Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pathlib
|
2 |
+
import gradio as gr
|
3 |
+
import transformers
|
4 |
+
from transformers import AutoTokenizer
|
5 |
+
from transformers import ModelForCausalLM
|
6 |
+
from transformers import GenerationConfig
|
7 |
+
from typing import List, Dict, Union
|
8 |
+
from typing import Any, TypeVar
|
9 |
+
|
10 |
+
Pathable = Union[str, pathlib.Path]
|
11 |
+
|
12 |
+
def load_model(name: str) -> Any:
|
13 |
+
return ModelForCausalLM.from_pretrained(name)
|
14 |
+
|
15 |
+
def load_tokenizer(name: str) -> Any:
|
16 |
+
return AutoTokenizer.from_pretrained(name)
|
17 |
+
|
18 |
+
def create_generator():
|
19 |
+
return GenerationConfig(
|
20 |
+
temperature=1.0,
|
21 |
+
top_p=0.75,
|
22 |
+
num_beams=4,
|
23 |
+
)
|
24 |
+
|
25 |
+
def generate_prompt(instruction, input=None):
|
26 |
+
if input:
|
27 |
+
return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
|
28 |
+
|
29 |
+
### Instruction:
|
30 |
+
{instruction}
|
31 |
+
|
32 |
+
### Input:
|
33 |
+
{input}
|
34 |
+
|
35 |
+
### Response:"""
|
36 |
+
else:
|
37 |
+
return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
|
38 |
+
|
39 |
+
### Instruction:
|
40 |
+
{instruction}
|
41 |
+
|
42 |
+
### Response:"""
|
43 |
+
|
44 |
+
|
45 |
+
|
46 |
+
|
47 |
+
def evaluate(instruction, input=None):
|
48 |
+
prompt = generate_prompt(instruction, input)
|
49 |
+
inputs = tokenizer(prompt, return_tensors="pt")
|
50 |
+
input_ids = inputs["input_ids"].cuda()
|
51 |
+
generation_output = model.generate(
|
52 |
+
input_ids=input_ids,
|
53 |
+
generation_config=generation_config,
|
54 |
+
return_dict_in_generate=True,
|
55 |
+
output_scores=True,
|
56 |
+
max_new_tokens=256
|
57 |
+
)
|
58 |
+
for s in generation_output.sequences:
|
59 |
+
output = tokenizer.decode(s)
|
60 |
+
print("Response:", output.split("### Response:")[1].strip())
|