Spaces:
Running
on
Zero
Running
on
Zero
File size: 7,708 Bytes
d9c19b7 |
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 |
from typing import List, Optional, Tuple, Union
import torch
import torch.nn as nn
from transformers import AutoConfig, AutoModelForCausalLM, \
LlamaConfig, LlamaModel, LlamaForCausalLM
from transformers.modeling_outputs import CausalLMOutputWithPast
from transformers.generation.utils import GenerateOutput
from oryx.model.oryx_arch import OryxMetaModel, OryxMetaForCausalLM
class OryxConfig(LlamaConfig):
model_type = "oryx_llama"
class OryxLlamaModel(OryxMetaModel, LlamaModel):
config_class = OryxConfig
def __init__(self, config: LlamaConfig):
super(OryxLlamaModel, self).__init__(config)
class OryxLlamaForCausalLM(LlamaForCausalLM, OryxMetaForCausalLM):
config_class = OryxConfig
def __init__(self, config):
LlamaForCausalLM.__init__(self, config)
self.model = OryxLlamaModel(config)
self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
# Initialize weights and apply final processing
self.post_init()
def get_model(self):
return self.model
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,
images_highres: Optional[List[torch.FloatTensor]] = None,
image_sizes: Optional[List[List[int]]] = None,
return_dict: Optional[bool] = None,
modalities: Optional[List[str]] = ["image"],
) -> 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(input_ids, position_ids, attention_mask, past_key_values, labels, images,
modalities, image_sizes, images_highres)
if labels is None:
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,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict
)
else:
return self.forward_llm_efficient(
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 forward_llm_efficient(self, input_ids, attention_mask, position_ids, past_key_values, inputs_embeds, labels, use_cache, output_attentions, output_hidden_states, return_dict):
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
# decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
outputs = self.model(
input_ids=input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
past_key_values=past_key_values,
inputs_embeds=inputs_embeds,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
hidden_states = outputs[0]
hidden_dim = hidden_states.size(-1)
shift_labels = labels[..., 1:].contiguous().reshape(-1)
shift_hidden_states = hidden_states[..., :-1, :].contiguous().reshape(-1, hidden_dim)
assert shift_labels.size(0) == shift_hidden_states.size(0)
mask = shift_labels > -1
seen_tokens = mask.float().sum().item()
if not seen_tokens > 0:
logits = self.lm_head(shift_hidden_states[0:2])
loss = logits.sum() * 0
print("No tokens seen")
print(shift_labels)
else:
shift_labels = shift_labels[mask]
shift_hidden_states = shift_hidden_states[mask, :]
logits = self.lm_head(shift_hidden_states)
logits = logits.float()
loss_fct = nn.CrossEntropyLoss()
loss = loss_fct(logits, shift_labels)
if not return_dict:
output = (logits,) + outputs[1:]
return (loss,) + output if loss is not None else output
return CausalLMOutputWithPast(
loss=loss,
logits=logits,
past_key_values=outputs.past_key_values,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
@torch.no_grad()
def generate(
self,
inputs: Optional[torch.Tensor] = None,
images: Optional[torch.Tensor] = None,
image_sizes: Optional[torch.Tensor] = None,
**kwargs,
) -> Union[GenerateOutput, torch.LongTensor]:
modalities = kwargs.pop("modalities", None)
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(
inputs,
position_ids,
attention_mask,
None,
None,
images,
modalities,
image_sizes=image_sizes
)
else:
inputs_embeds = self.get_model().embed_tokens(inputs)
return super().generate(
position_ids=position_ids,
attention_mask=attention_mask,
inputs_embeds=inputs_embeds,
**kwargs
)
def prepare_inputs_for_generation(self, input_ids, past_key_values=None,
inputs_embeds=None, **kwargs):
images = kwargs.pop("images", None)
image_sizes = kwargs.pop("image_sizes", 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
if image_sizes is not None:
inputs['image_sizes'] = image_sizes
return inputs
if OryxConfig.model_type == "oryx":
OryxConfig.model_type = "oryx_llama" # directly set to Oryx_dev to avoid conflict with HF's Oryx
AutoConfig.register("oryx_llama", OryxConfig)
AutoModelForCausalLM.register(OryxConfig, OryxLlamaForCausalLM)
|