Commit
·
71fe2f3
1
Parent(s):
94eabde
Delete handler.py
Browse files- handler.py +0 -70
handler.py
DELETED
@@ -1,70 +0,0 @@
|
|
1 |
-
from typing import Dict, Any, List
|
2 |
-
import logging
|
3 |
-
|
4 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer
|
5 |
-
from peft import PeftConfig, PeftModel
|
6 |
-
import torch.cuda
|
7 |
-
|
8 |
-
dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0]==8 else torch.float16
|
9 |
-
|
10 |
-
# LOGGER = logging.getLogger(__name__)
|
11 |
-
# logging.basicConfig(level=logging.INFO)
|
12 |
-
# device = "cuda" if torch.cuda.is_available() else "cpu"
|
13 |
-
|
14 |
-
|
15 |
-
class EndpointHandler():
|
16 |
-
def __init__(self, path=""):
|
17 |
-
tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True)
|
18 |
-
model = AutoModelForCausalLM.from_pretrained(
|
19 |
-
path,
|
20 |
-
return_dict=True,
|
21 |
-
device_map="auto",
|
22 |
-
load_in_8bit=True,
|
23 |
-
torch_dtype=dtype,
|
24 |
-
trust_remote_code=True,
|
25 |
-
)
|
26 |
-
|
27 |
-
generation_config = model.generation_config
|
28 |
-
generation_config.max_new_tokens=512
|
29 |
-
generation_config.temperation = 0
|
30 |
-
generation_config.num_return_sequences=1
|
31 |
-
generation_config.pad_token_id = tokenizer.eos_token_id
|
32 |
-
generation_config.eos_token_id = tokenizer.eos_token_id
|
33 |
-
self.generation_config = generation_config
|
34 |
-
|
35 |
-
self.pipeline = transformers.pipeline(
|
36 |
-
"text-generation",model=model,tokenizer=tokenizer
|
37 |
-
)
|
38 |
-
|
39 |
-
|
40 |
-
# config = PeftConfig.from_pretrained(path)
|
41 |
-
# model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, load_in_8bit=True, device_map='auto')
|
42 |
-
# self.tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
|
43 |
-
# # Load the Lora model
|
44 |
-
# self.model = PeftModel.from_pretrained(model, path)
|
45 |
-
|
46 |
-
def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
|
47 |
-
# """
|
48 |
-
# Args:
|
49 |
-
# data (Dict): The payload with the text prompt and generation parameters.
|
50 |
-
# """
|
51 |
-
# LOGGER.info(f"Received data: {data}")
|
52 |
-
# Get inputs
|
53 |
-
prompt = data.pop("inputs", None)
|
54 |
-
# parameters = data.pop("parameters", None)
|
55 |
-
# if prompt is None:
|
56 |
-
# raise ValueError("Missing prompt.")
|
57 |
-
# # Preprocess
|
58 |
-
# input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(device)
|
59 |
-
# # Forward
|
60 |
-
# LOGGER.info(f"Start generation.")
|
61 |
-
# if parameters is not None:
|
62 |
-
# output = self.model.generate(input_ids=input_ids, **parameters)
|
63 |
-
# else:
|
64 |
-
# output = self.model.generate(input_ids=input_ids)
|
65 |
-
# # Postprocess
|
66 |
-
# prediction = self.tokenizer.decode(output[0])
|
67 |
-
# LOGGER.info(f"Generated text: {prediction}")
|
68 |
-
# return {"generated_text": prediction}
|
69 |
-
result = self.pipeline(prompt,generation_config=self.generation_config)
|
70 |
-
return result
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|