Upload code
Browse files- configuration_prot2text.py +74 -0
- modeling_prot2text.py +200 -0
configuration_prot2text.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
""" Prot2Text configuration"""
|
2 |
+
|
3 |
+
from transformers.configuration_utils import PretrainedConfig
|
4 |
+
from transformers import AutoConfig
|
5 |
+
from transformers.utils import logging
|
6 |
+
|
7 |
+
|
8 |
+
logger = logging.get_logger(__name__)
|
9 |
+
|
10 |
+
|
11 |
+
class Prot2TextConfig(PretrainedConfig):
|
12 |
+
model_type = "prot2text"
|
13 |
+
keys_to_ignore_at_inference = ["past_key_values"]
|
14 |
+
_keys_to_ignore_on_load_missing = [r"transformer"]
|
15 |
+
|
16 |
+
def __init__(
|
17 |
+
self,
|
18 |
+
cross_esm_graph=True,
|
19 |
+
decoder_start_token_id=50257,
|
20 |
+
early_stopping=True,
|
21 |
+
eos_token_id=50258,
|
22 |
+
bos_token_id=50257,
|
23 |
+
esm=True,
|
24 |
+
esm_model_name="facebook/esm2_t6_8M_UR50D",
|
25 |
+
gpt_model_name="gpt2",
|
26 |
+
length_penalty=2.0,
|
27 |
+
max_new_tokens=256,
|
28 |
+
no_repeat_ngram_size=3,
|
29 |
+
pad_token_id=50256,
|
30 |
+
prot2text_version="1.1",
|
31 |
+
rgcn=True,
|
32 |
+
rgc_input_dim=67,
|
33 |
+
rgcn_n_layers=6,
|
34 |
+
gpt_config=None,
|
35 |
+
esm_config=None,
|
36 |
+
**kwargs,
|
37 |
+
):
|
38 |
+
self.cross_esm_graph = cross_esm_graph
|
39 |
+
self.decoder_start_token_id = decoder_start_token_id
|
40 |
+
self.early_stopping = early_stopping
|
41 |
+
self.eos_token_id = eos_token_id
|
42 |
+
self.esm = esm
|
43 |
+
self.esm_model_name = esm_model_name
|
44 |
+
self.gpt_model_name = gpt_model_name
|
45 |
+
self.length_penalty = length_penalty
|
46 |
+
self.max_new_tokens = max_new_tokens
|
47 |
+
self.no_repeat_ngram_size = no_repeat_ngram_size
|
48 |
+
self.pad_token_id = pad_token_id
|
49 |
+
self.prot2text_version = prot2text_version
|
50 |
+
self.rgcn = rgcn
|
51 |
+
self.rgc_input_dim = rgc_input_dim
|
52 |
+
self.rgcn_n_layers = rgcn_n_layers
|
53 |
+
if gpt_config is None:
|
54 |
+
self.gpt_config = AutoConfig.from_pretrained(gpt_model_name,
|
55 |
+
_name_or_path= gpt_model_name,
|
56 |
+
is_encoder_decoder=True,
|
57 |
+
use_cache=False,
|
58 |
+
add_cross_attention=True,
|
59 |
+
bos_token_id=bos_token_id,
|
60 |
+
decoder_start_token_id=decoder_start_token_id,
|
61 |
+
eos_token_id=eos_token_id,
|
62 |
+
max_new_tokens=max_new_tokens,
|
63 |
+
pad_token_id=50256,
|
64 |
+
vocab_size=50259,
|
65 |
+
num_beams=1,
|
66 |
+
max_length=256,
|
67 |
+
min_length=1).to_dict()
|
68 |
+
else:
|
69 |
+
self.gpt_config = gpt_config
|
70 |
+
if esm_config is None:
|
71 |
+
self.esm_config = AutoConfig.from_pretrained(esm_model_name).to_dict()
|
72 |
+
self.esm_config = esm_config
|
73 |
+
|
74 |
+
super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
|
modeling_prot2text.py
ADDED
@@ -0,0 +1,200 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import GPT2Config, AutoTokenizer, GPT2Config
|
2 |
+
from transformers import PretrainedConfig, PreTrainedModel
|
3 |
+
import transformers
|
4 |
+
from typing import Optional, Tuple, Callable
|
5 |
+
import torch
|
6 |
+
import torch.nn as nn
|
7 |
+
from transformers.modeling_utils import PreTrainedModel, PretrainedConfig
|
8 |
+
from .utils import CABlock, _GPT2LMHeadModel
|
9 |
+
from .configuration_prot2text import Prot2TextConfig
|
10 |
+
from transformers.generation.configuration_utils import GenerationConfig
|
11 |
+
from transformers.generation.logits_process import LogitsProcessorList
|
12 |
+
from transformers.generation.stopping_criteria import StoppingCriteriaList
|
13 |
+
|
14 |
+
|
15 |
+
class Prot2TextModel(PreTrainedModel):
|
16 |
+
config_class = Prot2TextConfig
|
17 |
+
_keys_to_ignore_on_load_missing = [r"transformer"]
|
18 |
+
base_model_prefix = "decoder"
|
19 |
+
def __init__(self, config):
|
20 |
+
super().__init__(config)
|
21 |
+
|
22 |
+
self.gpt_config = GPT2Config.from_dict(config.gpt_config)
|
23 |
+
|
24 |
+
# define the GPT2 decoder
|
25 |
+
self.decoder = _GPT2LMHeadModel(self.gpt_config)
|
26 |
+
|
27 |
+
# if using ESM to encode protein's sequence, define the ESM layer, the Projection layer and the fusion layer
|
28 |
+
if config.esm:
|
29 |
+
self.esm_config = PretrainedConfig.from_dict(config.esm_config)
|
30 |
+
self.esm = transformers.EsmModel(self.esm_config)
|
31 |
+
self.to_embedding = nn.Linear(self.esm_config.hidden_size, self.gpt_config.n_embd)
|
32 |
+
if config.cross_esm_graph and config.rgcn:
|
33 |
+
self.h = nn.ModuleList([CABlock(self.gpt_config, layer_idx=i) for i in range(4)])
|
34 |
+
self.ln_f = nn.LayerNorm(self.gpt_config.n_embd, eps=self.gpt_config.layer_norm_epsilon)
|
35 |
+
|
36 |
+
self.config = config
|
37 |
+
|
38 |
+
|
39 |
+
def get_encoder(self):
|
40 |
+
return self.encoder
|
41 |
+
|
42 |
+
def get_decoder(self):
|
43 |
+
return self.decoder
|
44 |
+
|
45 |
+
def get_input_embeddings(self):
|
46 |
+
if hasattr(self, "transformer"):
|
47 |
+
return self.transformer.wte
|
48 |
+
return self.decoder.transformer.wte
|
49 |
+
|
50 |
+
def warm_up(self, gpt_model=None, esm_model=None):
|
51 |
+
if esm_model is not None:
|
52 |
+
self.esm = transformers.EsmModel.from_pretrained(esm_model)
|
53 |
+
if gpt_model is not None:
|
54 |
+
self.decoder = _GPT2LMHeadModel.from_pretrained(gpt_model, add_cross_attention=True, use_cache=False)
|
55 |
+
self.decoder.resize_token_embeddings(self.gpt_config.vocab_size)
|
56 |
+
self.decoder.config = self.gpt_config
|
57 |
+
|
58 |
+
|
59 |
+
def forward(self,
|
60 |
+
encoder_input_ids: Optional[torch.LongTensor] = None,
|
61 |
+
edge_index: Optional[torch.LongTensor] = None,
|
62 |
+
batch: Optional[torch.LongTensor] = None,
|
63 |
+
x: Optional[torch.FloatTensor] = None,
|
64 |
+
edge_type: Optional[torch.LongTensor] = None,
|
65 |
+
decoder_input_ids: Optional[torch.LongTensor] = None,
|
66 |
+
past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
|
67 |
+
past_key_values_graph_esm: Optional[Tuple[Tuple[torch.Tensor]]] = None,
|
68 |
+
decoder_attention_mask: Optional[torch.FloatTensor] = None,
|
69 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
70 |
+
token_type_ids: Optional[torch.LongTensor] = None,
|
71 |
+
position_ids: Optional[torch.LongTensor] = None,
|
72 |
+
head_mask: Optional[torch.FloatTensor] = None,
|
73 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
74 |
+
encoder_hidden_states: Optional[torch.Tensor] = None,
|
75 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
76 |
+
labels: Optional[torch.LongTensor] = None,
|
77 |
+
use_cache: Optional[bool] = None,
|
78 |
+
output_attentions: Optional[bool] = None,
|
79 |
+
output_hidden_states: Optional[bool] = None,
|
80 |
+
return_dict: Optional[bool] = None,
|
81 |
+
get_graph_emb: Optional[bool] = False,
|
82 |
+
**delete_args,
|
83 |
+
):
|
84 |
+
use_cache = use_cache if use_cache is not None else self.gpt_config.use_cache
|
85 |
+
return_dict = return_dict if return_dict is not None else self.gpt_config.use_return_dict
|
86 |
+
|
87 |
+
|
88 |
+
if decoder_input_ids is not None and len(decoder_input_ids.size()) == 3:
|
89 |
+
decoder_input_ids = decoder_input_ids.squeeze(0)
|
90 |
+
|
91 |
+
if self.config.esm:
|
92 |
+
if self.config.prot2text_version=='1.0':
|
93 |
+
if encoder_input_ids.size()[1] != 1021:
|
94 |
+
raise ValueError("For this version of the model you need to PAD/Truncate the amino acid sequence for the ESM model to 1021")
|
95 |
+
|
96 |
+
esm_emb = self.esm(input_ids=encoder_input_ids, attention_mask=attention_mask, return_dict=return_dict).last_hidden_state
|
97 |
+
esm_emb = self.to_embedding(esm_emb)
|
98 |
+
graph_emb = esm_emb
|
99 |
+
else:
|
100 |
+
attention_mask = None
|
101 |
+
if self.config.prot2text_version=='1.0':
|
102 |
+
attention_mask = None
|
103 |
+
if get_graph_emb:
|
104 |
+
return graph_emb
|
105 |
+
|
106 |
+
transformer_outputs = self.decoder(input_ids=decoder_input_ids,
|
107 |
+
past_key_values=past_key_values,
|
108 |
+
attention_mask=decoder_attention_mask,
|
109 |
+
token_type_ids=token_type_ids,
|
110 |
+
position_ids=position_ids,
|
111 |
+
head_mask=head_mask,
|
112 |
+
inputs_embeds=inputs_embeds,
|
113 |
+
encoder_hidden_states=graph_emb,
|
114 |
+
encoder_attention_mask=attention_mask,
|
115 |
+
labels=labels,
|
116 |
+
use_cache=use_cache,
|
117 |
+
output_attentions=output_attentions,
|
118 |
+
output_hidden_states=output_hidden_states,
|
119 |
+
return_dict=return_dict,
|
120 |
+
)
|
121 |
+
|
122 |
+
return transformer_outputs
|
123 |
+
|
124 |
+
@torch.no_grad()
|
125 |
+
def generate_protein_description(self,
|
126 |
+
protein_pdbID=None,
|
127 |
+
protein_sequence=None,
|
128 |
+
edge_index: Optional[torch.LongTensor] = None,
|
129 |
+
x: Optional[torch.FloatTensor] = None,
|
130 |
+
edge_type: Optional[torch.LongTensor] = None,
|
131 |
+
tokenizer=None,
|
132 |
+
device='cpu'
|
133 |
+
):
|
134 |
+
|
135 |
+
if self.config.esm and not self.config.rgcn and protein_sequence==None:
|
136 |
+
raise ValueError(
|
137 |
+
"The model you are trying to use is based only on protein sequence, please provide an amino-acid protein_sequence"
|
138 |
+
)
|
139 |
+
if self.config.rgcn and protein_pdbID==None and (x==None or edge_index==None or edge_type==None):
|
140 |
+
raise ValueError(
|
141 |
+
"The model you are trying to use is based on protein structure, please provide a AlphaFold ID (you must have to have internet connection using protein_pdbID, or provide the triplet inputs: x (node features), edge_index and edge_type"
|
142 |
+
)
|
143 |
+
if self.config.esm:
|
144 |
+
esmtokenizer = AutoTokenizer.from_pretrained(self.config.esm_model_name)
|
145 |
+
|
146 |
+
if protein_pdbID==None and protein_sequence==None:
|
147 |
+
raise ValueError(
|
148 |
+
"you need to provide either a protein AlphaFold Id or an amino-acid sequence"
|
149 |
+
)
|
150 |
+
|
151 |
+
|
152 |
+
seq = esmtokenizer([protein_sequence], add_special_tokens=True, truncation=True, max_length=1021, padding='max_length', return_tensors="pt")
|
153 |
+
inputs={}
|
154 |
+
inputs['encoder_input_ids'] = seq['input_ids']
|
155 |
+
inputs['attention_mask'] = seq['attention_mask']
|
156 |
+
inputs['decoder_input_ids'] = inputs['encoder_input_ids'][:,0:1].clone()
|
157 |
+
inputs['decoder_input_ids'][:,0] = tokenizer.bos_token_id
|
158 |
+
|
159 |
+
self.to(device)
|
160 |
+
inputs = {k: v.to(device=device, non_blocking=True) if hasattr(v, 'to') else v for k, v in inputs.items()}
|
161 |
+
encoder_state = dict()
|
162 |
+
encoder_state['hidden_states'] = self(**inputs, get_graph_emb=True, output_attentions=True)
|
163 |
+
generated = tokenizer.batch_decode(self.decoder.generate(input_ids=inputs['decoder_input_ids'], encoder_outputs=encoder_state, use_cache=True), skip_special_tokens=True)
|
164 |
+
|
165 |
+
return generated[0].replace('<|stop_token|>', '').replace('<|graph_token|>', '')
|
166 |
+
|
167 |
+
@torch.no_grad()
|
168 |
+
def generate(self,
|
169 |
+
inputs: Optional[torch.Tensor] = None,
|
170 |
+
generation_config: Optional[GenerationConfig] = None,
|
171 |
+
logits_processor: Optional[LogitsProcessorList] = None,
|
172 |
+
stopping_criteria: Optional[StoppingCriteriaList] = None,
|
173 |
+
prefix_allowed_tokens_fn: Optional[Callable[[int, torch.Tensor], List[int]]] = None,
|
174 |
+
synced_gpus: Optional[bool] = None,
|
175 |
+
assistant_model: Optional["PreTrainedModel"] = None,
|
176 |
+
streamer: Optional["BaseStreamer"] = None,
|
177 |
+
**kwargs,
|
178 |
+
):
|
179 |
+
encoder_state = self(**kwargs, get_graph_emb=True)
|
180 |
+
input_ids = kwargs['decoder_input_ids']
|
181 |
+
attention_mask = kwargs['decoder_attention_mask']
|
182 |
+
kwargs['encoder_attention_mask'] = kwargs['attention_mask']
|
183 |
+
if not self.config.cross_esm_graph and self.config.rgcn and self.config.esm:
|
184 |
+
t_add = torch.ones((kwargs['encoder_attention_mask'].size(0), 1)).to(kwargs['encoder_attention_mask'].get_device())
|
185 |
+
kwargs['encoder_attention_mask'] = torch.cat((t_add, kwargs['encoder_attention_mask']), dim=1)
|
186 |
+
for key in ['edge_index', 'edge_type', 'x', 'encoder_input_ids', 'decoder_input_ids', 'decoder_attention_mask', 'batch', 'attention_mask', 'max_length',
|
187 |
+
'_num_nodes', 'node_id', 'name', 'sequence', 'distance_matrix', 'distance', 'coordinates', 'ptr', 'num_nodes',]:
|
188 |
+
if key in kwargs.keys():
|
189 |
+
kwargs.pop(key)
|
190 |
+
return self.decoder.generate(input_ids=input_ids,
|
191 |
+
generation_config=generation_config,
|
192 |
+
logits_processor=logits_processor,
|
193 |
+
stopping_criteria=stopping_criteria,
|
194 |
+
prefix_allowed_tokens_fn=prefix_allowed_tokens_fn,
|
195 |
+
synced_gpus=synced_gpus,
|
196 |
+
assistant_model=assistant_model,
|
197 |
+
streamer=streamer,
|
198 |
+
encoder_outputs={'hidden_states': encoder_state, 'attentions':0},
|
199 |
+
**kwargs
|
200 |
+
)
|