Spaces:
Sleeping
Sleeping
improve model loading
Browse files- tasks/text.py +5 -28
tasks/text.py
CHANGED
@@ -37,38 +37,16 @@ class TextClassifier:
|
|
37 |
|
38 |
try:
|
39 |
# Initialize tokenizer
|
40 |
-
self.tokenizer = AutoTokenizer.from_pretrained(
|
41 |
-
TOKENIZER_NAME,
|
42 |
-
model_max_length=8192,
|
43 |
-
padding_side='right',
|
44 |
-
truncation_side='right'
|
45 |
-
)
|
46 |
-
|
47 |
-
# Load model configuration
|
48 |
-
model_config = {
|
49 |
-
"architectures": ["ModernBertForSequenceClassification"],
|
50 |
-
"model_type": "modernbert",
|
51 |
-
"num_labels": 8,
|
52 |
-
"problem_type": "single_label_classification",
|
53 |
-
"hidden_size": 768,
|
54 |
-
"num_attention_heads": 12,
|
55 |
-
"num_hidden_layers": 22,
|
56 |
-
"intermediate_size": 1152,
|
57 |
-
"max_position_embeddings": 8192,
|
58 |
-
"torch_dtype": "float32",
|
59 |
-
"transformers_version": "4.48.3",
|
60 |
-
"layer_norm_eps": 1e-05
|
61 |
-
}
|
62 |
|
63 |
# Initialize model
|
64 |
-
self.model =
|
65 |
MODEL_NAME,
|
66 |
-
|
67 |
-
ignore_mismatched_sizes=True
|
68 |
-
trust_remote_code=True
|
69 |
).to(self.device)
|
70 |
|
71 |
-
# Convert to half precision
|
72 |
self.model = self.model.half()
|
73 |
self.model.eval()
|
74 |
|
@@ -79,7 +57,6 @@ class TextClassifier:
|
|
79 |
raise
|
80 |
|
81 |
def process_batch(self, batch):
|
82 |
-
"""Process a batch of texts and return their predictions"""
|
83 |
try:
|
84 |
# Move batch to device
|
85 |
input_ids = batch['input_ids'].to(self.device)
|
|
|
37 |
|
38 |
try:
|
39 |
# Initialize tokenizer
|
40 |
+
self.tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
# Initialize model
|
43 |
+
self.model = BertForSequenceClassification.from_pretrained(
|
44 |
MODEL_NAME,
|
45 |
+
num_labels=8,
|
46 |
+
ignore_mismatched_sizes=True
|
|
|
47 |
).to(self.device)
|
48 |
|
49 |
+
# Convert to half precision and eval mode
|
50 |
self.model = self.model.half()
|
51 |
self.model.eval()
|
52 |
|
|
|
57 |
raise
|
58 |
|
59 |
def process_batch(self, batch):
|
|
|
60 |
try:
|
61 |
# Move batch to device
|
62 |
input_ids = batch['input_ids'].to(self.device)
|