Spaces:
Running
Running
BenchmarkBot
commited on
Commit
β’
a18f8de
1
Parent(s):
bee5389
added single and multi gpu inference
Browse files
app.py
CHANGED
@@ -26,16 +26,13 @@ SORTING_COLUMN = ["Throughput (tokens/s) β¬οΈ"]
|
|
26 |
llm_perf_dataset_repo = load_dataset_repo(LLM_PERF_DATASET_REPO, OPTIMUM_TOKEN)
|
27 |
|
28 |
|
29 |
-
def get_benchmark_df():
|
30 |
if llm_perf_dataset_repo:
|
31 |
llm_perf_dataset_repo.git_pull()
|
32 |
|
33 |
# load
|
34 |
df = pd.read_csv(
|
35 |
-
"./llm-perf-dataset/reports/
|
36 |
-
|
37 |
-
# remove quantized models
|
38 |
-
df = df[df["backend.quantization"].isnull()]
|
39 |
|
40 |
# preprocess
|
41 |
df["model"] = df["model"].apply(make_clickable_model)
|
@@ -61,19 +58,34 @@ with demo:
|
|
61 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
62 |
|
63 |
with gr.Tabs(elem_classes="tab-buttons") as tabs:
|
64 |
-
with gr.TabItem("
|
65 |
-
|
66 |
-
|
67 |
-
-
|
68 |
-
-
|
69 |
-
|
70 |
-
gr.
|
71 |
-
|
|
|
72 |
leaderboard_table_lite = gr.components.Dataframe(
|
73 |
-
value=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
datatype=COLUMNS_DATATYPES,
|
75 |
headers=NEW_COLUMNS,
|
76 |
-
elem_id="
|
77 |
)
|
78 |
|
79 |
with gr.Row():
|
|
|
26 |
llm_perf_dataset_repo = load_dataset_repo(LLM_PERF_DATASET_REPO, OPTIMUM_TOKEN)
|
27 |
|
28 |
|
29 |
+
def get_benchmark_df(benchmark):
|
30 |
if llm_perf_dataset_repo:
|
31 |
llm_perf_dataset_repo.git_pull()
|
32 |
|
33 |
# load
|
34 |
df = pd.read_csv(
|
35 |
+
f"./llm-perf-dataset/reports/{benchmark}/inference_report.csv")
|
|
|
|
|
|
|
36 |
|
37 |
# preprocess
|
38 |
df["model"] = df["model"].apply(make_clickable_model)
|
|
|
58 |
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
|
59 |
|
60 |
with gr.Tabs(elem_classes="tab-buttons") as tabs:
|
61 |
+
with gr.TabItem("π₯οΈ A100-80GB Benchmark ποΈ", elem_id="A100-benchmark", id=0):
|
62 |
+
|
63 |
+
SINGLE_A100_TEXT = """<h4>Specifications:</h4>
|
64 |
+
- Single-GPU (1)
|
65 |
+
- Singleton Batch (1)
|
66 |
+
- Thousand Tokens (1000)"""
|
67 |
+
gr.HTML(SINGLE_A100_TEXT)
|
68 |
+
|
69 |
+
single_A100_df = get_benchmark_df(benchmark="1xA100-80GB")
|
70 |
leaderboard_table_lite = gr.components.Dataframe(
|
71 |
+
value=single_A100_df,
|
72 |
+
datatype=COLUMNS_DATATYPES,
|
73 |
+
headers=NEW_COLUMNS,
|
74 |
+
elem_id="1xA100-table",
|
75 |
+
)
|
76 |
+
|
77 |
+
MULTI_A100_TEXT = """<h4>Specifications:</h4>
|
78 |
+
- Multi-GPU (4)
|
79 |
+
- Singleton Batch (1)
|
80 |
+
- Thousand Tokens (1000)"""
|
81 |
+
gr.HTML(MULTI_A100_TEXT)
|
82 |
+
|
83 |
+
multi_A100_df = get_benchmark_df(benchmark="4xA100-80GB")
|
84 |
+
leaderboard_table_full = gr.components.Dataframe(
|
85 |
+
value=multi_A100_df,
|
86 |
datatype=COLUMNS_DATATYPES,
|
87 |
headers=NEW_COLUMNS,
|
88 |
+
elem_id="4xA100-table",
|
89 |
)
|
90 |
|
91 |
with gr.Row():
|