File size: 25,560 Bytes
94da716
 
 
 
 
 
 
 
 
 
 
 
 
7331cf8
94da716
 
 
 
 
 
 
 
 
9125121
94da716
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9125121
94da716
9125121
94da716
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ba5c98e
 
 
 
 
94da716
 
 
 
 
 
 
 
 
 
 
 
 
 
 
705386d
94da716
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4fc00f2
94da716
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import io
import os

import requests
import torch
import torch.nn as nn
import torch.nn.functional as F
# from petrel_client.client import Client
from PIL import Image, ImageFile
from torch.nn.utils import rnn
from types import SimpleNamespace
from peft import LoraConfig, TaskType, get_peft_model
from transformers import LlamaTokenizer, LlamaForCausalLM, LlamaConfig

import numpy as np
# from header import *

from transformers import StoppingCriteria, StoppingCriteriaList

from .CLIP import load as load_clip
from .PROCESS import data
from .modeling_llama import LlamaForCausalLM
from .utils.pcl_utils import MEAN_COLOR_RGB, RandomCuboid, random_sampling
from .conversations import conversation_dict, default_conversation

ImageFile.LOAD_TRUNCATED_IMAGES = True

# sov: start of vision part; eov: end of vision part
VISION_TAGS = {
    'pos': {'image': '<image>', 'pcl': '<pcl>'},
    'sov': {'image': '<Img>', 'pcl': '<Pcl>'},
    'eov': {'image': '</Img>', 'pcl': '</Pcl>'},
}
ModalityType = SimpleNamespace(
    VISION="vision",
    TEXT="text",
    AUDIO="audio",
    THERMAL="thermal",
    DEPTH="depth",
    IMU="imu",
)

class StoppingCriteriaSub(StoppingCriteria):

    def __init__(self, stops = [], encounters=1):
        super().__init__()
        self.stops = stops
        self.ENCOUNTERS = encounters

    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor):
        stop_count = 0
        for stop in self.stops:
            stop_count = (stop == input_ids[0]).sum().item()
        if stop_count >= self.ENCOUNTERS:
            return True
        return False


class MyStoppingCriteria(StoppingCriteria):
    def __init__(self, stops, input_ids):
        super().__init__()
        self.stops = [torch.tensor(stop).to('cuda:0') for stop in stops]
        self.stop_flag = [0]*input_ids.shape[0]

    def check_stop(self, input_ids):
        for stop in self.stops:
            if torch.all((stop == input_ids[-len(stop):])).item():
                return True
        return False

    def __call__(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        flag = 1
        for id, output_id in enumerate(output_ids):
            if self.stop_flag[id] == 1:
                continue
            if self.check_stop(output_id):
                self.stop_flag[id] = 1
            else:
                flag = 0
        if flag == 1:
            return True
        return False


def build_one_instance(tokenizer, conversation, vision_type='image'):
    pos = VISION_TAGS['pos'][vision_type]
    # sov = VISION_TAGS['sov'][vision_type]
    eov = VISION_TAGS['eov'][vision_type]

    text_list = []
    turn_num = len(conversation)
    input_ids, target_ids = [], []
    for i in range(turn_num):
        turn = conversation[i]
        role = turn['from']
        if i == 0: # the first human turn
            assert role == 'human'
            turn['value'] = turn['value'].replace(f'{pos}\n', '').replace(f'\n{pos}', '')
            text = f'{eov} ' + turn['value'] + '\n### Assistant:'
            one_input_id = tokenizer(text, add_special_tokens=False).input_ids
            input_ids += one_input_id
            target_ids += [-100]*len(one_input_id) # do not perform loss regression on human prompt
        else:
            if role == 'human':
                text = 'Human: ' + turn['value'] + '\n### Assistant:'
                one_input_id = tokenizer(text, add_special_tokens=False).input_ids
                input_ids += one_input_id
                target_ids += [-100]*len(one_input_id)
            elif role == 'gpt':
                text = turn['value'] + '\n###'
                one_input_id = tokenizer(text, add_special_tokens=False).input_ids
                input_ids += one_input_id
                target_ids += one_input_id
            else:
                raise Exception('Wrong Role!!!')
        text_list.append(text)
        assert len(input_ids) == len(target_ids)
    return text_list, input_ids, target_ids


def process_batch_instance(tokenizer, batch_of_conversations, max_tgt_len, vision_type='image'):
    batch_input_ids, batch_target_ids = [], []
    for conversation in batch_of_conversations:
        _, one_input_ids, one_target_ids = build_one_instance(tokenizer, conversation, vision_type=vision_type)
        batch_input_ids.append(torch.LongTensor(one_input_ids))
        batch_target_ids.append(torch.LongTensor(one_target_ids))
    input_ids = rnn.pad_sequence(batch_input_ids, batch_first=True, padding_value=tokenizer.pad_token_id)
    target_ids = rnn.pad_sequence(batch_target_ids, batch_first=True, padding_value=-100)
    assert input_ids.size() == target_ids.size()
    input_ids = input_ids[:,:max_tgt_len]
    target_ids = target_ids[:,:max_tgt_len]
    attention_mask = input_ids.ne(tokenizer.pad_token_id)
    assert attention_mask.size() == input_ids.size()
    return input_ids, target_ids, attention_mask.long()


def make_prompt_start(system_header=False, vision_type='image', task_type='normal'):
    # TODO: choose prefix according to task type
    PROMPT_START = f'### Human: {VISION_TAGS["sov"][vision_type]}'
    if system_header:
        if task_type == 'normal':
            return f"{default_conversation.system}\n\n" + PROMPT_START
        else:
            return [f"{conversation_dict[task]}\n\n" + PROMPT_START for task in task_type]
    else:
        return PROMPT_START


class LAMMPEFTModel(nn.Module):

    '''LoRA for LLaMa model'''

    def __init__(self, **args):
        super(LAMMPEFTModel, self).__init__()
        self.args = args
        # self.client = Client('~/petreloss.conf')
        self.client = None

        self.vision_type = args['vision_type'] if 'vision_type' in args else 'image'
        encoder_pretrain = args['encoder_pretrain'] if 'encoder_pretrain' in args else 'clip'
        self.encoder_pretrain = encoder_pretrain
        assert encoder_pretrain in ['imagebind', 'clip', 'epcl'], f'Encoder_pretrain: {encoder_pretrain} Not Implemented'
        if not encoder_pretrain == 'clip' or os.path.isfile(args['encoder_ckpt_path']):
            encoder_ckpt_path = args['encoder_ckpt_path'] 
        elif not os.path.isfile(args['encoder_ckpt_path']):
            encoder_ckpt_path = 'ViT-L/14'
            
        vicuna_ckpt_path = args['vicuna_ckpt_path']
        
        system_header = args['system_header'] if 'system_header' in args else False
        stage = args['stage']

        # TODO: checkout vision token number; for ImageBind = 1; Defaultly to use 1 global token for this
        # -1 for last embedding; -2 for transformer output
        self.vision_feature_type = args['vision_feature_type']
        self.num_vision_token = args['num_vision_token']

        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
        print (f'Initializing [{encoder_pretrain}] visual encoder from {encoder_ckpt_path} [{device}]...')

        # TODO: Make sure the number of vision tokens is correct
        if args['encoder_pretrain'].lower() == 'clip':
            clip_encoder, self.visual_preprocess = load_clip(encoder_ckpt_path, device=device)
            self.visual_encoder = clip_encoder.visual
            if self.vision_feature_type == 'global':          # global feature from CLIP
                self.vision_hidden_size = 768
                self.num_vision_token = 1
                assert self.num_vision_token == 1, 'Only 1 global token is available!'
            elif self.vision_feature_type == 'local':        # patch features from CLIP ViT
                self.vision_hidden_size = 1024
                self.num_vision_token = min(self.num_vision_token, 256)         # may cut partial tokens

        # freeze vision encoder
        for name, param in self.visual_encoder.named_parameters():
            param.requires_grad = False
        self.visual_encoder.eval()
        print ('Visual encoder initialized.')

        print (f'Initializing language decoder from {vicuna_ckpt_path} ...')
        # add the lora module
        peft_config = LoraConfig(
            task_type=TaskType.CAUSAL_LM, 
            inference_mode=False, 
            r=self.args['lora_r'], 
            lora_alpha=self.args['lora_alpha'], 
            lora_dropout=self.args['lora_dropout'],
            target_modules=self.args['lora_target_modules']
        )

        self.llama_model = LlamaForCausalLM.from_pretrained(vicuna_ckpt_path)
        self.llama_model = get_peft_model(self.llama_model, peft_config)
        self.llama_model.print_trainable_parameters()

        self.llama_tokenizer = LlamaTokenizer.from_pretrained(vicuna_ckpt_path, use_fast=False)
        self.llama_tokenizer.pad_token = self.llama_tokenizer.eos_token
        self.llama_tokenizer.padding_side = "right"
        print ('Language decoder initialized.')

        self.llama_proj = nn.Linear(
            self.vision_hidden_size, self.llama_model.config.hidden_size
        )
        print ('LLaMa projection layer initialized.')

        self.max_tgt_len = args['max_tgt_len']
        self.system_header = system_header
        self.device = torch.cuda.current_device()

    def encode_image(self, image_paths):
        """encode images to llama inputs

        :param tupe image_paths: (bsz, )
        :return tensor, tensor: input feature to llama, attention mask to llama
        """
        if self.encoder_pretrain == 'imagebind':
            inputs = {ModalityType.VISION: data.load_and_transform_vision_data(image_paths, self.device)}
            # convert into visual dtype
            inputs = {key: inputs[key].to(self.llama_model.dtype) for key in inputs}
            with torch.no_grad():
                embeddings = self.visual_encoder(inputs)
                image_embeds = embeddings['vision']                     # bsz x 1024
            inputs_llama = self.llama_proj(image_embeds).unsqueeze(1)   # bsz x 1 x llama_size
            atts_llama = torch.ones(inputs_llama.size()[:-1], dtype=torch.long).to(self.device) # bsz x 1
            return inputs_llama, atts_llama
        elif self.encoder_pretrain == 'clip':
            inputs = self.load_and_transform_vision_data_clip(image_paths, self.device)     # bsz x 3 x 224 x 224
            inputs = inputs.to(self.llama_model.dtype)                                      # clip requires torch.float32
            inputs_llama = self.clip_encode_image(inputs)
            atts_llama = torch.ones(inputs_llama.size()[:-1], dtype=torch.long).to(self.device)                     # bsz x 1/256
            return inputs_llama, atts_llama
        
    def my_encode_image(self, images):
        """encoder loaded image objects"""
        if self.encoder_pretrain == 'clip':
            inputs = data.transform_vision_data(images, self.device)                    # bsz x 3 x 224 x 224
            inputs_llama = self.clip_encode_image(inputs)                               # bsz x 1/256 x llama_size
            atts_llama = torch.ones(inputs_llama.size()[:-1], dtype=torch.long).to(self.device)                     # bsz x 1/256
            return inputs_llama, atts_llama
        else:
            raise NotImplementedError("Encoder pretrain [{}] not implemented".format(self.encoder_pretrain))
    
    def encode_pcl(self, pcl_paths):
        # load pcl data
        inputs = self.load_and_transform_pcl_data(pcl_paths, self.device) # bsz x 40000 x 3
        
        inputs = inputs.to(self.llama_model.dtype)                                  # clip requires torch.float32
        with torch.no_grad():
            if self.vision_feature_type == 'global':
                raise NotImplementedError("Global feature not implemented for pcl")
            elif self.vision_feature_type == 'local':
                embeddings = self.visual_encoder(inputs)[1][:, :self.num_vision_token]      # bsz x 256 x 1024;
                image_embeds = embeddings.reshape(-1, self.vision_hidden_size).to(self.llama_model.dtype)       # bsz*num vision token x 1024
                inputs_llama = self.llama_proj(image_embeds).reshape(-1, self.num_vision_token, self.llama_model.config.hidden_size) # bsz x num_vision_token x llama_size
                atts_llama = torch.ones(inputs_llama.size()[:-1], dtype=torch.long).to(self.device)                     # bsz x 1/256
        return inputs_llama, atts_llama
    
    def clip_encode_image(self, inputs):
        inputs = inputs.to(self.llama_model.dtype)                                  # clip requires torch.float32
        with torch.no_grad():
            if self.vision_feature_type == 'global':
                embeddings = self.visual_encoder(inputs)                            # bsz x 768
                image_embeds = embeddings.to(self.llama_model.dtype)
                inputs_llama = self.llama_proj(image_embeds).unsqueeze(1)           # bsz x 1 x llama_size
            elif self.vision_feature_type == 'local':
                embeddings = self.visual_encoder.forward_patch_features(inputs)[:, :self.num_vision_token]      # bsz x self.num_vision_token x 1024
                image_embeds = embeddings.reshape(-1, self.vision_hidden_size).to(self.llama_model.dtype)       # bsz*num vision token x 1024
                inputs_llama = self.llama_proj(image_embeds).reshape(-1, self.num_vision_token, self.llama_model.config.hidden_size) # bsz x num_vision_token x llama_size
            else:
                raise NotImplementedError("{} not Implemented".format(self.vision_feature_type))
        return inputs_llama

    def load_and_transform_vision_data_clip(self, image_paths, device):
        if image_paths is None:
            return None
        image_ouputs = []
        for image_path in image_paths:
            if os.path.exists(image_path):
                image = Image.open(image_path)
            elif image_path.startswith('s3://') and self.client is not None:
                image = Image.open(io.BytesIO(self.client.get(image_path, update_cache=True))).convert("RGB")
            elif image_path.startswith('http://'):
                image = Image.open(requests.get(image_path, stream=True).raw)
            else:
                print("can not load image: ", image_path)
            image_outpt = self.visual_preprocess(image).to(device)                      # 3 x 224 x 224
            image_ouputs.append(image_outpt)
        return torch.stack(image_ouputs, dim=0)                                         # B x 3 x 224 x 224
    
    def load_and_transform_pcl_data(self, pcl_paths, device):
        if pcl_paths is None:
            return None
        pcl_output = []
        for pcl_path in pcl_paths:
            mesh_vertices = np.load(pcl_path)         # 150000, 3
            if not self.use_color:
                point_cloud = mesh_vertices[:, 0:3]  # do not use color for now
            else:
                point_cloud = mesh_vertices[:, 0:6]
                point_cloud[:, 3:] = (point_cloud[:, 3:] - MEAN_COLOR_RGB) / 256.0
            
            if self.use_height:
                floor_height = np.percentile(point_cloud[:, 2], 0.99)
                height = point_cloud[:, 2] - floor_height
                point_cloud = np.concatenate([point_cloud, np.expand_dims(height, 1)], 1)
            
            point_cloud, _ = random_sampling(
                point_cloud, self.num_points, return_choices=True
            )
            pcl_output.append(torch.from_numpy(point_cloud))
        return torch.stack(pcl_output, dim=0).to(device) # bsz x num_points x 3

    def prompt_wrap(self, img_embeds, input_ids, target_ids, attention_mask, system_header, task_type):
        '''
            input_ids, target_ids, attention_mask: bsz x s2
        '''
        input_ids = input_ids.to(self.device)           # bsz x s2
        target_ids = target_ids.to(self.device)         # bsz x s2
        attention_mask = attention_mask.to(self.device) # bsz x s2

        batch_size = img_embeds.shape[0]

        # return list of headers if multiple tasks
        p_before = make_prompt_start(system_header=system_header, vision_type=self.vision_type, task_type=task_type)
        if isinstance(p_before, list):
            p_before_tokens = [self.llama_tokenizer(p, 
                return_tensors="pt", add_special_tokens=False).input_ids[0].to(self.device) for p in p_before]
            # TODO: fix bug here
            p_before_token_ids = rnn.pad_sequence(p_before_tokens, batch_first=True, padding_value=self.llama_tokenizer.pad_token_id) # bsz x s1
            p_before_attn_mask = p_before_token_ids.ne(self.llama_tokenizer.pad_token_id)
        else:
            p_before_tokens = self.llama_tokenizer(p_before, 
                return_tensors="pt", add_special_tokens=False).to(self.device)  # [s1, s1...] list of batch size
            p_before_token_ids = p_before_tokens.input_ids.expand(batch_size, -1) # bsz x s1
            p_before_attn_mask = p_before_tokens.attention_mask.expand(batch_size, -1) # bsz x s1
        # peft model need deeper call
        p_before_embeds = self.llama_model.model.model.embed_tokens(p_before_token_ids) #.expand(batch_size, -1, -1) # bsz x s1 x embed_dim
        p_after_embeds = self.llama_model.model.model.embed_tokens(input_ids).expand(batch_size, -1, -1) # bsz x s2 x embed_dim
        bos = torch.ones([batch_size, 1],
                         dtype=p_before_token_ids.dtype,
                         device=p_before_token_ids.device) * self.llama_tokenizer.bos_token_id # bsz x 1
        bos_embeds = self.llama_model.model.model.embed_tokens(bos) # bsz x 1 x embed_dim
        inputs_embeds = torch.cat([bos_embeds, p_before_embeds, img_embeds, p_after_embeds], dim=1) # bsz x (1+s1+NumToken+s2) x embed_dim

        # make target ids for prefix part
        empty_targets = (
            torch.ones([batch_size, 1 + p_before_embeds.size()[1] + self.num_vision_token], # 1 (bos) + s1 + num_image_tokens (image vector)
                       dtype=torch.long).to(self.device).fill_(-100)
        ) # bsz x (1 + s1 + 1)
        targets = torch.cat([empty_targets, target_ids], dim=1) # bsz x (1 + s1 + num_image_tokens + s2)
        assert inputs_embeds.size()[1] == targets.size()[1]

        # atts_prefix = torch.ones([batch_size, 1 + p_before_embeds.size()[1] + self.num_vision_token], dtype=torch.long).to(self.device) # bsz x (1[bos] + s1 +num_image_tokens)
        atts_bos = torch.ones([batch_size, 1], dtype=torch.long).to(self.device) # bsz x 1
        atts_img = torch.ones([batch_size, self.num_vision_token], dtype=torch.long).to(self.device) # bsz x num_image_tokens
        attention_mask = torch.cat([atts_bos, p_before_attn_mask, atts_img, attention_mask], dim=1)
        assert attention_mask.size() == targets.size() # bsz x (1 + s1 + num_image_tokens + s2)
        return inputs_embeds, targets, attention_mask

    def forward(self, inputs):
        """Model Forward in training

        :param class inputs: model itself
        :raises ValueError: valueerror if not image or pcl
        :return list: loss & token acc
        """
        # image_paths = inputs['image_paths']
        assert self.vision_type == inputs['vision_type']    # single modal case
        task_type = inputs['task_type']
        vision_paths = inputs['vision_paths']
        if self.vision_type == 'image':
            vision_embeds, _ = self.encode_image(vision_paths)
        elif self.vision_type == 'pcl':
            vision_embeds, _ = self.encode_pcl(vision_paths)        # Bsz x N token x C
        else:
            raise ValueError('vision type [{}] not supported'.format(self.vision_type))

        output_texts = inputs['output_texts']
        input_ids, target_ids, attention_mask = process_batch_instance(self.llama_tokenizer, output_texts, self.max_tgt_len, self.vision_type)
        inputs_embeds, targets, attention_mask = self.prompt_wrap(vision_embeds, input_ids, target_ids, attention_mask, self.system_header, task_type)

        outputs = self.llama_model(
            inputs_embeds=inputs_embeds,
            attention_mask=attention_mask,
            return_dict=True,
            labels=targets,
        )
        loss = outputs.loss
        # calculate the token accuarcy
        chosen_tokens = torch.max(outputs.logits, dim=-1)[1][:, 1: -1]    # [B, S-1]
        labels = targets[:, 2:]
        gen_acc = (chosen_tokens.reshape(-1) == labels.reshape(-1)).to(torch.long)    # [B*S]
        valid_mask = (labels != -100).reshape(-1)
        valid_tokens = gen_acc & valid_mask    # [B*S]
        gen_acc = valid_tokens.sum().item() / valid_mask.sum().item()
        return loss, gen_acc

    def extract_multimodal_feature(self, inputs):
        """Extract multimodal features from the input in Generation (Test)

        :param Dict inputs: input dict; modality: path
        :return _type_: _description_
        """
        features = []
        if inputs['image_paths']:
            image_embeds, _ = self.encode_image(inputs['image_paths'])
            features.append(image_embeds)
        if 'images' in inputs and inputs['images']:        # image objects input in testing
            image_embeds, _ = self.my_encode_image(inputs['images'])
            return image_embeds
            # features.append(image_embeds)
        if 'pcl_paths' in inputs and inputs['pcl_paths']:
            pcl_embeds, _ = self.encode_pcl(inputs['pcl_paths'])
            features.append(pcl_embeds)
        # TODO: Cautions HERE! Multimodality allowed in test ONLY!
        feature_embeds = torch.cat(features).sum(dim=0).unsqueeze(0)        # sum all modality features together
        return feature_embeds

    def prepare_generation_embedding(self, inputs):
        """prepare for generation

        :param class inputs: model
        :return Dict: generation input
        """
        eov = VISION_TAGS['eov'][self.vision_type]
        # TODO: add System header & image token size
        prompt_list = inputs['prompt']           # questions from user
        if len(inputs['modality_embeds']) == 1:
            feature_embeds = inputs['modality_embeds'][0]
        else:
            feature_embeds = self.extract_multimodal_feature(inputs)
            inputs['modality_embeds'].append(feature_embeds)

        batch_size = feature_embeds.shape[0]
        p_before = make_prompt_start(vision_type=self.vision_type)      # no system header in test
        p_before_tokens = self.llama_tokenizer(p_before, 
            return_tensors="pt", add_special_tokens=False).to(self.device)
        p_before_embeds = self.llama_model.model.model.embed_tokens(p_before_tokens.input_ids).expand(batch_size, -1, -1) # bsz x s1 x embed_dim
        p_after_embeds_list = []
        p_after_tokens_list = []
        for prompt in prompt_list:
            # text = '</Img> ' + prompt + '\n### Assistant:'
            text = f'{eov} ' + prompt + '\n### Assistant:'
            p_after_tokens = self.llama_tokenizer(text, add_special_tokens=False, return_tensors='pt').to(self.device)

            p_after_tokens_list.append(p_after_tokens.input_ids.squeeze(0))

        p_after_tokens = rnn.pad_sequence(p_after_tokens_list, batch_first=True, padding_value=self.llama_tokenizer.pad_token_id)

        p_after_embeds = self.llama_model.model.model.embed_tokens(p_after_tokens)
        
        # text = f'{eov} ' + prompt + '\n### Assistant:'
        # p_after_tokens = self.llama_tokenizer(text, add_special_tokens=False, return_tensors='pt').to(self.device)
        # p_after_embeds = self.llama_model.model.model.embed_tokens(p_after_tokens.input_ids).expand(batch_size, -1, -1) # bsz x s1 x embed_dim
        bos = torch.ones([batch_size, 1],
                         dtype=p_before_tokens.input_ids.dtype,
                         device=p_before_tokens.input_ids.device) * self.llama_tokenizer.bos_token_id # bsz x 1
        bos_embeds = self.llama_model.model.model.embed_tokens(bos) # bsz x 1 x embed_dim
        # print(bos_embeds.shape, p_before_embeds.shape, feature_embeds.shape, p_after_embeds.shape)
        inputs_embeds = torch.cat([bos_embeds, p_before_embeds, feature_embeds, p_after_embeds], dim=1) # bsz x (1+s1+NumVisionToken+s2) x embed_dim
        return inputs_embeds

    def generate(self, inputs):
        '''
            inputs = {
                'image_paths': optional,
                'audio_paths': optional
                'video_paths': optional
                'thermal_paths': optional
                'mode': generation mode,
                'prompt': human input prompt,
                'max_tgt_len': generation length,
                'top_p': top_p,
                'temperature': temperature
                'modality_embeds': None or torch.tensor
                'modality_cache': save the image cache
            }
        '''
        input_embeds = self.prepare_generation_embedding(inputs)
        # stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops=[2277], encounters=1)])
        stopping_criteria = StoppingCriteriaList([MyStoppingCriteria([[2277]], input_embeds)])
        outputs = self.llama_model.generate(
            inputs_embeds=input_embeds,
            max_new_tokens=inputs['max_tgt_len'],
            top_p=inputs['top_p'],
            temperature=inputs['temperature'],
            do_sample=True,
            use_cache=True,
            stopping_criteria=stopping_criteria,
        )
        #output_text = self.llama_tokenizer.decode(outputs[0][:-2], skip_special_tokens=True)
        output_text = self.llama_tokenizer.batch_decode(outputs, skip_special_tokens=True)
        return output_text