filtered out some columns

app.py

@@ -1,28 +1,25 @@
 import os
 import gradio as gr
 import pandas as pd
-from huggingface_hub import HfApi
-from huggingface_hub import Repository
+from huggingface_hub import HfApi, Repository
 from apscheduler.schedulers.background import BackgroundScheduler
 
-from src.assets.text_content import *
-from src.assets.css_html_js import custom_css
+from src.assets.text_content import TITLE, INTRODUCTION_TEXT
+from src.assets.css_html_js import custom_css, get_window_url_params
 
 OPTIMUM_TOKEN = os.environ.get("OPTIMUM_TOKEN", None)
 
 LLM_PERF_LEADERBOARD_REPO = "optimum/llm-perf-leaderboard"
 LLM_PERF_DATASET_REPO = "optimum/llm-perf"
 
-api = HfApi()
-
 
 def restart_space():
-    api.restart_space(
+    HfApi().restart_space(
         repo_id=LLM_PERF_LEADERBOARD_REPO, token=OPTIMUM_TOKEN
     )
 
 
-def load_all_info_from_hub():
+def load_dataset_repo():
     llm_perf_repo = None
     if OPTIMUM_TOKEN:
         print("Loading LLM-Perf-Dataset from Hub...")
@@ -37,29 +34,26 @@ def load_all_info_from_hub():
     return llm_perf_repo
 
 
-llm_perf_repo = load_all_info_from_hub()
-
-
-def has_no_nan_values(df, columns):
-    return df[columns].notna().all(axis=1)
-
-
-def has_nan_values(df, columns):
-    return df[columns].isna().any(axis=1)
-
-
 def get_leaderboard_df():
     if llm_perf_repo:
         llm_perf_repo.git_pull()
 
     df = pd.read_csv("./llm-perf/reports/cuda_1_100/inference_report.csv")
-    print(df.columns)
+    df = df[["model", "backend.name", "backend.torch_dtype", "backend.quantization",
+             "generate.latency(s)", "generate.throughput(tokens/s)"]]
 
-    return df
+    df.rename(columns={
+        "model": "Model",
+        "backend.name": "Backend",
+        "backend.torch_dtype": "Torch dtype",
+        "backend.quantization": "Quantization",
+        "generate.latency(s)": "Latency (s)",
+        "generate.throughput(tokens/s)": "Throughput (tokens/s)"
+    }, inplace=True)
 
+    df.sort_values(by=["Throughput (tokens/s)"], ascending=False, inplace=True)
 
-original_df = get_leaderboard_df()
-leaderboard_df = original_df.copy()
+    return df
 
 
 def refresh():
@@ -68,14 +62,16 @@ def refresh():
     return leaderboard_df
 
 
+llm_perf_repo = load_dataset_repo()
+
 demo = gr.Blocks(css=custom_css)
 with demo:
     gr.HTML(TITLE)
     gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
-    print("rendering tab...")
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        with gr.TabItem("🏅 LLM-Perf Benchmark", elem_id="llm-perf-benchmark-tab-table", id=0):
+        with gr.TabItem("Vanilla Benchmark", elem_id="vanilla-benchmark", id=0):
+            leaderboard_df = get_leaderboard_df()
             leaderboard_table_lite = gr.components.Dataframe(
                 value=leaderboard_df,
                 headers=leaderboard_df.columns.tolist(),

src/assets/text_content.py

@@ -1,8 +1,5 @@
 TITLE = """<h1 align="center" id="space-title">🤗 Open LLM-Perf Leaderboard</h1>"""
 
 INTRODUCTION_TEXT = f"""
-The 🤗 Open LLM-Perf Leaderboard aims to benchmark the performance of large language models (LLMs) on different backends and hardwares using [Optimum-Benchmark](https://github.com/huggingface/optimum-benchmark)
-"""
-
-LLM_BENCHMARKS_TEXT = f"""
+The 🤗 Open LLM-Perf Leaderboard aims to benchmark the performance (latency & throughput) of Large Language Models (LLMs) on different backends and hardwares using [Optimum-Benchmark](https://github.com/huggingface/optimum-benchmark)
 """