Spaces:
Runtime error
Runtime error
shrimantasatpati
commited on
Commit
•
92df928
1
Parent(s):
c4be01c
Updated app.py
Browse files
app.py
CHANGED
@@ -1,12 +1,6 @@
|
|
1 |
import streamlit as st
|
2 |
-
import transformers
|
3 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
4 |
import torch
|
5 |
-
# device = "cpu"
|
6 |
-
# if torch.cuda.is_available():
|
7 |
-
# device = "cuda"
|
8 |
-
# if torch.backends.mps.is_available():
|
9 |
-
# device = "mps"
|
10 |
|
11 |
# torch.set_default_device("cuda")
|
12 |
# Load the Phi 2 model and tokenizer
|
@@ -19,9 +13,7 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
19 |
"microsoft/phi-2",
|
20 |
device_map="cpu",
|
21 |
trust_remote_code=True,
|
22 |
-
# offload_folder="offload",
|
23 |
torch_dtype=torch.float32
|
24 |
-
# torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
25 |
)
|
26 |
|
27 |
# Streamlit UI
|
@@ -34,14 +26,14 @@ prompt = st.text_area("Enter your prompt:", """Write a short summary about how t
|
|
34 |
if st.button("Generate Output"):
|
35 |
with torch.no_grad():
|
36 |
token_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt",
|
37 |
-
|
38 |
)
|
39 |
output_ids = model.generate(
|
40 |
token_ids.to(model.device),
|
41 |
-
max_new_tokens=512,
|
42 |
do_sample=True,
|
43 |
temperature=0.3,
|
44 |
-
|
45 |
)
|
46 |
|
47 |
output = tokenizer.decode(output_ids[0][token_ids.size(1):])
|
|
|
1 |
import streamlit as st
|
|
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
# torch.set_default_device("cuda")
|
6 |
# Load the Phi 2 model and tokenizer
|
|
|
13 |
"microsoft/phi-2",
|
14 |
device_map="cpu",
|
15 |
trust_remote_code=True,
|
|
|
16 |
torch_dtype=torch.float32
|
|
|
17 |
)
|
18 |
|
19 |
# Streamlit UI
|
|
|
26 |
if st.button("Generate Output"):
|
27 |
with torch.no_grad():
|
28 |
token_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt",
|
29 |
+
return_attention_mask=False
|
30 |
)
|
31 |
output_ids = model.generate(
|
32 |
token_ids.to(model.device),
|
33 |
+
# max_new_tokens=512,
|
34 |
do_sample=True,
|
35 |
temperature=0.3,
|
36 |
+
max_length=200
|
37 |
)
|
38 |
|
39 |
output = tokenizer.decode(output_ids[0][token_ids.size(1):])
|