added single and multi gpu inference

app.py

@@ -26,16 +26,13 @@ SORTING_COLUMN = ["Throughput (tokens/s) ⬆️"]
 llm_perf_dataset_repo = load_dataset_repo(LLM_PERF_DATASET_REPO, OPTIMUM_TOKEN)
 
 
-def get_benchmark_df():
+def get_benchmark_df(benchmark):
     if llm_perf_dataset_repo:
         llm_perf_dataset_repo.git_pull()
 
     # load
     df = pd.read_csv(
-        "./llm-perf-dataset/reports/cuda_1_100/inference_report.csv")
-
-    # remove quantized models
-    df = df[df["backend.quantization"].isnull()]
+        f"./llm-perf-dataset/reports/{benchmark}/inference_report.csv")
 
     # preprocess
     df["model"] = df["model"].apply(make_clickable_model)
@@ -61,19 +58,34 @@ with demo:
     gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        with gr.TabItem("📊 A100-80GB Benchmark 🏋️", elem_id="a100-benchmark", id=0):
-            dataframe_text = """<h4>Specifications:</h4>
-            - Single and Multi-GPU Setup
-            - Batch Size: 1
-            - Generated Tokens: 100"""
-
-            gr.Markdown(dataframe_text, elem_classes="markdown-text")
-            benchmark_df = get_benchmark_df()
+        with gr.TabItem("🖥️ A100-80GB Benchmark 🏋️", elem_id="A100-benchmark", id=0):
+
+            SINGLE_A100_TEXT = """<h4>Specifications:</h4>
+            - Single-GPU (1)
+            - Singleton Batch (1)
+            - Thousand Tokens (1000)"""
+            gr.HTML(SINGLE_A100_TEXT)
+
+            single_A100_df = get_benchmark_df(benchmark="1xA100-80GB")
             leaderboard_table_lite = gr.components.Dataframe(
-                value=benchmark_df,
+                value=single_A100_df,
+                datatype=COLUMNS_DATATYPES,
+                headers=NEW_COLUMNS,
+                elem_id="1xA100-table",
+            )
+
+            MULTI_A100_TEXT = """<h4>Specifications:</h4>
+            - Multi-GPU (4)
+            - Singleton Batch (1)
+            - Thousand Tokens (1000)"""
+            gr.HTML(MULTI_A100_TEXT)
+
+            multi_A100_df = get_benchmark_df(benchmark="4xA100-80GB")
+            leaderboard_table_full = gr.components.Dataframe(
+                value=multi_A100_df,
                 datatype=COLUMNS_DATATYPES,
                 headers=NEW_COLUMNS,
-                elem_id="pytorch-A100-benchmark",
+                elem_id="4xA100-table",
             )
 
     with gr.Row():