Spaces:
Runtime error
Runtime error
Commit
·
d889137
1
Parent(s):
98f80c6
updated base model
Browse files- __pycache__/quantization.cpython-311.pyc +0 -0
- app.py +63 -38
- flagged/log.csv +2 -0
- flagged/throughput Comparison/49a4c8006ae895a1b75f/image.webp +0 -0
- mest.tar +0 -0
__pycache__/quantization.cpython-311.pyc
ADDED
Binary file (5.25 kB). View file
|
|
app.py
CHANGED
@@ -1,80 +1,105 @@
|
|
1 |
import gradio as gr
|
2 |
-
from transformers import pipeline
|
3 |
-
import numpy as np
|
4 |
-
from time import perf_counter
|
5 |
from setfit import SetFitModel
|
6 |
from optimum.onnxruntime import ORTModelForFeatureExtraction
|
7 |
from quantization import OnnxSetFitModel
|
8 |
-
|
9 |
-
|
|
|
|
|
|
|
|
|
10 |
|
11 |
-
|
|
|
12 |
ort_model = ORTModelForFeatureExtraction.from_pretrained("hsmashiana/optimized_model_hpml", file_name="model_quantized.onnx")
|
13 |
tokenizer = AutoTokenizer.from_pretrained("hsmashiana/optimized_model_hpml")
|
14 |
model3 = OnnxSetFitModel(ort_model, tokenizer, model1.model_head)
|
15 |
|
16 |
-
decode = {0:"World",1:"Sports",2:"Business",3:"Sci/Tech"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
def compare_models(text):
|
19 |
-
|
20 |
-
# result2 = model2(text)
|
21 |
-
# # Including model names in the results
|
22 |
-
# output1 = {"Model": "BERT Base Uncased", "Output": result1}
|
23 |
-
# output2 = {"Model": "RoBERTa Base", "Output": result2}
|
24 |
-
# return output1, output2
|
25 |
-
|
26 |
times = []
|
27 |
-
|
|
|
28 |
for _ in range(5):
|
29 |
model1([text])
|
|
|
30 |
# Measure the execution time of model predictions
|
31 |
for _ in range(20):
|
32 |
start = perf_counter()
|
33 |
out1 = model1([text])
|
34 |
end = perf_counter()
|
35 |
times.append(end - start)
|
36 |
-
# Calculate mean and standard deviation of latency
|
37 |
-
avg_latency_ms_model_1 = np.mean(times) * 1000
|
38 |
|
39 |
-
|
40 |
-
# # Warm-up phase to ensure fair timing
|
41 |
-
# for _ in range(5):
|
42 |
-
# model2([text])
|
43 |
-
# # Measure the execution time of model predictions
|
44 |
-
# for _ in range(20):
|
45 |
-
# start = perf_counter()
|
46 |
-
# out2 = model2([text])
|
47 |
-
# end = perf_counter()
|
48 |
-
# times.append(end - start)
|
49 |
-
# # Calculate mean and standard deviation of latency
|
50 |
-
# avg_latency_ms_model_2 = np.mean(times) * 1000
|
51 |
|
52 |
times = []
|
53 |
-
|
|
|
54 |
for _ in range(5):
|
55 |
model3.predict([text])
|
|
|
56 |
# Measure the execution time of model predictions
|
57 |
for _ in range(20):
|
58 |
start = perf_counter()
|
59 |
-
out3 = model3([text])
|
60 |
end = perf_counter()
|
61 |
times.append(end - start)
|
62 |
-
|
63 |
avg_latency_ms_model_3 = np.mean(times) * 1000
|
64 |
|
65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
-
# Create a Gradio interface
|
68 |
iface = gr.Interface(
|
69 |
fn=compare_models,
|
70 |
inputs="text",
|
71 |
outputs=[
|
72 |
-
gr.components.JSON(label="Base
|
73 |
-
gr.components.JSON(label="
|
|
|
74 |
],
|
75 |
title="Compare Sentence Classification Models",
|
76 |
-
description="Enter a sentence to see how each model classifies it."
|
|
|
77 |
)
|
78 |
|
79 |
-
# Run the interface
|
80 |
iface.launch()
|
|
|
1 |
import gradio as gr
|
2 |
+
from transformers import pipeline, AutoTokenizer
|
|
|
|
|
3 |
from setfit import SetFitModel
|
4 |
from optimum.onnxruntime import ORTModelForFeatureExtraction
|
5 |
from quantization import OnnxSetFitModel
|
6 |
+
import numpy as np
|
7 |
+
from time import perf_counter
|
8 |
+
import matplotlib.pyplot as plt
|
9 |
+
from io import BytesIO
|
10 |
+
from PIL import Image
|
11 |
+
import io
|
12 |
|
13 |
+
# Load the models
|
14 |
+
model1 = SetFitModel.from_pretrained("hsmashiana/basemodel_hpml")
|
15 |
ort_model = ORTModelForFeatureExtraction.from_pretrained("hsmashiana/optimized_model_hpml", file_name="model_quantized.onnx")
|
16 |
tokenizer = AutoTokenizer.from_pretrained("hsmashiana/optimized_model_hpml")
|
17 |
model3 = OnnxSetFitModel(ort_model, tokenizer, model1.model_head)
|
18 |
|
19 |
+
decode = {0: "World", 1: "Sports", 2: "Business", 3: "Sci/Tech"}
|
20 |
+
|
21 |
+
def plot_throughput_bar_chart(throughput_model1, throughput_model2):
|
22 |
+
labels = ['Base model', 'Optimized model']
|
23 |
+
throughputs = [throughput_model1, throughput_model2]
|
24 |
+
|
25 |
+
plt.figure(figsize=(8, 6))
|
26 |
+
plt.bar(labels, throughputs, color=['blue', 'navy'])
|
27 |
+
plt.xlabel('Models')
|
28 |
+
plt.ylabel('Throughput (tokens/second)')
|
29 |
+
plt.title('Model Throughput Comparison')
|
30 |
+
plt.tight_layout()
|
31 |
+
|
32 |
+
# Create a PIL Image from the plot
|
33 |
+
buf = io.BytesIO()
|
34 |
+
plt.savefig(buf, format='png')
|
35 |
+
buf.seek(0)
|
36 |
+
img = Image.open(buf)
|
37 |
+
plt.close()
|
38 |
+
return img
|
39 |
|
40 |
def compare_models(text):
|
41 |
+
inputs = tokenizer(text, return_tensors="pt")
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
times = []
|
43 |
+
|
44 |
+
# Warm-up phase to ensure fair timing
|
45 |
for _ in range(5):
|
46 |
model1([text])
|
47 |
+
|
48 |
# Measure the execution time of model predictions
|
49 |
for _ in range(20):
|
50 |
start = perf_counter()
|
51 |
out1 = model1([text])
|
52 |
end = perf_counter()
|
53 |
times.append(end - start)
|
|
|
|
|
54 |
|
55 |
+
avg_latency_ms_model_1 = np.mean(times) * 1000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
|
57 |
times = []
|
58 |
+
|
59 |
+
# Warm-up phase to ensure fair timing
|
60 |
for _ in range(5):
|
61 |
model3.predict([text])
|
62 |
+
|
63 |
# Measure the execution time of model predictions
|
64 |
for _ in range(20):
|
65 |
start = perf_counter()
|
66 |
+
out3 = model3.predict([text])
|
67 |
end = perf_counter()
|
68 |
times.append(end - start)
|
69 |
+
|
70 |
avg_latency_ms_model_3 = np.mean(times) * 1000
|
71 |
|
72 |
+
throughput_tokens_per_sec1 = inputs['input_ids'].size(1) / (avg_latency_ms_model_1 / 1000)
|
73 |
+
throughput_tokens_per_sec2 = inputs['input_ids'].size(1) / (avg_latency_ms_model_3 / 1000)
|
74 |
+
|
75 |
+
plot_data = plot_throughput_bar_chart(throughput_tokens_per_sec1, throughput_tokens_per_sec2)
|
76 |
+
|
77 |
+
result1 = {
|
78 |
+
"Base Model": {
|
79 |
+
"answer": decode[out1.numpy()[0]],
|
80 |
+
"average time (ms)": avg_latency_ms_model_1,
|
81 |
+
"throughput (tokens/sec)": throughput_tokens_per_sec1
|
82 |
+
}}
|
83 |
+
result2 = {
|
84 |
+
"Optimized Model": {
|
85 |
+
"answer": decode[out3[0]],
|
86 |
+
"average time (ms)": avg_latency_ms_model_3,
|
87 |
+
"throughput (tokens/sec)": throughput_tokens_per_sec2
|
88 |
+
}}
|
89 |
+
|
90 |
+
return result1, result2, plot_data
|
91 |
|
|
|
92 |
iface = gr.Interface(
|
93 |
fn=compare_models,
|
94 |
inputs="text",
|
95 |
outputs=[
|
96 |
+
gr.components.JSON(label="Base Model"),
|
97 |
+
gr.components.JSON(label="Optimized Model"),
|
98 |
+
gr.components.Image(label="throughput Comparison")
|
99 |
],
|
100 |
title="Compare Sentence Classification Models",
|
101 |
+
description="Enter a sentence to see how each model classifies it and their throughputs.",
|
102 |
+
allow_flagging="never"
|
103 |
)
|
104 |
|
|
|
105 |
iface.launch()
|
flagged/log.csv
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
text,Base Model,Optimized Model,throughput Comparison,flag,username,timestamp
|
2 |
+
hellool,"{""Base Model"": {""answer"": ""Business"", ""average time (ms)"": 9.624537501076702, ""throughput (tokens/sec)"": 415.60438613829683}}","{""Optimized Model"": {""answer"": ""Business"", ""average time (ms)"": 1.6875000983418431, ""throughput (tokens/sec)"": 2370.370232233139}}",flagged/throughput Comparison/49a4c8006ae895a1b75f/image.webp,,,2024-05-04 20:48:16.264716
|
flagged/throughput Comparison/49a4c8006ae895a1b75f/image.webp
ADDED
![]() |
mest.tar
ADDED
Binary file (1.02 kB). View file
|
|