update

app.py

@@ -40,6 +40,7 @@ with demo:
                     datatype_checkboxes,
                     optimization_checkboxes,
                     quantization_checkboxes,
+                    kernels_checkboxes,
                 ) = create_control_panel(machine=machine)
                 ####################### HARDWARE SUBTABS #######################
                 with gr.Tabs(elem_classes="subtabs"):
@@ -73,6 +74,7 @@ with demo:
                     datatype_checkboxes,
                     optimization_checkboxes,
                     quantization_checkboxes,
+                    kernels_checkboxes,
                     # interactive
                     columns_checkboxes,
                     search_bar,

src/assets.py

@@ -48,6 +48,9 @@ custom_css = """
 #quantization-checkboxes span {
     font-size: 16px !important;
 }
+#kernel-checkboxes span {
+    font-size: 16px !important;
+}
 
 #leaderboard-table td:first-child,
 #leaderboard-table th:first-child {

src/map.py

@@ -54,7 +54,7 @@ def get_lat_score_mem_fig(llm_perf_df):
         height=600,
     )
     # update x range with 95 percentile of
-    fig.update_xaxes(range=[0, copy_df["End-to-End (s)"].quantile(0.95)])
+    fig.update_xaxes(range=[-0.5, copy_df["End-to-End (s)"].quantile(0.90)])
 
     return fig
 
@@ -62,6 +62,7 @@ def get_lat_score_mem_fig(llm_perf_df):
 def create_lat_score_mem_plot(llm_perf_df):
     # descriptive text
     gr.HTML("👆 Hover over the points 👆 for additional information. ", elem_id="text")
+    gr.HTML("📊 We only show the top 90% LLMs based on latency ⌛", elem_id="text")
     # get figure
     fig = get_lat_score_mem_fig(llm_perf_df)
     # create plot

src/panel.py

@@ -13,14 +13,14 @@ def create_control_panel(machine: str):
     machine_textbox = gr.Textbox(value=machine, visible=False)
     with gr.Accordion("Control Panel 🎛️", open=False, elem_id="control-panel"):
         with gr.Row():
-            with gr.Column(scale=1, variant="panel"):
+            with gr.Column(scale=2, variant="panel"):
                 score_slider = gr.Slider(
                     label="Open LLM Score (%) 📈",
                     info="🎚️ Slide to minimum Open LLM score",
                     value=0,
                     elem_id="threshold-slider",
                 )
-            with gr.Column(scale=1, variant="panel"):
+            with gr.Column(scale=2, variant="panel"):
                 memory_slider = gr.Slider(
                     label="Peak Memory (MB) 📈",
                     info="🎚️ Slide to maximum Peak Memory",
@@ -54,7 +54,8 @@ def create_control_panel(machine: str):
                     info="☑️ Select the optimization",
                     elem_id="optimization-checkboxes",
                 )
-            with gr.Column(scale=2, variant="panel"):
+        with gr.Row():
+            with gr.Column(scale=1, variant="panel"):
                 quantization_checkboxes = gr.CheckboxGroup(
                     label="Quantizations 🗜️",
                     choices=[
@@ -75,6 +76,27 @@ def create_control_panel(machine: str):
                     elem_id="quantization-checkboxes",
                     elem_classes="boxed-option",
                 )
+            with gr.Column(scale=1, variant="panel"):
+                kernels_checkboxes = gr.CheckboxGroup(
+                    label="Kernels ⚛️",
+                    choices=[
+                        "No Kernel",
+                        "GPTQ.ExllamaV1",
+                        "GPTQ.ExllamaV2",
+                        "AWQ.GEMM",
+                        "AWQ.GEMV",
+                    ],
+                    value=[
+                        "No Kernel",
+                        "GPTQ.ExllamaV1",
+                        "GPTQ.ExllamaV2",
+                        "AWQ.GEMM",
+                        "AWQ.GEMV",
+                    ],
+                    info="☑️ Select the custom kernels",
+                    elem_id="kernel-checkboxes",
+                    elem_classes="boxed-option",
+                )
         with gr.Row():
             filter_button = gr.Button(
                 value="Filter 🚀",
@@ -91,6 +113,7 @@ def create_control_panel(machine: str):
         datatype_checkboxes,
         optimization_checkboxes,
         quantization_checkboxes,
+        kernels_checkboxes,
     )
 
 
@@ -103,6 +126,7 @@ def filter_rows_fn(
     precisions,
     attentions,
     quantizations,
+    kernels,
     # interactive
     columns,
     search,
@@ -116,6 +140,7 @@ def filter_rows_fn(
         & llm_perf_df["Precision 📥"].isin(precisions)
         & llm_perf_df["Attention 👁️"].isin(attentions)
         & llm_perf_df["Quantization 🗜️"].isin(quantizations)
+        & llm_perf_df["Kernel ⚛️"].isin(kernels)
         & (llm_perf_df["Open LLM Score (%)"] >= score)
         & (llm_perf_df["Memory (MB)"] <= memory)
     ]
@@ -154,6 +179,7 @@ def create_control_callback(
     datatype_checkboxes,
     optimization_checkboxes,
     quantization_checkboxes,
+    kernels_checkboxes,
     # interactive
     columns_checkboxes,
     search_bar,
@@ -179,6 +205,7 @@ def create_control_callback(
             datatype_checkboxes,
             optimization_checkboxes,
             quantization_checkboxes,
+            kernels_checkboxes,
             # interactive
             columns_checkboxes,
             search_bar,

src/utils.py

@@ -97,7 +97,7 @@ def process_kernels(x):
     ):
         return "AWQ.GEMV"
     else:
-        return "Unquantized"
+        return "No Kernel"
 
 
 # def change_tab(query_param):