Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import gradio as gr
|
3 |
+
import spaces
|
4 |
+
from transformers import AutoTokenizer, AutoModel
|
5 |
+
|
6 |
+
|
7 |
+
model_name = "teknium/OpenHermes-2.5-Mistral-7B"
|
8 |
+
token = os.environ['hf_token']
|
9 |
+
|
10 |
+
model = AutoModel.from_pretrained(model_name).cuda()
|
11 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
12 |
+
|
13 |
+
|
14 |
+
system_prompt = '''You are given an input text for a chat interface. Propose auto-completion to the text. You have several roles:
|
15 |
+
- Fight under-specification: if the user does not provide sufficient context, propose them a set of relevant suggestions.
|
16 |
+
- Complete text: The text provided to you is in the making. If you have a good idea for how to complete - make suggestions.
|
17 |
+
|
18 |
+
Make sure the suggestions are valid completions of the text! No need for them to complete the text completely.
|
19 |
+
Suggest only up to 5 works ahead.
|
20 |
+
'''
|
21 |
+
|
22 |
+
@spaces.GPU
|
23 |
+
def generate(text):
|
24 |
+
data = [
|
25 |
+
{'role': 'system', 'content': system_prompt},
|
26 |
+
{'role': 'user', 'content': text}
|
27 |
+
]
|
28 |
+
tokenized = tokenizer.apply_chat_template(data, return_tensors='pt')
|
29 |
+
return tokenizer.deocode(model.generate(**tokenized).squeeze(0))
|