Created handler.py
Browse files- handler.py +50 -0
handler.py
ADDED
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Any, Dict
|
2 |
+
import torch
|
3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
|
4 |
+
|
5 |
+
dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.float16
|
6 |
+
|
7 |
+
class EndpointHandler:
|
8 |
+
def __init__(self, path=""):
|
9 |
+
|
10 |
+
quantization_config = BitsAndBytesConfig(load_in_8bit=True)
|
11 |
+
tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code = True)
|
12 |
+
model = AutoModelForCausalLM.from_pretrained(
|
13 |
+
path,
|
14 |
+
return_dict = True,
|
15 |
+
device_map = "auto",
|
16 |
+
torch_dtype = dtype,
|
17 |
+
trust_remote_code = True,
|
18 |
+
quantization_config=quantization_config
|
19 |
+
)
|
20 |
+
|
21 |
+
gen_config = model.generation_config
|
22 |
+
gen_config.max_new_tokens = 256
|
23 |
+
gen_config.num_return_sequences = 1
|
24 |
+
gen_config.pad_token_id = tokenizer.eos_token_id
|
25 |
+
gen_config.eos_token_id = tokenizer.eos_token_id
|
26 |
+
|
27 |
+
self.generation_config = gen_config
|
28 |
+
|
29 |
+
self.pipeline = pipeline(
|
30 |
+
'text-generation', model=model, tokenizer=tokenizer
|
31 |
+
)
|
32 |
+
|
33 |
+
|
34 |
+
|
35 |
+
def __call__(self, data: Dict[dict, Any]) -> Dict[str, Any]:
|
36 |
+
prompt = data.pop("inputs", data)
|
37 |
+
|
38 |
+
instruction = "Create a list of chords,a corresponding scale to improve with, title, and style along with an example in ABC notation based on this input in JSON format."
|
39 |
+
|
40 |
+
full_prompt = f"""<s>
|
41 |
+
### Instruction:
|
42 |
+
{instruction}
|
43 |
+
### Input:
|
44 |
+
{prompt}
|
45 |
+
### Response:
|
46 |
+
"""
|
47 |
+
|
48 |
+
result = self.pipeline(full_prompt, generation_config = self.generation_config)[0]
|
49 |
+
|
50 |
+
return result
|