|
import gradio as gr |
|
import models |
|
import pandas as pd |
|
|
|
text = "<h1 style='text-align: center; color: midnightblue; font-size: 40px;'>TCO Comparison Calculator" |
|
text0 = "<h1 style='text-align: center; color: midnightblue; font-size: 30px;'>Describe your use case" |
|
text1 = "<h1 style='text-align: center; color: midnightblue; font-size: 25px;'>First option" |
|
text2 = "<h1 style='text-align: center; color: midnightblue; font-size: 25px;'>Second option" |
|
text3 = "<h1 style='text-align: center; color: midnightblue; font-size: 30px;'>Compute and compare TCOs" |
|
description=f""" |
|
<p>In this demo application, we help you compare different AI model services, such as Open source or SaaS solutions.</p> |
|
<p>First, you'll have to choose how you want to use the AI model service based on your use case. Then, select the two model service solutions you'd like to compare. Depending on the solution you chose, you could be able to modify some parameters of the set-up. Eventually, we will provide you with the cost of deployment for the selected model services, as a function of the number of requests. You can compare both solutions to evaluate which one best suits your needs.</p> |
|
""" |
|
|
|
def on_use_case_change(use_case): |
|
if use_case == "Summarize": |
|
return gr.update(value=500), gr.update(value=200) |
|
elif use_case == "Question-Answering": |
|
return gr.update(value=300), gr.update(value=300) |
|
else: |
|
return gr.update(value=50), gr.update(value=10) |
|
|
|
def compare(tco1, tco2): |
|
r = tco1 / tco2 |
|
if r < 1: |
|
comparison_result = f"Second solution's cost/request is {1/r:.5f} times more expensive than the first" |
|
elif r > 1: |
|
comparison_result = f"Second solution's cost/request is {r:.5f} times cheaper than the first" |
|
else: |
|
comparison_result = "Both solutions will cost the same." |
|
return comparison_result |
|
|
|
def update_plot(tco1, tco2, dropdown, dropdown2, labour_cost1, labour_cost2): |
|
|
|
request_ranges = [100, 200, 300, 400, 500, 1000, 10000] |
|
costs_tco1 = [(tco1 * req + labour_cost1) for req in request_ranges] |
|
costs_tco2 = [(tco2 * req + labour_cost2) for req in request_ranges] |
|
|
|
data = pd.DataFrame({ |
|
"Number of requests": request_ranges * 2, |
|
"Cost ($)": costs_tco1 + costs_tco2, |
|
"AI model service": [dropdown] * len(request_ranges) + [dropdown2] * len(request_ranges) |
|
} |
|
) |
|
return gr.LinePlot.update(data, x="Number of requests", y="Cost ($)",color="AI model service",color_legend_position="bottom", title="Total Cost of Model Serving for one month", height=300, width=500, tooltip=["Number of requests", "Cost ($)", "AI model service"]) |
|
|
|
with gr.Blocks(theme=gr.themes.Soft()) as demo: |
|
Models: list[models.BaseTCOModel] = [models.OpenAIModel, models.CohereModel, models.OpenSourceLlama2Model] |
|
model_names = [Model().get_name() for Model in Models] |
|
gr.Markdown(value=text) |
|
gr.Markdown(value=description) |
|
|
|
with gr.Row(): |
|
with gr.Column(): |
|
|
|
|
|
with gr.Row(): |
|
use_case = gr.Dropdown(["Summarize", "Question-Answering", "Classification"], value="Question-Answering", label="Describe your use case") |
|
with gr.Accordion("Open to customize the number of input and output tokens for your use case", open=False): |
|
with gr.Row(): |
|
input_tokens = gr.Slider(minimum=1, maximum=1000, value=300, step=1, label="Number of input token", info="We put a value that we find best suit your use case choice but feel free to adjust", interactive=True) |
|
output_tokens = gr.Slider(minimum=1, maximum=1000, value=300, step=1, label="Number of output token", info="We put a value that we find best suit your use case choice but feel free to adjust", interactive=True) |
|
with gr.Row(visible=False): |
|
num_users = gr.Number(value="1000", interactive = True, label="Number of users for your service") |
|
|
|
use_case.change(on_use_case_change, inputs=use_case, outputs=[input_tokens, output_tokens]) |
|
|
|
with gr.Row(): |
|
with gr.Column(): |
|
|
|
page1 = models.ModelPage(Models) |
|
dropdown = gr.Dropdown(model_names, interactive=True, label="AI service option 1") |
|
with gr.Accordion("Open for more information on the computation parameters for your first AI service option", open=False): |
|
page1.render() |
|
|
|
with gr.Column(): |
|
|
|
page2 = models.ModelPage(Models) |
|
dropdown2 = gr.Dropdown(model_names, interactive=True, label="AI service option 2") |
|
with gr.Accordion("Open for more information on the computation parameters for your second AI service option", open=False): |
|
page2.render() |
|
|
|
dropdown.change(page1.make_model_visible, inputs=[dropdown, use_case], outputs=page1.get_all_components()) |
|
dropdown2.change(page2.make_model_visible, inputs=[dropdown2, use_case], outputs=page2.get_all_components()) |
|
|
|
|
|
compute_tco_btn = gr.Button("Compute cost/request and TCOs", size="lg") |
|
tco1 = gr.State() |
|
tco2 = gr.State() |
|
labour_cost1 = gr.State() |
|
labour_cost2 = gr.State() |
|
|
|
with gr.Row(): |
|
with gr.Column(): |
|
tco_output = gr.Text("Output 1: ", label="Cost/request for the first option", info="This is only the infrastructure cost per request for deployment, the labor cost still has to be added for a Total Cost of Model Serving") |
|
latency_info = gr.Markdown() |
|
with gr.Accordion("Open to see the formula", open=False): |
|
tco_formula = gr.Markdown() |
|
|
|
with gr.Column(): |
|
tco_output2 = gr.Text("Output 2: ", label="Cost/request for the second option", info="This is only the infrastructure cost per request for deployment, the labor cost still has to be added for a Total Cost of Model Serving") |
|
latency_info2 = gr.Markdown() |
|
with gr.Accordion("Open to see the formula", open=False): |
|
tco_formula2 = gr.Markdown() |
|
|
|
with gr.Row(): |
|
with gr.Column(scale=1): |
|
ratio = gr.Text("Ratio: ", label="Ratio of cost/request for both solutions") |
|
with gr.Column(scale=3): |
|
plot = gr.LinePlot() |
|
|
|
compute_tco_btn.click(page1.compute_cost_per_token, inputs=page1.get_all_components_for_cost_computing() + [dropdown, input_tokens, output_tokens], outputs=[tco_output, tco1, tco_formula, latency_info, labour_cost1]).then(page2.compute_cost_per_token, inputs=page2.get_all_components_for_cost_computing() + [dropdown2, input_tokens, output_tokens], outputs=[tco_output2, tco2, tco_formula2, latency_info2, labour_cost2]).then(compare, inputs=[tco1, tco2], outputs=ratio).then(update_plot, inputs=[tco1, tco2, dropdown, dropdown2, labour_cost1, labour_cost2], outputs=plot) |
|
|
|
demo.launch(debug=True) |
