Update handler.py
Browse files- handler.py +18 -6
handler.py
CHANGED
@@ -20,21 +20,33 @@ class EndpointHandler():
|
|
20 |
|
21 |
self.tokenizer = transformers.AutoTokenizer.from_pretrained('EleutherAI/gpt-neox-20b')
|
22 |
print("tokenizer created ", datetime.now())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
self.generate_text = transformers.pipeline(
|
24 |
model=self.model,
|
25 |
tokenizer=self.tokenizer,
|
|
|
26 |
task='text-generation',
|
27 |
return_full_text=True,
|
28 |
temperature=0.1,
|
29 |
top_p=0.15,
|
30 |
top_k=0,
|
31 |
-
|
32 |
repetition_penalty=1.1
|
33 |
)
|
34 |
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
print(inputs)
|
39 |
-
res = self.generate_text("Explain to me the difference between nuclear fission and fusion." , max_length= 60)
|
40 |
return res
|
|
|
20 |
|
21 |
self.tokenizer = transformers.AutoTokenizer.from_pretrained('EleutherAI/gpt-neox-20b')
|
22 |
print("tokenizer created ", datetime.now())
|
23 |
+
|
24 |
+
|
25 |
+
stop_token_ids = self.tokenizer.convert_tokens_to_ids(["<|endoftext|>"])
|
26 |
+
|
27 |
+
class StopOnTokens(StoppingCriteria):
|
28 |
+
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs):
|
29 |
+
for stop_id in stop_token_ids:
|
30 |
+
if input_ids[0][-1] == stop_id:
|
31 |
+
return True
|
32 |
+
return False
|
33 |
+
|
34 |
+
stopping_criteria = StoppingCriteriaList([StopOnTokens()])
|
35 |
+
|
36 |
self.generate_text = transformers.pipeline(
|
37 |
model=self.model,
|
38 |
tokenizer=self.tokenizer,
|
39 |
+
stopping_criteria=stopping_criteria,
|
40 |
task='text-generation',
|
41 |
return_full_text=True,
|
42 |
temperature=0.1,
|
43 |
top_p=0.15,
|
44 |
top_k=0,
|
45 |
+
max_new_tokens=64,
|
46 |
repetition_penalty=1.1
|
47 |
)
|
48 |
|
49 |
+
|
50 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
51 |
+
res = self.generate_text("Explain to me the difference between nuclear fission and fusion.")
|
|
|
|
|
52 |
return res
|