Spaces:
Running
Running
File size: 8,258 Bytes
7f9a235 d1cb523 7f9a235 d1cb523 7f9a235 d1cb523 7f9a235 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 |
import gradio as gr
def get_base_backend_config(backend_name="pytorch"):
return [
# seed
gr.Textbox(
value=42,
label=f"{backend_name}.seed",
info="Sets seed for reproducibility",
),
# inter_op_num_threads
gr.Textbox(
value="null",
label=f"{backend_name}.inter_op_num_threads",
info="Use null for default and -1 for cpu_count()",
),
# intra_op_num_threads
gr.Textbox(
value="null",
label=f"{backend_name}.intra_op_num_threads",
info="Use null for default and -1 for cpu_count()",
),
# initial_isolation_check
gr.Checkbox(
value=True,
label=f"{backend_name}.initial_isolation_check",
info="Makes sure that initially, no other process is running on the target device",
),
# continous_isolation_check
gr.Checkbox(
value=True,
label=f"{backend_name}.continous_isolation_check",
info="Makes sure that throughout the benchmark, no other process is running on the target device",
),
# delete_cache
gr.Checkbox(
value=False,
label=f"{backend_name}.delete_cache",
info="Deletes model cache (weights & configs) after benchmark is done",
),
]
def get_pytorch_config():
return get_base_backend_config(backend_name="pytorch") + [
# no_weights
gr.Checkbox(
value=False,
label="pytorch.no_weights",
info="Generates random weights instead of downloading pretrained ones",
),
# # device_map
# gr.Dropdown(
# value="null",
#
# label="pytorch.device_map",
# choices=["null", "auto", "sequential"],
# info="Use null for default and `auto` or `sequential` the same way as in `from_pretrained`",
# ),
# torch_dtype
gr.Dropdown(
value="null",
label="pytorch.torch_dtype",
choices=["null", "bfloat16", "float16", "float32", "auto"],
info="Use null for default and `auto` for automatic dtype selection",
),
# amp_autocast
gr.Checkbox(
value=False,
label="pytorch.amp_autocast",
info="Enables Pytorch's native Automatic Mixed Precision",
),
# amp_dtype
gr.Dropdown(
value="null",
label="pytorch.amp_dtype",
info="Use null for default",
choices=["null", "bfloat16", "float16"],
),
# torch_compile
gr.Checkbox(
value=False,
label="pytorch.torch_compile",
info="Compiles the model with torch.compile",
),
# bettertransformer
gr.Checkbox(
value=False,
label="pytorch.bettertransformer",
info="Applies optimum.BetterTransformer for fastpath anf optimized attention",
),
# quantization_scheme
gr.Dropdown(
value="null",
choices=["null", "gptq", "bnb"],
label="pytorch.quantization_scheme",
info="Use null for no quantization",
),
# # use_ddp
# gr.Checkbox(
# value=False,
#
# label="pytorch.use_ddp",
# info="Uses DistributedDataParallel for multi-gpu training",
# ),
# peft_strategy
gr.Textbox(
value="null",
label="pytorch.peft_strategy",
),
]
def get_onnxruntime_config():
return get_base_backend_config(backend_name="onnxruntime")
# no_weights
# no_weights: bool = False
# # export options
# export: bool = True
# use_cache: bool = True
# use_merged: bool = False
# torch_dtype: Optional[str] = None
# # provider options
# provider: str = "${infer_provider:${device}}"
# device_id: Optional[int] = "${oc.deprecated:backend.provider_options.device_id}"
# provider_options: Dict[str, Any] = field(default_factory=lambda: {"device_id": "${infer_device_id:${device}}"})
# # inference options
# use_io_binding: bool = "${is_gpu:${device}}"
# enable_profiling: bool = "${oc.deprecated:backend.session_options.enable_profiling}"
# session_options: Dict[str, Any] = field(
# default_factory=lambda: {"enable_profiling": "${is_profiling:${benchmark.name}}"}
# )
# # optimization options
# optimization: bool = False
# optimization_config: Dict[str, Any] = field(default_factory=dict)
# # quantization options
# quantization: bool = False
# quantization_config: Dict[str, Any] = field(default_factory=dict)
# # calibration options
# calibration: bool = False
# calibration_config: Dict[str, Any] = field(default_factory=dict)
# # null, O1, O2, O3, O4
# auto_optimization: Optional[str] = None
# auto_optimization_config: Dict[str, Any] = field(default_factory=dict)
# # null, arm64, avx2, avx512, avx512_vnni, tensorrt
# auto_quantization: Optional[str] = None
# auto_quantization_config: Dict[str, Any] = field(default_factory=dict)
# # ort-training is basically a different package so we might need to seperate these two backends in the future
# use_inference_session: bool = "${is_inference:${benchmark.name}}"
# # training options
# use_ddp: bool = False
# ddp_config: Dict[str, Any] = field(default_factory=dict)
# # peft options
# peft_strategy: Optional[str] = None
# peft_config: Dict[str, Any] = field(default_factory=dict)
def get_openvino_config():
return get_base_backend_config(backend_name="openvino")
def get_neural_compressor_config():
return get_base_backend_config(backend_name="neural-compressor")
def get_text_generation_inference_config():
return get_base_backend_config(backend_name="text-generation-inference")
def get_inference_config():
return [
# duration
gr.Textbox(
value=10,
label="inference.duration",
info="Minimum duration of benchmark in seconds",
),
# warmup runs
gr.Textbox(
value=10,
label="inference.warmup_runs",
info="Number of warmup runs before measurements",
),
# memory
gr.Checkbox(
value=False,
label="inference.memory",
info="Measures the peak memory footprint",
),
# energy
gr.Checkbox(
value=False,
label="inference.energy",
info="Measures energy consumption and carbon emissions",
),
# input_shapes
gr.Dataframe(
type="array",
value=[[2, 16]],
row_count=(1, "static"),
col_count=(2, "dynamic"),
label="inference.input_shapes",
headers=["batch_size", "sequence_length"],
info="Controllable input shapes, add more columns for more inputs",
),
# forward kwargs
gr.Dataframe(
type="array",
value=[[False]],
headers=["return_dict"],
row_count=(1, "static"),
col_count=(1, "dynamic"),
label="inference.forward_kwargs",
info="Keyword arguments for the forward pass, add more columns for more arguments",
),
]
def get_training_config():
return [
# warmup steps
gr.Textbox(
value=40,
label="training.warmup_steps",
),
# dataset_shapes
gr.Dataframe(
type="array",
value=[[500, 16]],
headers=["dataset_size", "sequence_length"],
row_count=(1, "static"),
col_count=(2, "dynamic"),
label="training.dataset_shapes",
),
# training_arguments
gr.Dataframe(
value=[[2]],
type="array",
row_count=(1, "static"),
col_count=(1, "dynamic"),
label="training.training_arguments",
headers=["per_device_train_batch_size"],
),
]
|