Upload handler.py
Browse files- handler.py +2 -2
handler.py
CHANGED
|
@@ -28,7 +28,7 @@ class EndpointHandler():
|
|
| 28 |
self.tokenizer.pad_token = '[PAD]'
|
| 29 |
self.tokenizer.padding_side = 'left'
|
| 30 |
|
| 31 |
-
bnb_config = BitsAndBytesConfig(load_in_8bit=True, bnb_8bit_compute_dtype=torch.float16)
|
| 32 |
|
| 33 |
self.model = AutoModel.from_pretrained(
|
| 34 |
model_dir,
|
|
@@ -58,7 +58,7 @@ class EndpointHandler():
|
|
| 58 |
return self.tokenizer(text + self.tokenizer.eos_token, max_length=self.max_length, truncation=True, return_tensors='pt').to(self.device)
|
| 59 |
|
| 60 |
|
| 61 |
-
def extract_attn_vec(self):
|
| 62 |
return self.model._modules['layers'][-1].self_attn.attn_vec
|
| 63 |
|
| 64 |
|
|
|
|
| 28 |
self.tokenizer.pad_token = '[PAD]'
|
| 29 |
self.tokenizer.padding_side = 'left'
|
| 30 |
|
| 31 |
+
#bnb_config = BitsAndBytesConfig(load_in_8bit=True, bnb_8bit_compute_dtype=torch.float16)
|
| 32 |
|
| 33 |
self.model = AutoModel.from_pretrained(
|
| 34 |
model_dir,
|
|
|
|
| 58 |
return self.tokenizer(text + self.tokenizer.eos_token, max_length=self.max_length, truncation=True, return_tensors='pt').to(self.device)
|
| 59 |
|
| 60 |
|
| 61 |
+
def extract_attn_vec(self, model):
|
| 62 |
return self.model._modules['layers'][-1].self_attn.attn_vec
|
| 63 |
|
| 64 |
|