changed optimizations layout and added them to plot

app.py

@@ -29,14 +29,13 @@ COLUMNS_MAPPING = {
     "model": "Model 🤗",
     "backend.name": "Backend 🏭",
     "backend.torch_dtype": "Load Dtype 📥",
-    "num_parameters": "#️⃣ Parameters 📏",
+    "optimizations": "Optimizations 🛠️",
     #
     "forward.peak_memory(MB)": "Peak Memory (MB) ⬇️",
     "generate.throughput(tokens/s)": "Throughput (tokens/s) ⬆️",
     "average": "Average Open LLM Score ⬆️",
     #
-    "backend.bettertransformer": "BetterTransformer 🤖",
-    "backend.load_in_8bit": "LLM.int8 🗜️",
+    "num_parameters": "#️⃣ Parameters 📏",
 }
 COLUMNS_DATATYPES = [
     "markdown",
@@ -49,7 +48,6 @@ COLUMNS_DATATYPES = [
     "markdown",
     #
     "str",
-    "str",
 ]
 SORTING_COLUMN = ["Throughput (tokens/s) ⬆️"]
 
@@ -66,6 +64,13 @@ def get_benchmark_df(benchmark="1xA100-80GB"):
     scores_df = pd.read_csv(f"./llm-perf-dataset/reports/additional_data.csv")
     bench_df = bench_df.merge(scores_df, on="model", how="left")
 
+    bench_df["optimizations"] = bench_df[
+        ["backend.bettertransformer", "backend.load_in_8bit"]
+    ].apply(
+        lambda x: "BetterTransformer 🤖" if x[0] else "LLM.int8 🗜️" if x[1] else "None",
+        axis=1,
+    )
+
     return bench_df
 
 
@@ -100,6 +105,7 @@ def get_benchmark_plot(bench_df):
             "model",
             "backend.name",
             "backend.torch_dtype",
+            "optimizations",
             "forward.peak_memory(MB)",
             "generate.throughput(tokens/s)",
         ],
@@ -129,8 +135,9 @@ def get_benchmark_plot(bench_df):
                 "Model: %{customdata[0]}",
                 "Backend: %{customdata[1]}",
                 "Datatype: %{customdata[2]}",
-                "Peak Memory (MB): %{customdata[3]}",
-                "Throughput (tokens/s): %{customdata[4]}",
+                "Optimizations: %{customdata[3]}",
+                "Peak Memory (MB): %{customdata[4]}",
+                "Throughput (tokens/s): %{customdata[5]}",
                 "Average Open LLM Score: %{y}",
                 "Per 1000 Tokens Latency (s): %{x}",
             ]

src/assets/css_html_js.py

@@ -25,10 +25,9 @@ custom_css = """
     border: none;
 }
 
-/* Limit the width of the first AutoEvalColumn so that names don't expand too much */
 table td:first-child,
 table th:first-child {
-    max-width: 400px;
+    max-width: 300px;
     overflow: auto;
     white-space: nowrap;
 }