patches_t.view(local_patches_t.shape[0], grid_size[0], grid_size[1],
                                                           -1).permute(0, 3, 1, 2)
                        pooled_local_patches_t = local_patches_t.mean(dim=[2, 3])
                        normed_pooled_local_patches_t = F.normalize(pooled_local_patches_t, dim=-1)
                        loss_cosine = 1.0 - (normed_pooled_roi_global_features * normed_pooled_local_patches_t).sum(-1).mean()    # global roi & local roi

                    elif args.distill_align == 'roi2cls':       # indirect alignment with text
                        if args.distill_type=='active':
                            normed_local_pooled_t = F.normalize(features["subset_image_pooled"], dim=-1)
                        else:
                            normed_local_pooled_t = F.normalize(features_t["subset_image_pooled"], dim=-1)
                        loss_cosine = 1.0 - (normed_pooled_roi_global_features * normed_local_pooled_t).sum(-1).mean()  # global roi & local pooled

                    elif args.distill_align == 'combined':
                        if args.distill_type == 'active':
                            local_patches_t = features["subset_patches"]
                            local_image_pooled_t = features["subset_image_pooled"]
                        else:
                            local_patches_t = features_t["subset_patches"]
                            local_image_pooled_t = features_t["subset_image_pooled"]

                        local_patches_t = local_patches_t.view(local_patches_t.shape[0], grid_size[0], grid_size[1],
                                                           -1).permute(0, 3, 1, 2)
                        pooled_local_patches_t = local_patches_t.mean(dim=[2, 3])
                        normed_pooled_local_patches_t = F.normalize(pooled_local_patches_t, dim=-1)

                        normed_local_pooled_t = F.normalize(local_image_pooled_t, dim=-1)

                        loss_roi2roi = 1.0 - (normed_pooled_roi_global_features * normed_pooled_local_patches_t).sum(-1).mean()
                        loss_roi2pooled = 1.0 - (normed_pooled_roi_global_features * normed_local_pooled_t).sum(-1).mean()
                        loss_cosine = 0.5 * loss_roi2roi + 0.5 * loss_roi2pooled
                    else:
                        raise ValueError(f"Unknown distill_align: {args.distill_align}")
                    losses["distill"] = loss_cosine

                # ============ local_itc loss ============
                if "local_itc" in args.loss_type:       # ["cls", "pooled", "roi"]
                    total_len = args.batch_size * args.max_boxes

                    if args.local_itc_align == "cls":
                        normed_local_pooled = F.normalize(features["local_image_pooled"], dim=-1)
                        local_feat = normed_local_pooled

                    elif args.local_itc_align == "pooled":
                        local_patches = features["local_patches"]
                        local_patches = local_patches.view(local_patches.shape[0], grid_size[0], grid_size[1],
                                                           -1).permute(0, 3, 1, 2)
                        pooled_local_patches = local_patches.mean(dim=[2, 3])
                        normed_pooled_local_patches = F.normalize(pooled_local_patches, dim=-1)

                        local_feat = normed_pooled_local_patches

                    elif args.local_itc_align == "roi":
                        local_feat = extract_normed_roi_features(
                            features["global_patches"], bboxes_list, grid_size)

                    else:
                        raise ValueError(f"Unknown local_itc_align: {args.local_itc_align}")

                    padded_normed_local_pooled, valid_len, feature_len = pad_tensor(local_feat, total_len)

                    normed_local_text_pooled = F.normalize(features["local_text_pooled"], dim=-1)
                    padded_normed_local_text_pooled, _, _ = pad_tensor(normed_local_text_pooled, total_len)

                    padding_mask = torch.zeros(total_len, dtype=torch.bool, device=device)
                    padding_mask[:valid_len] = True

                    padded_local_categories = torch.full((total_len,), -1, dtype=local_categories.dtype,
                                                         device=local_categories.device)
                    padded_local_categories[:valid_len] = local_categories

                    if mpcl_loss:
                        region_itc_loss = mpcl_loss(padded_normed_local_pooled, padded_normed_local_text_pooled, logit_scale, padded_local_categories)
                    else:
                        region_itc_loss = loss(padded_normed_local_pooled, padded_normed_local_text_pooled, logit_scale, padding_mask=padding_mask)

                    losses["region_itc"] = region_itc_loss

                # ============ global_itc loss ============
                if "global_itc" in args.loss_type:
                    normed_global_pooled = F.normalize(features["global_image_pooled"], dim=-1)
                    normed_global_text_pooled = F.normalize(features["global_text_pooled"], dim=-1)
                    assert normed_global_pooled.shape[0] == normed_global_text_pooled.shape[0]
                    global_itc_loss = loss(normed_global_pooled, normed_global_text_pooled, logit_scale)
                    losses["global_itc"] = global_itc_loss

                total_loss = losses.get("distill", 0) * args.w_d + losses.get("region_itc", 0) * args.w_l + losses.get("global_itc", 0) * args.w_g
                losses["loss"] = total_loss

            backward(total_loss, scaler)


        elif method == 'vanilla':
            # images, texts = batch
            images, texts = batch["global_image"], batch["global_text"]
            images = images.to(device=device, dtype=input_dtype, non_blocking=True)
            texts = texts.to(device=device, non_blocking=True)
            if args.accum_freq == 1:
                with autocast():
                    model_out = model(images, texts)
                    logit_scale = model_out["logit_scale"]
                    losses = loss(**model_out, output_dict=True)

                    total_loss = sum(losses.values())
                    losses["loss"] = total_loss

                backward(total_loss, scaler)
            else:
                # First, cache the features without any gradient tracking.
                with torch.no_grad():
                    with autocast():
                        model_out = model(images, texts)

                        for f in ("logit_scale", "logit_bias"):
                            model_out.pop(f, None)

                        for key, val in model_out.items():
                            if key in accum_features:
                                accum_features[key].append(val)
                            else:
                                accum_features[key] = [val]

                    accum_images.append(images)
                    accum_texts.append(texts)

                # If (i + 1) % accum_freq is not zero, move on to the next batch.
                if ((i + 1) % args.accum_freq) > 0:
                    # FIXME this makes data time logging unreliable when accumulating
                    continue

                # Now, ready to take gradients for the last accum_freq batches.
                # Re-do the forward pass for those batches, and use the cached features from the other batches as negatives.
                # Call backwards each time, but only step optimizer at the end.
                optimizer.zero_grad()
                for j in range(args.accum_freq):
                    images = accum_images[j]
                    texts = accum_texts[j]
                    with autocast():
                        model_out = model(images, texts)

                        inputs_no_accum = {}
                        inputs_no_accum["logit_scale"] = logit_scale = model_out.pop("logit_scale")
                        if "logit_bias" in model_out:
                            inputs_no_accum["logit_bias"] = model_out.pop("logit_bias")

                        inputs = {}
                        for key, val in accum_features.items():
                            accumulated = accum_features[key]
                            inputs[key] = torch.cat(accumulated[:j] + [model_out[key]] + accumulated[j + 1:])

                        losses = loss(**inputs, **inputs_no_accum, output_dict=True)
                        del inputs
                        del inputs_no_accum
                        total_loss = sum(losses.values())
                        losses["loss"] = total_loss

                    backward(total_loss, scaler)

        if teacher and args.distill_type == 'ema':
            # EMA update for the teacher
            momentum = args.EMA_momentum
            with torch.no_grad():
                for param_q, param_k in zip(model.parameters(), teacher.parameters()):
                    param_k.data.mul_(momentum).add_(
                        (1 - momentum) * param_q.detach().data)

        if scaler is not None:
            if args.horovod:
                optimizer.synchronize()
                scaler.unscale_(optimizer)
                if args.grad_clip_norm is not None:
                    torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip_norm, norm_type=2.0)
                with optimizer.skip_synchronize():
                    scaler.step(optimizer)
            else:
                if args.grad_clip_norm is not None:
                    scaler.unscale_(optimizer)
                    torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip_norm, norm_type=2.0)
                scaler.step(optimizer)
            scaler.update()
        else:
            if args.grad_clip_norm is not None:
                torch.nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip_norm, norm_type=2.0)
            optimizer.step()

        # reset gradient accum, if enabled
        if args.accum_freq > 1:
            accum_images, accum_texts, accum_features = [], [], {}

        # Note: we clamp to 4.6052 = ln(100), as in the original paper.
        with torch.no_grad():
            unwrap_model(model).logit_scale.clamp_(0, math.log(100))
            if teacher and args.distill_type == 'ema': unwrap_model(teacher).logit_scale.clamp_(0, math.log(100))

        batch_time_m.update(time.time() - end)
        end = time.time()
        batch_count = i_accum + 1
        if is_master(args) and (i_accum % args.log_every_n_steps == 0 or batch_count == num_batches_per_epoch):
            batch_size = dataloader.batch_size
            num_samples = batch_count * batch_size * args.accum_freq * args.world_size
            samples_per_epoch = dataloader.num_samples
            percent_complete = 100.0 * batch_count / num_batches_per_epoch

            # NOTE loss is coarsely sampled, just master node and per log update
            for key, val in losses.items():
                if key not in losses_m:
                    losses_m[key] = AverageMeter()
                losses_m[key].update(val.item(), batch_size)

            logit_scale_scalar = logit_scale.item()
            loss_log = " ".join(
                [
                    f"{loss_name.capitalize()}: {loss_m.val:#.5g} ({loss_m.avg:#.5g})"
                    for loss_name, loss_m in losses_m.items()
                ]
            )
            samples_per_second = args.accum_freq * args.batch_size * args.world_size / batch_time_m.val
            samples_per_second_per_gpu = args.accum_freq * args.batch_size / batch_time_m.val
            eta_total = batch_time_m.total_sum / (i+1) * (args.epochs - epoch - (i+1)/dataloader.num_batches) * dataloader.num_batches
            logging.info(
                f"Train Epoch: {epoch} [{num_samples:>{sample_digits}}/{samples_per_epoch} ({percent_complete:.0f}%)] "
                # f"ETA: {eta_total:.3f}s "
                f"ETA: {int(eta_total // 3600):02}h{int((eta_total % 3600) // 60):02}m{int(eta_total % 60):02}s "
                f"Data (t): {data_time_m.avg:.3f} "
                f"Batch (t): {batch_time_m.avg:.3f}, {samples_per_second:#g}/s, {samples_per_second_per_gpu:#g}/s/gpu "
                f"LR: {optimizer.param_groups[0]['lr']:5f} "
                f"Logit Scale: {logit_scale_scalar:.3f} " + loss_log
            )

            # Save train loss / etc. Using non avg meter values as loggers have their own smoothing
            log_data = {
                "data_time": data_time_m.val,
                "batch_time": batch_time_m.val,
                "samples_per_second": samples_per_second,
                "samples_per_second_per_gpu": samples_per_second_per_gpu,
                "scale": logit_scale_scalar,
                "lr": optimizer.param_groups[0]["lr"]
            }
            log_data.update({name:val.val for name,val in losses_m.items()})

            log_data = {"train/" + name: val for name, val in log_data.items()}

            if tb_writer is not None:
                for name, val in log_data.items():
                    tb_writer.add_scalar(name, val, step)

            if args.wandb:
                assert wandb is not None, 'Please install wandb.'
                log_data['step'] = step  # for backwards compatibility
                wandb.log(log_data, step=step)

            # resetting batch / data time meters per log window
            batch_time_m.reset()
            data_time_m.reset()

    # end for


def evaluate(model, data, epoch, args, tb_writer=None, tokenizer=None):
    metrics = {}
    if not is_master(args):
        return metrics
    device = torch.device(args.device)
    model.eval()

    zero_shot_metrics = zero_shot_eval(model, data, epoch, args, tokenizer=tokenizer)
    metrics.update(zero_shot_metrics)

    autocast = get_autocast(args.precision, device_type=device.type)
    input_dtype = get_input_dtype(args.precision)

    if 'val' in data and (args.val_frequency and ((epoch % args.val_frequency) == 0 or epoch == args.epochs)):
        dataloader = data['val'].dataloader
        num_samples = 0
        samples_per_val = dataloader.num_samples

        # FIXME this does not scale past small eval datasets
        # all_image_features @ all_text_features will blow up memory and compute very quickly
        cumulative_loss = 0.0
        cumulative_gen_loss = 0.0
        all_image_features, all_text_features = [], []
        with torch.inference_mode():
            for i, batch in enumerate(dataloader):
                images, texts = batch
                images = images.to(device=device, dtype=input_dtype, non_blocking=True)
                texts = texts.to(device=device, non_blocking=True)

                with autocast():
                    model_out = model(images, texts)
                    image_features = model_out["image_features"]
                    text_features = model_out["text_features"]
                    logit_scale = model_out["logit_scale"]
                    # features are accumulated in CPU tensors, otherwise GPU memory exhausted quickly
                    # however, system RAM is easily exceeded and compute time becomes problematic
                    all_image_features.append(image_features.cpu())
                    all_text_features.append(text_features.cpu())
                    logit_scale = logit_scale.mean()
                    logits_per_image = logit_scale * image_features @ text_features.t()
                    logits_per_text = logits_per_image.t()

                    batch_size = images.shape[0]
                    labels = torch.arange(batch_size, device=device).long()
                    total_loss = (
                        F.cross_entropy(logits_per_image, labels) +
                        F.cross_entropy(logits_per_text, labels)
                    ) / 2

                    gen_loss = maybe_compute_generative_loss(model_out)

                cumulative_loss += total_loss * batch_size
                num_samples += batch_size
                if is_master(args) and (i % 100) == 0:
                    logging.info(
                        f"Eval Epoch: {epoch} [{num_samples} / {samples_per_val}]\t"
                        f"Clip Loss: {cumulative_loss / num_samples:.6f}\t")

                    if gen_loss is not None:
                        cumulative_gen_loss += gen_loss * batch_size
                        logging.info(
                            f"Generative Loss: {cumulative_gen_loss / num_samples:.6f}\t")

            val_metrics = get_clip_metrics(
                image_features=torch.cat(all_image_features),
                text_features=torch.cat(all_text_features),
                logit_scale=logit_scale.cpu(),
            )
            loss = cumulative_loss / num_samples
            metrics.update(
                {**val_metrics, "clip_val_loss": loss.item(), "epoch": epoch, "num_samples": num_samples}
            )
            if gen_loss is not None:
                gen_loss = cumulative_gen_loss / num_samples
                metrics.update({"val_generative_loss": gen_loss.item()})

    if not metrics:
        return metrics

    logging.info(
        f"Eval Epoch: {epoch} "
        + "\t".join([f"{k}: {round(v, 4):.4f}" for k, v in metrics.items()])
    )

    log_data = {"val/" + name: val for name, val in metrics.items()}

    if args.save_logs:
        if tb_writer is not None:
            for name, val in log_data.items():
                tb_writer.add_scalar(name, val, epoch)

        with open(os.path.join(args.checkpoint_path, "results.jsonl"), "a+") as f:
            f.write(json.dumps(metrics))
            f.write("\n")

    if args.wandb:
        assert wandb is not None, 'Please install wandb.'
        if 'train' in data:
            dataloader = data['train'].dataloader
            num_batches_per_epoch = dataloader.num_batches // args.accum_freq
            step = num_batches_per_epoch * epoch
        else:
            step = None
        log_data['epoch'] = epoch
        wandb.log(log_data, step=step)

    return metrics


def get_clip_metrics(image_features, text_features, logit_scale):
    metrics = {}
    logits_per_image = (logit_scale * image_features @ text_features.t()).detach().cpu()
    logits_per_text = logits_per_image.t().detach().cpu()

    logits = {"image_to_text": logits_per_image, "text_to_image": logits_per_text}
    ground_truth = torch.arange(len(text_features)).view(-1, 1)

    for name, logit in logits.items():
        ranking = torch.argsort(logit, descending=True)
        preds = torch.where(ranking == ground_truth)[1]
        preds = preds.detach().cpu().numpy()
        metrics[f"{name}_mean_rank"] = preds.mean() + 1
        metrics[f"{name}_median_rank"] = np.floor(np.median(preds)) + 1
        for k in [1, 5, 10]:
            metrics[f"{name}_R@{k}"] = np.mean(preds < k)

    return metrics


def maybe_compute_generative_loss(model_out):
    if "logits" in model_out and "labels" in model_out:
        token_logits = model_out["logits"]
        token_labels = model_out["labels"]
        return F.cross_entropy(token_logits.permute(0, 2, 1), token_labels)