File size: 8,215 Bytes
8f83af3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#    Copyright 2023 Haotian Liu
#
#    Licensed under the Apache License, Version 2.0 (the "License");
#    you may not use this file except in compliance with the License.
#    You may obtain a copy of the License at
#
#        http://www.apache.org/licenses/LICENSE-2.0
#
#    Unless required by applicable law or agreed to in writing, software
#    distributed under the License is distributed on an "AS IS" BASIS,
#    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#    See the License for the specific language governing permissions and
#    limitations under the License.


from typing import List, Optional, Tuple, Union
import warnings

import torch
import torch.nn as nn


from transformers import AutoConfig, AutoModelForCausalLM, AutoModel, PretrainedConfig
                        #  StableLMEpochConfig, StableLMEpochModel, StableLMEpochForCausalLM
from transformers.modeling_utils import cached_file, CONFIG_NAME, extract_commit_hash, is_peft_available, find_adapter_config_file, json, os
from transformers.models.auto.auto_factory import _BaseAutoModelClass, _get_model_class
from transformers.dynamic_module_utils import resolve_trust_remote_code, get_class_from_dynamic_module


from transformers.modeling_outputs import CausalLMOutputWithPast

import pdb


import sys
from .llava_arch import LlavaMetaModel, LlavaMetaForCausalLM
from .modeling_stablelm_epoch import StableLMEpochForCausalLM, StableLMEpochModel, StableLMEpochConfig
from .generation_utils import build_allava_input


################ stableLM ###############################

class LlavaStableLM_1_6bConfig(StableLMEpochConfig):
    model_type = "llava_stablelm_1_6b"

# class LlavaStableLMModel(LlavaMetaModel, AutoModel):
class LlavaStableLMModel(LlavaMetaModel, StableLMEpochModel):
    config_class = LlavaStableLM_1_6bConfig

    def __init__(self, config: AutoConfig):
        super(LlavaStableLMModel, self).__init__(config)



class LlavaStableLM_1_6bForCausalLM(StableLMEpochForCausalLM, LlavaMetaForCausalLM):
    config_class = LlavaStableLM_1_6bConfig
    

    def __init__(self, config, init_vision_encoder_from_ckpt=True):
        config._attn_implementation = "flash_attention_2"

        super(StableLMEpochForCausalLM, self).__init__(config)

        self.model = LlavaStableLMModel(config)
        if hasattr(self.model, '_use_flash_attention_2'):
            assert self.model._use_flash_attention_2, 'flash attn is not enabled. check it out!'
        # self.pretraining_tp = config.pretraining_tp
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        
        if init_vision_encoder_from_ckpt:
            vision_tower = self.get_vision_tower()
            print(f'loading from CLIP first. This should only be used at inference!!!')
            vision_tower.load_model()
            
        # Initialize weights and apply final processing
        self.post_init()


    def get_model(self):
        return self.model
    
    def get_tokenizer(self):
        return self.tokenizer
    
    def get_processor(self):
        return self.model.vision_tower.image_processor

    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        images: Optional[torch.FloatTensor] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, CausalLMOutputWithPast]:

        if inputs_embeds is None:
            (
                input_ids,
                position_ids,
                attention_mask,
                past_key_values,
                inputs_embeds,
                labels
            # ) = self.prepare_inputs_labels_for_multimodal(
            ) = self.prepare_inputs_labels_for_multimodal_new(
                input_ids,
                position_ids,
                attention_mask,
                past_key_values,
                labels,
                images
            )

        return super().forward(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            labels=labels,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict
        )

    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs):
        images = kwargs.pop("images", None)
        _inputs = super().prepare_inputs_for_generation(
            input_ids, past_key_values=past_key_values, inputs_embeds=inputs_embeds, **kwargs
        )
        if images is not None:
            _inputs['images'] = images
        return _inputs

    @torch.no_grad()
    def generate(
        self,
        inputs: Optional[torch.Tensor] = None,
        images: Optional[torch.Tensor] = None,
        **kwargs,
    ) :
        position_ids = kwargs.pop("position_ids", None)
        attention_mask = kwargs.pop("attention_mask", None)
        if "inputs_embeds" in kwargs:
            raise NotImplementedError("`inputs_embeds` is not supported")

        if images is not None:
            (
                inputs,
                position_ids,
                attention_mask,
                _,
                inputs_embeds,
                _
            ) = self.prepare_inputs_labels_for_multimodal_new(
                inputs,
                position_ids,
                attention_mask,
                None,
                None,
                images
            )
        else:
            inputs_embeds = self.get_model().embed_tokens(inputs)

        # print(inputs_embeds.shape)
        return super().generate(
            position_ids=None,
            attention_mask=None,
            inputs_embeds=inputs_embeds,
            **kwargs
        )
    

    def chat(
        self, 
        texts: Optional[str | list[list[str, str]]], 
        images: Optional[str | list[str]] = None, 
        history: Optional[list[str]] = None, 
        stream = False, 
        return_history = False, 
        **kwargs
    ):
        '''
        texts: if `str`, then generate for a single round; if list[dict], 
        images: str (optional), local path to an image.
        '''
        use_cache = kwargs.pop('use_cache', True)


        ############################
        # merge history
        ############################            
        input_ids, image_tensors, history = build_allava_input(
            tokenizer = self.get_tokenizer(), 
            processor = self.get_processor(), 
            texts = texts, 
            images = images, 
            history=history,
            return_history=return_history,
            device = self.device
        )

        ############################
        # generate response
        ############################
        # with torch.autocast(device_type='cuda'):
        if 'cuda' in str(self.device):
            device_type = 'cuda'
        else:
            device_type = 'cpu'

        with torch.autocast(device_type=device_type, dtype=self.dtype):
            output_ids = self.generate(
                inputs=input_ids,
                images=image_tensors,
                use_cache=use_cache,
                **kwargs)

        answer = self.get_tokenizer().decode(output_ids[0, :], skip_special_tokens=True).strip()

        if return_history:
            history[-1][-1] = answer
            return answer, history
        return answer


AutoConfig.register("llava_stablelm_1_6b", LlavaStableLM_1_6bConfig)
# AutoConfig.register("stablelm_epoch", LlavaStableLMConfig)
AutoModelForCausalLM.register(LlavaStableLM_1_6bConfig, LlavaStableLM_1_6bForCausalLM)