Spaces:
Sleeping
Sleeping
File size: 7,576 Bytes
ab5f5f1 dc685a9 ab5f5f1 a8a6326 ab5f5f1 0232cf1 ab5f5f1 0232cf1 a8a6326 0232cf1 ab5f5f1 0232cf1 ab5f5f1 0232cf1 ab5f5f1 dc685a9 ab5f5f1 dc685a9 ab5f5f1 0232cf1 ab5f5f1 0232cf1 ab5f5f1 0232cf1 a8a6326 0232cf1 ab5f5f1 dc685a9 ab5f5f1 0232cf1 ab5f5f1 0232cf1 ab5f5f1 0232cf1 a8a6326 0232cf1 ab5f5f1 dc685a9 ab5f5f1 a8a6326 0232cf1 a8a6326 0232cf1 29307cd a8a6326 0232cf1 a8a6326 0232cf1 a8a6326 0232cf1 a8a6326 0232cf1 a8a6326 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 |
import gradio as gr
from src.llm_perf import get_llm_perf_df
from src.leaderboard import get_leaderboard_df
from src.latency_score_memory import get_lat_score_mem_fig
from src.bettertransformer import get_bt_prefill_fig, get_bt_decode_fig
from src.flashattentionv2 import get_fa2_prefill_fig, get_fa2_decode_fig
from src.quantization_kernels import get_quant_prefill_fig, get_quant_decode_fig
def create_control_panel(machine: str = "hf-dgx-01"):
# controls
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"):
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"):
memory_slider = gr.Slider(
label="Peak Memory (MB) 📈",
info="🎚️ Slide to maximum Peak Memory",
minimum=0,
maximum=80 * 1024,
value=80 * 1024,
elem_id="memory-slider",
)
with gr.Column(scale=1):
backend_checkboxes = gr.CheckboxGroup(
label="Backends 🏭",
choices=["pytorch"],
value=["pytorch"],
info="☑️ Select the backends",
elem_id="backend-checkboxes",
)
with gr.Row():
with gr.Column(scale=1, variant="panel"):
datatype_checkboxes = gr.CheckboxGroup(
label="Load DTypes 📥",
choices=["float32", "float16", "bfloat16"],
value=["float32", "float16", "bfloat16"],
info="☑️ Select the load data types",
elem_id="dtype-checkboxes",
)
with gr.Column(scale=1, variant="panel"):
optimization_checkboxes = gr.CheckboxGroup(
label="Optimizations 🛠️",
choices=["None", "BetterTransformer", "FlashAttentionV2"],
value=["None", "BetterTransformer", "FlashAttentionV2"],
info="☑️ Select the optimization",
elem_id="optimization-checkboxes",
)
with gr.Column(scale=2):
quantization_checkboxes = gr.CheckboxGroup(
label="Quantizations 🗜️",
choices=[
"None",
"BnB.4bit",
"BnB.8bit",
"GPTQ.4bit",
"GPTQ.4bit+ExllamaV1",
"GPTQ.4bit+ExllamaV2",
"AWQ.4bit+GEMM",
"AWQ.4bit+GEMV",
],
value=[
"None",
"BnB.4bit",
"BnB.8bit",
"GPTQ.4bit",
"GPTQ.4bit+ExllamaV1",
"GPTQ.4bit+ExllamaV2",
"AWQ.4bit+GEMM",
"AWQ.4bit+GEMV",
],
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_textbox,
score_slider,
memory_slider,
backend_checkboxes,
datatype_checkboxes,
optimization_checkboxes,
quantization_checkboxes,
)
def filter_fn(
machine,
# inputs
score,
memory,
backends,
datatypes,
optimizations,
quantizations,
# interactive
columns,
search,
):
raw_df = get_llm_perf_df(machine=machine)
filtered_df = raw_df[
# raw_df["Model 🤗"].str.contains(model, case=False)
raw_df["Backend 🏭"].isin(backends)
& raw_df["DType 📥"].isin(datatypes)
& raw_df["Optimization 🛠️"].isin(optimizations)
& raw_df["Quantization 🗜️"].isin(quantizations)
& (raw_df["Open LLM Score (%)"] >= score)
& (raw_df["Allocated Memory (MB)"] <= memory)
]
filtered_leaderboard_df = select_fn(machine, columns, search)
filtered_lat_score_mem_fig = get_lat_score_mem_fig(filtered_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 [
filtered_leaderboard_df,
filtered_lat_score_mem_fig,
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_textbox,
# inputs
score_slider,
memory_slider,
backend_checkboxes,
datatype_checkboxes,
optimization_checkboxes,
quantization_checkboxes,
# interactive
columns_checkboxes,
search_bar,
# outputs
leaderboard_table,
lat_score_mem_plot,
bt_prefill_plot,
bt_decode_plot,
fa2_prefill_plot,
fa2_decode_plot,
quant_prefill_plot,
quant_decode_plot,
):
filter_button.click(
fn=filter_fn,
inputs=[
# fixed
machine_textbox,
# inputs
score_slider,
memory_slider,
backend_checkboxes,
datatype_checkboxes,
optimization_checkboxes,
quantization_checkboxes,
# interactive
columns_checkboxes,
search_bar,
],
outputs=[
leaderboard_table,
lat_score_mem_plot,
bt_prefill_plot,
bt_decode_plot,
fa2_prefill_plot,
fa2_decode_plot,
quant_prefill_plot,
quant_decode_plot,
],
)
def select_fn(machine, columns, search):
raw_df = get_llm_perf_df(machine=machine)
selected_leaderboard_df = get_leaderboard_df(raw_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_textbox,
# interactive
columns_checkboxes,
search_bar,
# outputs
leaderboard_table,
):
columns_checkboxes.change(
fn=select_fn,
inputs=[machine_textbox, columns_checkboxes, search_bar],
outputs=[leaderboard_table],
)
search_bar.change(
fn=select_fn,
inputs=[machine_textbox, columns_checkboxes, search_bar],
outputs=[leaderboard_table],
)
|