Spaces:
Sleeping
Sleeping
adarshj322
commited on
Commit
•
6569da3
1
Parent(s):
6a5e8f0
model changes
Browse files- app.py +69 -2
- requirements.txt +2 -0
app.py
CHANGED
@@ -1,7 +1,74 @@
|
|
1 |
import gradio as gr
|
2 |
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
return "Hello " + name + "!!"
|
5 |
|
6 |
-
iface = gr.Interface(fn=
|
7 |
iface.launch()
|
|
|
1 |
import gradio as gr
|
2 |
|
3 |
+
|
4 |
+
import wandb
|
5 |
+
import torch
|
6 |
+
|
7 |
+
from transformers import GPT2Tokenizer,GPT2LMHeadModel
|
8 |
+
from peft import PeftModel
|
9 |
+
|
10 |
+
import os
|
11 |
+
|
12 |
+
import re
|
13 |
+
|
14 |
+
def clean_text(text):
|
15 |
+
# Lowercase the text
|
16 |
+
text = text.lower()
|
17 |
+
# Remove special characters
|
18 |
+
text = re.sub(r'\W', ' ', text)
|
19 |
+
# Remove extra white spaces
|
20 |
+
text = re.sub(r'\s+', ' ', text).strip()
|
21 |
+
return text
|
22 |
+
|
23 |
+
|
24 |
+
|
25 |
+
os.environ["WANDB_API_KEY"] = "d2ad0a7285379c0808ca816971d965fc242d0b5e"
|
26 |
+
|
27 |
+
wandb.login()
|
28 |
+
|
29 |
+
run = wandb.init(project="Email_subject_gen", job_type="model_loading")
|
30 |
+
|
31 |
+
artifact = run.use_artifact('Email_subject_gen/final_model:v0')
|
32 |
+
artifact_dir = artifact.download()
|
33 |
+
|
34 |
+
#tokenizer= GPT2Tokenizer.from_pretrained(artifact_dir)
|
35 |
+
|
36 |
+
MODEL_KEY = 'olm/olm-gpt2-dec-2022'
|
37 |
+
tokenizer= GPT2Tokenizer.from_pretrained(MODEL_KEY)
|
38 |
+
tokenizer.add_special_tokens({'pad_token':'{PAD}'})
|
39 |
+
|
40 |
+
model = GPT2LMHeadModel.from_pretrained(MODEL_KEY)
|
41 |
+
model.resize_token_embeddings(len(tokenizer))
|
42 |
+
model.config.dropout = 0.1 # Set dropout rate
|
43 |
+
model.config.attention_dropout = 0.1
|
44 |
+
|
45 |
+
model = PeftModel.from_pretrained(model, artifact_dir)
|
46 |
+
|
47 |
+
|
48 |
+
def generateSubject(email):
|
49 |
+
|
50 |
+
clean_text(email)
|
51 |
+
|
52 |
+
email = "<email>" + clean_text(email) + "<subject>"
|
53 |
+
|
54 |
+
prompts = list()
|
55 |
+
prompts.append(email)
|
56 |
+
tokenizer.padding_side='left'
|
57 |
+
prompts_batch_ids = tokenizer(prompts,
|
58 |
+
padding=True, truncation=True, return_tensors='pt').to(model.device)
|
59 |
+
output_ids = model.generate(
|
60 |
+
**prompts_batch_ids, max_new_tokens=10,
|
61 |
+
pad_token_id=tokenizer.pad_token_id)
|
62 |
+
outputs_batch = [seq.split('<subject>')[1] for seq in
|
63 |
+
tokenizer.batch_decode(output_ids, skip_special_tokens=True)]
|
64 |
+
tokenizer.padding_side='right'
|
65 |
+
|
66 |
+
return outputs_batch[0]
|
67 |
+
|
68 |
+
|
69 |
+
|
70 |
+
def predict(name):
|
71 |
return "Hello " + name + "!!"
|
72 |
|
73 |
+
iface = gr.Interface(fn=predict, inputs=gr.inputs.Textbox(), outputs="text")
|
74 |
iface.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
wandb
|
2 |
+
peft
|