File size: 2,266 Bytes
bf66e5a
366e62e
bf66e5a
 
 
 
 
 
833b301
000ad8b
 
bf66e5a
 
216cf30
bf66e5a
 
4f7f1d5
bf66e5a
 
 
 
 
 
000ad8b
0e224b1
b29c898
0e224b1
 
 
000ad8b
 
 
 
 
 
 
 
02ffbef
e1040a6
000ad8b
cd860a6
000ad8b
5e35500
392d92f
833b301
bf66e5a
 
 
 
 
 
 
 
 
 
366e62e
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
from typing import Dict, List, Any
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, StoppingCriteria, StoppingCriteriaList


class EndpointHandler():
    def __init__(self, path=""):
        # Preload all the elements you are going to need at inference.
        tokenizer = AutoTokenizer.from_pretrained(path)
        tokenizer.pad_token = tokenizer.eos_token
        self.model = AutoModelForCausalLM.from_pretrained(path)
        self.tokenizer = tokenizer
        self.stopping_criteria = StoppingCriteriaList([StopAtPeriodCriteria(tokenizer)])

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
       data args:
            inputs (:obj: `str`)
            kwargs
      Return:
            A :obj:`list` | `dict`: will be serialized and returned
        """
        inputs = data.pop("inputs", data)


        # 3070, 10456, [313, 334] corresponds to "(*", and we do not want to output a comment
        # 13 is a newline character
        # [1976, 441, 29889] is "Abort."
        # [2087, 29885, 4430, 29889] is "Admitted."
        bad_words_ids = [[3070], [313, 334], [10456], [13], [1976, 441, 29889], [2087, 29885, 4430, 29889]]

        input_ids = self.tokenizer.encode(inputs, return_tensors="pt")

        # Generate text using model.generate
        generated_ids = self.model.generate(
            input_ids,
            max_length=input_ids.shape[1] + 50,  # 50 new tokens
            bad_words_ids=bad_words_ids,
            temperature=1,
            top_k=40,
            stopping_criteria=self.stopping_criteria,
        )

        generated_text = self.tokenizer.decode(generated_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
        prediction = [{"generated_text": generated_text, "generated_ids": generated_ids[0][input_ids.shape[1]:].tolist()}]
        return prediction


class StopAtPeriodCriteria(StoppingCriteria):
    def __init__(self, tokenizer):
        self.tokenizer = tokenizer

    def __call__(self, input_ids, scores, **kwargs):
        # Decode the last generated token to text
        last_token_text = self.tokenizer.decode(input_ids[:, -1], skip_special_tokens=True)
        # Check if the decoded text ends with a period
        return '.' in last_token_text