File size: 2,205 Bytes
bf66e5a 366e62e bf66e5a 366e62e bf66e5a 4f7f1d5 bf66e5a 366e62e bf66e5a d1e4ec5 366e62e cd860a6 366e62e 02ffbef e1040a6 366e62e cd860a6 366e62e 2872c1c 366e62e 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 |
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)
self.tokenizer = tokenizer
self.model = AutoModelForCausalLM.from_pretrained(path)
self.tokenizer.pad_token = tokenizer.eos_token
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)
input_ids = self.tokenizer.encode(inputs, return_tensors="pt")
# Bad word: id 3070 corresponds to "(*", and we do not want to output a comment
prediction_ids = self.model.generate(
input_ids,
max_length=input_ids.shape[1] + 50,
stopping_criteria=self.stopping_criteria,
bad_words_ids=[[3070], [313, 334]],
temperature=1,
top_k=40,
# pad_token_id=self.tokenizer.eos_token_id,
# return_dict_in_generate=True, # To get more detailed output (optional)
)
# Decode the generated ids to text
# Exclude the input_ids length to get only the new tokens
print("Generated IDs:", prediction_ids[0, input_ids.shape[1]:])
prediction_text = self.tokenizer.decode(prediction_ids[0, input_ids.shape[1]:], skip_special_tokens=True)
return [{"generated_text": prediction_text}]
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 |