now with label title
Browse files
app.py
CHANGED
@@ -13,19 +13,19 @@ import pandas as pd
|
|
13 |
BENCHMARK_DATA = {
|
14 |
"Greedy Search": {
|
15 |
"DistilGPT2": {
|
16 |
-
"T4": [],
|
17 |
-
"3090": [],
|
18 |
-
"A100": [],
|
19 |
},
|
20 |
"GPT2": {
|
21 |
-
"T4": [],
|
22 |
-
"3090": [],
|
23 |
-
"A100": [],
|
24 |
},
|
25 |
"OPT-1.3B": {
|
26 |
-
"T4": [],
|
27 |
-
"3090": [],
|
28 |
-
"A100": [],
|
29 |
},
|
30 |
"GPTJ-6B": {
|
31 |
"T4": [],
|
@@ -152,7 +152,8 @@ def get_plot(model_name, generate_type):
|
|
152 |
)
|
153 |
g.despine(left=True)
|
154 |
g.set_axis_labels("GPU", "Generation time (ms)")
|
155 |
-
g.
|
|
|
156 |
return plt.gcf()
|
157 |
|
158 |
demo = gr.Blocks()
|
@@ -182,7 +183,7 @@ with demo:
|
|
182 |
["Pytorch Version", "1.11.0"],
|
183 |
["OS", "22.04 LTS (3090) / Debian 10 (other GPUs)"],
|
184 |
["CUDA", "11.6 (3090) / 11.3 (others GPUs)"],
|
185 |
-
["Number of
|
186 |
["Is there code to reproduce?", "Yes -- https://gist.github.com/gante/f0017e3f13ac11b0c02e4e4db351f52f"],
|
187 |
],
|
188 |
)
|
|
|
13 |
BENCHMARK_DATA = {
|
14 |
"Greedy Search": {
|
15 |
"DistilGPT2": {
|
16 |
+
"T4": [336.22, 3976.23, 115.84],
|
17 |
+
"3090": [158.38, 1835.82, 46.56],
|
18 |
+
"A100": [371.49, 4073.84, 60.94],
|
19 |
},
|
20 |
"GPT2": {
|
21 |
+
"T4": [607.31, 7140.23, 185.12],
|
22 |
+
"3090": [297.03, 3308.31, 76.68],
|
23 |
+
"A100": [691.75, 7323.60, 110.72],
|
24 |
},
|
25 |
"OPT-1.3B": {
|
26 |
+
"T4": [1303.41, 15939.07, 1488.15],
|
27 |
+
"3090": [428.33, 7259.43, 468.37],
|
28 |
+
"A100": [1125.00, 16713.63, 384.52],
|
29 |
},
|
30 |
"GPTJ-6B": {
|
31 |
"T4": [],
|
|
|
152 |
)
|
153 |
g.despine(left=True)
|
154 |
g.set_axis_labels("GPU", "Generation time (ms)")
|
155 |
+
g.set(title=f"{generate_type} generation time for {model_name} (no bar means OOM)")
|
156 |
+
g.legend.set_title("Framework")
|
157 |
return plt.gcf()
|
158 |
|
159 |
demo = gr.Blocks()
|
|
|
183 |
["Pytorch Version", "1.11.0"],
|
184 |
["OS", "22.04 LTS (3090) / Debian 10 (other GPUs)"],
|
185 |
["CUDA", "11.6 (3090) / 11.3 (others GPUs)"],
|
186 |
+
["Number of Runs", "100 (the first run was discarded to ignore compilation time)"],
|
187 |
["Is there code to reproduce?", "Yes -- https://gist.github.com/gante/f0017e3f13ac11b0c02e4e4db351f52f"],
|
188 |
],
|
189 |
)
|