Add RoFormer-slav
Browse files
README.md
CHANGED
|
@@ -71,3 +71,4 @@ model = AutoModelForMaskedLM.from_pretrained(
|
|
| 71 |
## Contact
|
| 72 |
|
| 73 |
Maxim Eremeev, maeremeev@edu.hse.ru
|
|
|
|
|
|
| 71 |
## Contact
|
| 72 |
|
| 73 |
Maxim Eremeev, maeremeev@edu.hse.ru
|
| 74 |
+
|
config.json
CHANGED
|
@@ -22,7 +22,7 @@
|
|
| 22 |
"use_cache": true,
|
| 23 |
"vocab_size": 50000,
|
| 24 |
"auto_map": {
|
| 25 |
-
"AutoConfig": "
|
| 26 |
-
"AutoModelForMaskedLM": "
|
| 27 |
}
|
| 28 |
}
|
|
|
|
| 22 |
"use_cache": true,
|
| 23 |
"vocab_size": 50000,
|
| 24 |
"auto_map": {
|
| 25 |
+
"AutoConfig": "config.RoFormerConfig",
|
| 26 |
+
"AutoModelForMaskedLM": "model.RoFormerForMaskedLM"
|
| 27 |
}
|
| 28 |
}
|
config.py
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from transformers import RoFormerConfig
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
def get_roformer_config(vocab_size: int, pad_token_id: int):
|
| 5 |
+
"""
|
| 6 |
+
Returns the RoFormer configuration with parameters optimized for Ancient Russian text.
|
| 7 |
+
"""
|
| 8 |
+
return RoFormerConfig(
|
| 9 |
+
vocab_size=vocab_size,
|
| 10 |
+
embedding_size=512,
|
| 11 |
+
hidden_size=512,
|
| 12 |
+
num_hidden_layers=6,
|
| 13 |
+
num_attention_heads=8,
|
| 14 |
+
intermediate_size=2048,
|
| 15 |
+
max_position_embeddings=514,
|
| 16 |
+
pad_token_id=pad_token_id,
|
| 17 |
+
rotary_value=False,
|
| 18 |
+
)
|
model.py
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from config import get_roformer_config
|
| 2 |
+
from transformers import RoFormerForMaskedLM
|
| 3 |
+
|
| 4 |
+
|
| 5 |
+
def get_model(vocab_size: int, pad_token_id: int):
|
| 6 |
+
"""
|
| 7 |
+
Initializes and returns a RoFormer model for Masked Language Modeling.
|
| 8 |
+
"""
|
| 9 |
+
config = get_roformer_config(vocab_size, pad_token_id)
|
| 10 |
+
model = RoFormerForMaskedLM(config)
|
| 11 |
+
return model
|