Spaces:
Sleeping
Sleeping
Merge branch 'main' of hf.co:spaces/valcore/Branchy-phi-2
Browse files- README.md +2 -2
- app.py +3 -1
- requirements.txt +2 -2
README.md
CHANGED
@@ -4,7 +4,7 @@ emoji: ⚡
|
|
4 |
colorFrom: purple
|
5 |
colorTo: pink
|
6 |
sdk: gradio
|
7 |
-
sdk_version: 4.
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
---
|
@@ -45,4 +45,4 @@ The command will start a local web server and open the application in your defau
|
|
45 |
|
46 |
- **app.py**: The main Streamlit application script.
|
47 |
- **requirements.txt**: Lists all the Python dependencies required by the project.
|
48 |
-
- **src/**: Contains the source code for the project.
|
|
|
4 |
colorFrom: purple
|
5 |
colorTo: pink
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 4.37.2
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
---
|
|
|
45 |
|
46 |
- **app.py**: The main Streamlit application script.
|
47 |
- **requirements.txt**: Lists all the Python dependencies required by the project.
|
48 |
+
- **src/**: Contains the source code for the project.
|
app.py
CHANGED
@@ -2,6 +2,7 @@ import gradio as gr
|
|
2 |
import torch
|
3 |
import pandas as pd
|
4 |
import plotly.graph_objects as go
|
|
|
5 |
from plotly.subplots import make_subplots
|
6 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
7 |
import time
|
@@ -73,7 +74,8 @@ def truncate_context(input_ids, max_length=2048):
|
|
73 |
if len(input_ids[0]) > max_length:
|
74 |
return input_ids[:, -max_length:]
|
75 |
return input_ids
|
76 |
-
|
|
|
77 |
def generate_response(message, chat_history, epsilon):
|
78 |
global data, stop_generation
|
79 |
data = pd.DataFrame(columns=["Time taken (in ms)", "Early exit depth", "Token"])
|
|
|
2 |
import torch
|
3 |
import pandas as pd
|
4 |
import plotly.graph_objects as go
|
5 |
+
import spaces
|
6 |
from plotly.subplots import make_subplots
|
7 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
8 |
import time
|
|
|
74 |
if len(input_ids[0]) > max_length:
|
75 |
return input_ids[:, -max_length:]
|
76 |
return input_ids
|
77 |
+
|
78 |
+
@spaces.GPU
|
79 |
def generate_response(message, chat_history, epsilon):
|
80 |
global data, stop_generation
|
81 |
data = pd.DataFrame(columns=["Time taken (in ms)", "Early exit depth", "Token"])
|
requirements.txt
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
-
gradio==4.
|
2 |
-
torch==2.
|
3 |
pandas==2.0.3
|
4 |
transformers==4.41.1
|
5 |
plotly==5.22.0
|
|
|
1 |
+
gradio==4.37.2
|
2 |
+
torch==2.2.0
|
3 |
pandas==2.0.3
|
4 |
transformers==4.41.1
|
5 |
plotly==5.22.0
|