Spaces:
Running
Running
File size: 5,834 Bytes
2fcb72a aa8b4d6 2fcb72a aa8b4d6 2fcb72a aa8b4d6 2fcb72a aa8b4d6 2fcb72a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 |
from typing import List
import gradio as gr
import pandas as pd
from src.leaderboard import get_leaderboard_df
from src.llm_perf import get_llm_perf_df
# from attention_implementations import get_attn_decode_fig, get_attn_prefill_fig
# from custom_kernels import get_kernel_decode_fig, get_kernel_prefill_fig
def create_control_panel(
machine: str,
backends: List[str],
hardware_provider: str,
hardware_type: str,
):
# controls
machine_value = gr.State(value=machine)
backends_value = gr.State(value=backends)
hardware_type_value = gr.State(value=hardware_type)
if hardware_provider == "ARM":
backends = ["llama_cpp"]
quantizations = ["Q8_0", "Q4_K_M", "Q4_0_4_4"]
else:
raise ValueError(f"Unknown hardware provider: {hardware_provider}")
with gr.Accordion("Control Panel", open=False, elem_id="control-panel"):
with gr.Row():
with gr.Column(scale=2, variant="panel"):
memory_slider = gr.Slider(
label="Model Size (GB)",
info="🎚️ Slide to maximum Model Size",
minimum=0,
maximum=16,
value=16,
elem_id="memory-slider",
)
with gr.Column(scale=1, variant="panel"):
quantization_checkboxes = gr.CheckboxGroup(
label="Quantizations",
choices=quantizations,
value=quantizations,
info="☑️ Select the quantization schemes",
elem_id="quantization-checkboxes",
elem_classes="boxed-option",
)
with gr.Row():
filter_button = gr.Button(
value="Filter 🚀",
elem_id="filter-button",
elem_classes="boxed-option",
)
return (
filter_button,
machine_value,
backends_value,
hardware_type_value,
memory_slider,
quantization_checkboxes,
)
def filter_rows_fn(
machine,
backends,
hardware_type,
# inputs
memory,
quantizations,
# interactive
columns,
search,
):
llm_perf_df = get_llm_perf_df(
machine=machine, backends=backends, hardware_type=hardware_type
)
# print(attentions)
# print(llm_perf_df["Attention 👁️"].unique())
filtered_llm_perf_df = llm_perf_df[
llm_perf_df["Model"].str.contains(search, case=False)
& llm_perf_df["Quantization"].isin(quantizations)
& llm_perf_df["Model Size (GB)"] <= memory
]
selected_filtered_llm_perf_df = select_columns_fn(
machine, backends, hardware_type, columns, search, filtered_llm_perf_df
)
# filtered_bt_prefill_fig = get_bt_prefill_fig(filtered_df)
# filtered_bt_decode_fig = get_bt_decode_fig(filtered_df)
# filtered_fa2_prefill_fig = get_fa2_prefill_fig(filtered_df)
# filtered_fa2_decode_fig = get_fa2_decode_fig(filtered_df)
# filtered_quant_prefill_fig = get_quant_prefill_fig(filtered_df)
# filtered_quant_decode_fig = get_quant_decode_fig(filtered_df)
return [
selected_filtered_llm_perf_df,
# filtered_bt_prefill_fig,
# filtered_bt_decode_fig,
# filtered_fa2_prefill_fig,
# filtered_fa2_decode_fig,
# filtered_quant_prefill_fig,
# filtered_quant_decode_fig,
]
def create_control_callback(
# button
filter_button,
# fixed
machine_value,
backends_value,
hardware_type_value,
# inputs
memory_slider,
quantization_checkboxes,
# interactive
columns_checkboxes,
search_bar,
# outputs
leaderboard_table,
# attn_prefill_plot,
# attn_decode_plot,
# fa2_prefill_plot,
# fa2_decode_plot,
# quant_prefill_plot,
# quant_decode_plot,
):
filter_button.click(
fn=filter_rows_fn,
inputs=[
# fixed
machine_value,
backends_value,
hardware_type_value,
# inputs
memory_slider,
quantization_checkboxes,
# interactive
columns_checkboxes,
search_bar,
],
outputs=[
leaderboard_table,
# attn_prefill_plot,
# attn_decode_plot,
# fa2_prefill_plot,
# fa2_decode_plot,
# quant_prefill_plot,
# quant_decode_plot,
],
)
def select_columns_fn(
machine, backends, hardware_type, columns, search, llm_perf_df=None
):
if llm_perf_df is None:
llm_perf_df = get_llm_perf_df(
machine=machine,
backends=backends,
hardware_type=hardware_type,
)
selected_leaderboard_df = get_leaderboard_df(llm_perf_df)
selected_leaderboard_df = selected_leaderboard_df[
selected_leaderboard_df["Model"].str.contains(search, case=False)
]
selected_leaderboard_df = selected_leaderboard_df[columns]
return selected_leaderboard_df
def create_select_callback(
# fixed
machine_value,
backends_value,
hardware_type_value,
# interactive
columns_checkboxes,
search_bar,
# outputs
leaderboard_table,
):
columns_checkboxes.change(
fn=select_columns_fn,
inputs=[
machine_value,
backends_value,
hardware_type_value,
columns_checkboxes,
search_bar,
],
outputs=[leaderboard_table],
)
search_bar.change(
fn=select_columns_fn,
inputs=[
machine_value,
backends_value,
hardware_type_value,
columns_checkboxes,
search_bar,
],
outputs=[leaderboard_table],
)
|