Spaces:
Running
Running
File size: 18,907 Bytes
50f19fa 29078ea 50f19fa 29078ea 50f19fa 0680f69 29078ea 50f19fa bdf4f51 50f19fa bdf4f51 98eb32d 50f19fa bdf4f51 98eb32d bdf4f51 7769b47 bdf4f51 98eb32d bdf4f51 98eb32d bdf4f51 98eb32d bdf4f51 db28a13 bdf4f51 db28a13 bdf4f51 98eb32d bdf4f51 a42de63 bdf4f51 2db3504 bdf4f51 50f19fa 0680f69 db28a13 7769b47 db28a13 7769b47 8e75c4f db28a13 7769b47 9793af4 7769b47 50f19fa db28a13 50f19fa 7769b47 50f19fa 0e893b5 0680f69 50f19fa 0e893b5 29078ea 50f19fa 2db3504 0e893b5 42b592e 847028b 7769b47 50f19fa 0680f69 50f19fa 9793af4 4424c49 9793af4 42b592e 2db3504 db28a13 7769b47 50f19fa db28a13 7769b47 29078ea 2db3504 50f19fa 0e893b5 7769b47 9793af4 7769b47 50f19fa db28a13 7769b47 0ad933c 98eb32d 0ad933c 98eb32d 0ad933c 50f19fa 0ad933c 98eb32d b1abf8e 98eb32d b1abf8e 98eb32d b1abf8e 98eb32d 9793af4 7769b47 0ad933c db28a13 98eb32d 0ad933c 50f19fa 7769b47 50f19fa 29078ea 9793af4 29078ea 50f19fa 0e893b5 7769b47 50f19fa 7769b47 fd49c3f 29078ea 4e90465 0680f69 50f19fa |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 |
from gradio.components import Component
import gradio as gr
import pandas as pd
from abc import ABC, abstractclassmethod
import inspect
class BaseTCOModel(ABC):
# TO DO: Find way to specify which component should be used for computing cost
def __setattr__(self, name, value):
if isinstance(value, Component):
self._components.append(value)
self.__dict__[name] = value
def __init__(self):
super(BaseTCOModel, self).__setattr__("_components", [])
self.use_case = None
def get_components(self) -> list[Component]:
return self._components
def get_components_for_cost_computing(self):
return self.components_for_cost_computing
def get_name(self):
return self.name
def register_components_for_cost_computing(self):
args = inspect.getfullargspec(self.compute_cost_per_token)[0][1:]
self.components_for_cost_computing = [self.__getattribute__(arg) for arg in args]
@abstractclassmethod
def compute_cost_per_token(self):
pass
@abstractclassmethod
def render(self):
pass
def set_name(self, name):
self.name = name
def set_latency(self, latency):
self.latency = latency
def get_latency(self):
return self.latency
class OpenAIModelGPT4(BaseTCOModel):
def __init__(self):
self.set_name("(SaaS) OpenAI GPT4")
self.set_latency("10s") #Default value for GPT4
super().__init__()
def render(self):
def define_cost_per_token(context_length):
if context_length == "128K":
cost_per_1k_input_tokens = 0.01
cost_per_1k_output_tokens = 0.03
else:
cost_per_1k_input_tokens = 0.06
cost_per_1k_output_tokens = 0.12
return cost_per_1k_input_tokens, cost_per_1k_output_tokens
self.context_length = gr.Dropdown(["128K"], value="128K", interactive=True,
label="Context size",
visible=False, info="Number of tokens the model considers when processing text")
self.input_tokens_cost_per_token = gr.Number(0.01, visible=False,
label="($) Price/1K input prompt tokens",
interactive=False
)
self.output_tokens_cost_per_token = gr.Number(0.03, visible=False,
label="($) Price/1K output prompt tokens",
interactive=False
)
self.info = gr.Markdown("The cost per input and output tokens values are from OpenAI's [pricing web page](https://openai.com/pricing)", interactive=False, visible=False)
self.context_length.change(define_cost_per_token, inputs=self.context_length, outputs=[self.input_tokens_cost_per_token, self.output_tokens_cost_per_token])
self.labor = gr.Number(0, visible=False,
label="($) Labor cost per month",
info="This is an estimate of the labor cost of the AI engineer in charge of deploying the model",
interactive=True
)
def compute_cost_per_token(self, input_tokens_cost_per_token, output_tokens_cost_per_token, labor):
cost_per_input_token = (input_tokens_cost_per_token / 1000)
cost_per_output_token = (output_tokens_cost_per_token / 1000)
return cost_per_input_token, cost_per_output_token, labor
class MistralO(BaseTCOModel):
def __init__(self):
self.set_name("(SaaS) Mistral API")
self.set_latency("5s") #Average latency value for GPT3.5 Turbo
super().__init__()
def render(self):
def define_cost_per_token(context_length):
if context_length == "4K":
cost_per_1k_input_tokens = 0.0015
cost_per_1k_output_tokens = 0.002
else:
cost_per_1k_input_tokens = 0.003
cost_per_1k_output_tokens = 0.004
return cost_per_1k_input_tokens, cost_per_1k_output_tokens
self.context_length = gr.Dropdown(choices=["4K", "16K"], value="4K", interactive=True,
label="Context size",
visible=False, info="Number of tokens the model considers when processing text")
self.input_tokens_cost_per_token = gr.Number(0.0015, visible=False,
label="($) Price/1K input prompt tokens",
interactive=False
)
self.output_tokens_cost_per_token = gr.Number(0.002, visible=False,
label="($) Price/1K output prompt tokens",
interactive=False
)
self.info = gr.Markdown("The cost per input and output tokens values are from OpenAI's [pricing web page](https://openai.com/pricing)", interactive=False, visible=False)
self.context_length.change(define_cost_per_token, inputs=self.context_length, outputs=[self.input_tokens_cost_per_token, self.output_tokens_cost_per_token])
self.labor = gr.Number(0, visible=False,
label="($) Labor cost per month",
info="This is an estimate of the labor cost of the AI engineer in charge of deploying the model",
interactive=True
)
def compute_cost_per_token(self, input_tokens_cost_per_token, output_tokens_cost_per_token, labor):
cost_per_input_token = (input_tokens_cost_per_token / 1000)
cost_per_output_token = (output_tokens_cost_per_token / 1000)
return cost_per_input_token, cost_per_output_token, labor
class OpenAIModelGPT3_5(BaseTCOModel):
def __init__(self):
self.set_name("(SaaS) OpenAI GPT3.5 Turbo")
self.set_latency("5s") #Average latency value for GPT3.5 Turbo
super().__init__()
def render(self):
def define_cost_per_token(context_length):
if context_length == "4K":
cost_per_1k_input_tokens = 0.0015
cost_per_1k_output_tokens = 0.002
else:
cost_per_1k_input_tokens = 0.003
cost_per_1k_output_tokens = 0.004
return cost_per_1k_input_tokens, cost_per_1k_output_tokens
self.context_length = gr.Dropdown(choices=["4K", "16K"], value="4K", interactive=True,
label="Context size",
visible=False, info="Number of tokens the model considers when processing text")
self.input_tokens_cost_per_token = gr.Number(0.0015, visible=False,
label="($) Price/1K input prompt tokens",
interactive=False
)
self.output_tokens_cost_per_token = gr.Number(0.002, visible=False,
label="($) Price/1K output prompt tokens",
interactive=False
)
self.info = gr.Markdown("The cost per input and output tokens values are from OpenAI's [pricing web page](https://openai.com/pricing)", interactive=False, visible=False)
self.context_length.change(define_cost_per_token, inputs=self.context_length, outputs=[self.input_tokens_cost_per_token, self.output_tokens_cost_per_token])
self.labor = gr.Number(0, visible=False,
label="($) Labor cost per month",
info="This is an estimate of the labor cost of the AI engineer in charge of deploying the model",
interactive=True
)
def compute_cost_per_token(self, input_tokens_cost_per_token, output_tokens_cost_per_token, labor):
cost_per_input_token = (input_tokens_cost_per_token / 1000)
cost_per_output_token = (output_tokens_cost_per_token / 1000)
return cost_per_input_token, cost_per_output_token, labor
class DIYLlama2Model(BaseTCOModel):
def __init__(self):
self.set_name("(Deploy yourself) Llama 2 70B")
self.set_latency("27s")
super().__init__()
def render(self):
def on_maxed_out_change(maxed_out, input_tokens_cost_per_token, output_tokens_cost_per_token):
output_tokens_cost_per_token = 0.06656
input_tokens_cost_per_token = 0.00052
r = maxed_out / 100
return input_tokens_cost_per_token * 0.65 / r, output_tokens_cost_per_token * 0.65/ r
self.source = gr.Markdown("""<span style="font-size: 16px; font-weight: 600; color: #212529;">Source</span>""", visible=False)
self.info = gr.Markdown("The cost per input and output tokens values below are from [these benchmark results](https://www.cursor.so/blog/llama-inference#user-content-fn-llama-paper) that were obtained using the following initial configurations.",
interactive=False,
visible=False)
self.vm = gr.Textbox(value="2x A100 80GB NVLINK",
visible=False,
label="Instance of VM with GPU",
)
self.vm_cost_per_hour = gr.Number(4.42, label="Instance cost ($) per hour",
interactive=False, visible=False)
self.info_vm = gr.Markdown("This price above is from [CoreWeave's pricing web page](https://www.coreweave.com/gpu-cloud-pricing)", interactive=False, visible=False)
self.maxed_out = gr.Slider(minimum=1, maximum=100, value=65, step=1, label="Maxed out", info="Estimated average percentage of total GPU memory that is used. The instantaneous value can go from very high when many users are using the service to very low when no one does.", visible=False)
self.info_maxed_out = gr.Markdown(r"""This percentage influences the input and output cost/token values, and more precisely the number of token/s. Here is the formula used:<br>
$CT = \frac{VM_C}{TS}$ where $TS = TS_{max} * \frac{MO}{100}$ <br>
with: <br>
$CT$ = Cost per Token (Input or output), <br>
$VM_C$ = VM Cost per second, <br>
$TS$ = Tokens per second (Input or output), <br>
$TS_{max}$ = Tokens per second when the GPU is maxed out at 100%, <br>
$MO$ = Maxed Out, <br>
""", interactive=False, visible=False)
self.input_tokens_cost_per_token = gr.Number(0.00052, visible=False,
label="($) Price/1K input prompt tokens",
interactive=False
)
self.output_tokens_cost_per_token = gr.Number(0.06656, visible=False,
label="($) Price/1K output prompt tokens",
interactive=False
)
self.maxed_out.change(on_maxed_out_change, inputs=[self.maxed_out, self.input_tokens_cost_per_token, self.output_tokens_cost_per_token], outputs=[self.input_tokens_cost_per_token, self.output_tokens_cost_per_token])
self.labor = gr.Number(5000, visible=False,
label="($) Labor cost per month",
info="This is an estimate of the labor cost of the AI engineer in charge of deploying the model",
interactive=True
)
def compute_cost_per_token(self, input_tokens_cost_per_token, output_tokens_cost_per_token, labor):
cost_per_input_token = (input_tokens_cost_per_token / 1000)
cost_per_output_token = (output_tokens_cost_per_token / 1000)
return cost_per_input_token, cost_per_output_token, labor
class DIYLlama2Model(BaseTCOModel):
def __init__(self):
self.set_name("(Deploy yourself) Llama 2/Mistral (et variante 7B)")
self.set_latency("6s")
super().__init__()
def render(self):
def on_maxed_out_change(maxed_out, input_tokens_cost_per_token, output_tokens_cost_per_token):
output_tokens_cost_per_token = 0.06656
input_tokens_cost_per_token = 0.00052
r = maxed_out / 100
return input_tokens_cost_per_token * 0.65 / r, output_tokens_cost_per_token * 0.65/ r
self.source = gr.Markdown("""<span style="font-size: 16px; font-weight: 600; color: #212529;">Source</span>""", visible=False)
self.info = gr.Markdown("The cost per input and output tokens values below are from [these benchmark results](https://www.cursor.so/blog/llama-inference#user-content-fn-llama-paper) that were obtained using the following initial configurations.",
interactive=False,
visible=False)
self.vm = gr.Textbox(value="2x A100 80GB NVLINK",
visible=False,
label="Instance of VM with GPU",
)
self.vm_cost_per_hour = gr.Number(4.42, label="Instance cost ($) per hour",
interactive=False, visible=False)
self.info_vm = gr.Markdown("This price above is from [CoreWeave's pricing web page](https://www.coreweave.com/gpu-cloud-pricing)", interactive=False, visible=False)
self.maxed_out = gr.Slider(minimum=1, maximum=100, value=65, step=1, label="Maxed out", info="Estimated average percentage of total GPU memory that is used. The instantaneous value can go from very high when many users are using the service to very low when no one does.", visible=False)
self.info_maxed_out = gr.Markdown(r"""This percentage influences the input and output cost/token values, and more precisely the number of token/s. Here is the formula used:<br>
$CT = \frac{VM_C}{TS}$ where $TS = TS_{max} * \frac{MO}{100}$ <br>
with: <br>
$CT$ = Cost per Token (Input or output), <br>
$VM_C$ = VM Cost per second, <br>
$TS$ = Tokens per second (Input or output), <br>
$TS_{max}$ = Tokens per second when the GPU is maxed out at 100%, <br>
$MO$ = Maxed Out, <br>
""", interactive=False, visible=False)
self.input_tokens_cost_per_token = gr.Number(0.00052, visible=False,
label="($) Price/1K input prompt tokens",
interactive=False
)
self.output_tokens_cost_per_token = gr.Number(0.06656, visible=False,
label="($) Price/1K output prompt tokens",
interactive=False
)
self.maxed_out.change(on_maxed_out_change, inputs=[self.maxed_out, self.input_tokens_cost_per_token, self.output_tokens_cost_per_token], outputs=[self.input_tokens_cost_per_token, self.output_tokens_cost_per_token])
self.labor = gr.Number(5000, visible=False,
label="($) Labor cost per month",
info="This is an estimate of the labor cost of the AI engineer in charge of deploying the model",
interactive=True
)
def compute_cost_per_token(self, input_tokens_cost_per_token, output_tokens_cost_per_token, labor):
cost_per_input_token = (input_tokens_cost_per_token / 1000)
cost_per_output_token = (output_tokens_cost_per_token / 1000)
return cost_per_input_token, cost_per_output_token, labor
class ModelPage:
def __init__(self, Models: BaseTCOModel):
self.models: list[BaseTCOModel] = []
for Model in Models:
model = Model()
self.models.append(model)
def render(self):
for model in self.models:
model.render()
model.register_components_for_cost_computing()
def get_all_components(self) -> list[Component]:
output = []
for model in self.models:
output += model.get_components()
return output
def get_all_components_for_cost_computing(self) -> list[Component]:
output = []
for model in self.models:
output += model.get_components_for_cost_computing()
return output
def make_model_visible(self, name:str, use_case: gr.Dropdown):
# First decide which indexes
output = []
for model in self.models:
if model.get_name() == name:
output+= [gr.update(visible=True)] * len(model.get_components())
# Set use_case value in the model
model.use_case = use_case
else:
output+= [gr.update(visible=False)] * len(model.get_components())
return output
def compute_cost_per_token(self, *args):
begin=0
current_model = args[-3]
current_input_tokens = args[-2]
current_output_tokens = args[-1]
for model in self.models:
model_n_args = len(model.get_components_for_cost_computing())
if current_model == model.get_name():
model_args = args[begin:begin+model_n_args]
cost_per_input_token, cost_per_output_token, labor_cost = model.compute_cost_per_token(*model_args)
model_tco = cost_per_input_token * current_input_tokens + cost_per_output_token * current_output_tokens
latency = model.get_latency()
return model_tco, latency, labor_cost
begin = begin+model_n_args |