Upload configuration_prot2text.py
Browse files- configuration_prot2text.py +73 -0
configuration_prot2text.py
ADDED
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
""" Prot2Text configuration"""
|
2 |
+
|
3 |
+
from transformers.configuration_utils import PretrainedConfig, AutoConfig
|
4 |
+
from transformers.utils import logging
|
5 |
+
|
6 |
+
|
7 |
+
logger = logging.get_logger(__name__)
|
8 |
+
|
9 |
+
|
10 |
+
class Prot2TextConfig(PretrainedConfig):
|
11 |
+
model_type = "prot2text"
|
12 |
+
keys_to_ignore_at_inference = ["past_key_values"]
|
13 |
+
_keys_to_ignore_on_load_missing = [r"transformer"]
|
14 |
+
|
15 |
+
def __init__(
|
16 |
+
self,
|
17 |
+
cross_esm_graph=True,
|
18 |
+
decoder_start_token_id=50257,
|
19 |
+
early_stopping=True,
|
20 |
+
eos_token_id=50258,
|
21 |
+
bos_token_id=50257,
|
22 |
+
esm=True,
|
23 |
+
esm_model_name="facebook/esm2_t6_8M_UR50D",
|
24 |
+
gpt_model_name="gpt2",
|
25 |
+
length_penalty=2.0,
|
26 |
+
max_new_tokens=256,
|
27 |
+
no_repeat_ngram_size=3,
|
28 |
+
pad_token_id=50256,
|
29 |
+
prot2text_version="1.1",
|
30 |
+
rgcn=True,
|
31 |
+
rgc_input_dim=67,
|
32 |
+
rgcn_n_layers=6,
|
33 |
+
gpt_config=None,
|
34 |
+
esm_config=None,
|
35 |
+
**kwargs,
|
36 |
+
):
|
37 |
+
self.cross_esm_graph = cross_esm_graph
|
38 |
+
self.decoder_start_token_id = decoder_start_token_id
|
39 |
+
self.early_stopping = early_stopping
|
40 |
+
self.eos_token_id = eos_token_id
|
41 |
+
self.esm = esm
|
42 |
+
self.esm_model_name = esm_model_name
|
43 |
+
self.gpt_model_name = gpt_model_name
|
44 |
+
self.length_penalty = length_penalty
|
45 |
+
self.max_new_tokens = max_new_tokens
|
46 |
+
self.no_repeat_ngram_size = no_repeat_ngram_size
|
47 |
+
self.pad_token_id = pad_token_id
|
48 |
+
self.prot2text_version = prot2text_version
|
49 |
+
self.rgcn = rgcn
|
50 |
+
self.rgc_input_dim = rgc_input_dim
|
51 |
+
self.rgcn_n_layers = rgcn_n_layers
|
52 |
+
if gpt_config is None:
|
53 |
+
self.gpt_config = AutoConfig.from_pretrained(gpt_model_name,
|
54 |
+
_name_or_path= gpt_model_name,
|
55 |
+
is_encoder_decoder=True,
|
56 |
+
use_cache=False,
|
57 |
+
add_cross_attention=True,
|
58 |
+
bos_token_id=bos_token_id,
|
59 |
+
decoder_start_token_id=decoder_start_token_id,
|
60 |
+
eos_token_id=eos_token_id,
|
61 |
+
max_new_tokens=max_new_tokens,
|
62 |
+
pad_token_id=50256,
|
63 |
+
vocab_size=50259,
|
64 |
+
num_beams=1,
|
65 |
+
max_length=256,
|
66 |
+
min_length=1).to_dict()
|
67 |
+
else:
|
68 |
+
self.gpt_config = gpt_config
|
69 |
+
if esm_config is None:
|
70 |
+
self.esm_config = AutoConfig.from_pretrained(esm_model_name).to_dict()
|
71 |
+
self.esm_config = esm_config
|
72 |
+
|
73 |
+
super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
|