Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -4,48 +4,55 @@ import torch
|
|
4 |
from concurrent.futures import ThreadPoolExecutor
|
5 |
from threading import Lock
|
6 |
|
7 |
-
# Global cache
|
8 |
CACHE_SIZE = 100
|
9 |
prediction_cache = {}
|
10 |
cache_lock = Lock()
|
11 |
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
17 |
-
device = 0
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
|
|
|
|
|
|
24 |
|
25 |
-
# Load both models concurrently at startup
|
26 |
with ThreadPoolExecutor() as executor:
|
27 |
-
sentiment_future = executor.submit(
|
28 |
-
emotion_future = executor.submit(
|
29 |
|
30 |
sentiment_pipeline = sentiment_future.result()
|
31 |
emotion_pipeline = emotion_future.result()
|
32 |
|
33 |
def analyze_text(text):
|
34 |
-
# Check cache first (
|
35 |
with cache_lock:
|
36 |
if text in prediction_cache:
|
37 |
return prediction_cache[text]
|
38 |
|
39 |
try:
|
40 |
-
#
|
41 |
with ThreadPoolExecutor() as executor:
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
emotion_result = emotion_future.result()[0]
|
47 |
|
48 |
-
#
|
49 |
result = {
|
50 |
"Sentiment": {sentiment_result['label']: round(sentiment_result['score'], 4)},
|
51 |
"Emotion": {emotion_result['label']: round(emotion_result['score'], 4)}
|
@@ -53,7 +60,7 @@ def analyze_text(text):
|
|
53 |
except Exception as e:
|
54 |
result = {"error": str(e)}
|
55 |
|
56 |
-
# Update cache with
|
57 |
with cache_lock:
|
58 |
if len(prediction_cache) >= CACHE_SIZE:
|
59 |
prediction_cache.pop(next(iter(prediction_cache)))
|
@@ -61,15 +68,13 @@ def analyze_text(text):
|
|
61 |
|
62 |
return result
|
63 |
|
64 |
-
# Gradio interface
|
65 |
-
|
66 |
-
|
67 |
-
demo = gr.Interface(
|
68 |
fn=analyze_text,
|
69 |
inputs=gr.Textbox(placeholder="Enter your text here...", label="Input Text"),
|
70 |
outputs=gr.JSON(label="Analysis Results"),
|
71 |
title="🚀 Fast Sentiment & Emotion Analysis",
|
72 |
-
description="An optimized application using
|
73 |
examples=[
|
74 |
["I'm thrilled to start this new adventure!"],
|
75 |
["This situation is making me really frustrated."],
|
@@ -79,8 +84,9 @@ demo = gr.Interface(
|
|
79 |
allow_flagging="never"
|
80 |
)
|
81 |
|
82 |
-
# Warm up the models
|
83 |
_ = analyze_text("Warming up models...")
|
84 |
|
85 |
if __name__ == "__main__":
|
86 |
-
|
|
|
|
4 |
from concurrent.futures import ThreadPoolExecutor
|
5 |
from threading import Lock
|
6 |
|
7 |
+
# Global cache and thread lock for thread-safe caching
|
8 |
CACHE_SIZE = 100
|
9 |
prediction_cache = {}
|
10 |
cache_lock = Lock()
|
11 |
|
12 |
+
def load_model(model_name):
|
13 |
+
"""
|
14 |
+
Loads the model with 8-bit quantization if a GPU is available.
|
15 |
+
On CPU, it loads the full model.
|
16 |
+
"""
|
17 |
+
if torch.cuda.is_available():
|
18 |
+
# Use 8-bit quantization and auto device mapping for GPU inference.
|
19 |
+
model = AutoModelForSequenceClassification.from_pretrained(
|
20 |
+
model_name, load_in_8bit=True, device_map="auto"
|
21 |
+
)
|
22 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
23 |
+
device = 0 # GPU index
|
24 |
+
else:
|
25 |
+
# CPU fallback: do not use quantization.
|
26 |
+
model = AutoModelForSequenceClassification.from_pretrained(model_name)
|
27 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
28 |
+
device = -1
|
29 |
+
|
30 |
+
return pipeline("text-classification", model=model, tokenizer=tokenizer, device=device)
|
31 |
+
|
32 |
+
# Load both models concurrently atartup.
|
33 |
|
|
|
34 |
with ThreadPoolExecutor() as executor:
|
35 |
+
sentiment_future = executor.submit(load_model, "cardiffnlp/twitter-roberta-base-sentiment")
|
36 |
+
emotion_future = executor.submit(load_model, "bhadresh-savani/bert-base-uncased-emotion")
|
37 |
|
38 |
sentiment_pipeline = sentiment_future.result()
|
39 |
emotion_pipeline = emotion_future.result()
|
40 |
|
41 |
def analyze_text(text):
|
42 |
+
# Check cache first (thread-safe)
|
43 |
with cache_lock:
|
44 |
if text in prediction_cache:
|
45 |
return prediction_cache[text]
|
46 |
|
47 |
try:
|
48 |
+
# Run both model inferences in parallel.
|
49 |
with ThreadPoolExecutor() as executor:
|
50 |
+
future_sentiment = executor.submit(sentiment_pipeline, text)
|
51 |
+
future_emotion = executor.submit(emotion_pipeline, text)
|
52 |
+
sentiment_result = future_sentiment.result()[0]
|
53 |
+
emotion_result = future_emotion.result()[0]
|
|
|
54 |
|
55 |
+
# Format the output with rounded scores.
|
56 |
result = {
|
57 |
"Sentiment": {sentiment_result['label']: round(sentiment_result['score'], 4)},
|
58 |
"Emotion": {emotion_result['label']: round(emotion_result['score'], 4)}
|
|
|
60 |
except Exception as e:
|
61 |
result = {"error": str(e)}
|
62 |
|
63 |
+
# Update cache with protection.
|
64 |
with cache_lock:
|
65 |
if len(prediction_cache) >= CACHE_SIZE:
|
66 |
prediction_cache.pop(next(iter(prediction_cache)))
|
|
|
68 |
|
69 |
return result
|
70 |
|
71 |
+
# Define the Gradio interface.
|
72 |
+
demo = gr.Interface(
|
|
|
|
|
73 |
fn=analyze_text,
|
74 |
inputs=gr.Textbox(placeholder="Enter your text here...", label="Input Text"),
|
75 |
outputs=gr.JSON(label="Analysis Results"),
|
76 |
title="🚀 Fast Sentiment & Emotion Analysis",
|
77 |
+
description="An optimized application using quantized models (when available) and parallel processing for fast inference.",
|
78 |
examples=[
|
79 |
["I'm thrilled to start this new adventure!"],
|
80 |
["This situation is making me really frustrated."],
|
|
|
84 |
allow_flagging="never"
|
85 |
)
|
86 |
|
87 |
+
# Warm up the models to reduce first-call latency.
|
88 |
_ = analyze_text("Warming up models...")
|
89 |
|
90 |
if __name__ == "__main__":
|
91 |
+
# In Spaces, binding to 0.0.0.0 is required.
|
92 |
+
demo.launch(server_name="0.0.0.0", server_port=7860)
|