added H4 score

app.py

@@ -17,10 +17,10 @@ COLUMNS_MAPPING = {
     "model": "Model 🤗",
     "backend.name": "Backend 🏭",
     "backend.torch_dtype": "Datatype 📥",
-    "generate.latency(s)": "Latency (s) ⬇️",
+    "average": "Average H4 Score ⬆️",
     "generate.throughput(tokens/s)": "Throughput (tokens/s) ⬆️",
 }
-COLUMNS_DATATYPES = ["markdown", "str", "str", "number", "number"]
+COLUMNS_DATATYPES = ["markdown", "str", "str", "number", "number", "number"]
 SORTING_COLUMN = ["Throughput (tokens/s) ⬆️"]
 
 
@@ -32,18 +32,24 @@ def get_benchmark_df(benchmark):
         llm_perf_dataset_repo.git_pull()
 
     # load
-    df = pd.read_csv(
+    bench_df = pd.read_csv(
         f"./llm-perf-dataset/reports/{benchmark}/inference_report.csv")
+    scores_df = pd.read_csv(
+        f"./llm-perf-dataset/reports/average_scores.csv")
+    # merge on model
+    bench_df = bench_df.merge(
+        scores_df, how="left", left_on="model", right_on="model")
+
     # preprocess
-    df["model"] = df["model"].apply(make_clickable_model)
+    bench_df["model"] = bench_df["model"].apply(make_clickable_model)
     # filter
-    df = df[list(COLUMNS_MAPPING.keys())]
+    bench_df = bench_df[list(COLUMNS_MAPPING.keys())]
     # rename
-    df.rename(columns=COLUMNS_MAPPING, inplace=True)
+    bench_df.rename(columns=COLUMNS_MAPPING, inplace=True)
     # sort
-    df.sort_values(by=SORTING_COLUMN, ascending=False, inplace=True)
+    bench_df.sort_values(by=SORTING_COLUMN, ascending=False, inplace=True)
 
-    return df
+    return bench_df
 
 
 def change_tab(query_param):