In this demo application, we help you compare different AI model services, such as Open source or SaaS solutions, based on the Total Cost of Ownership for their deployment. Please note that we focus on getting the service up and running, but not the maintenance that follows.
First, you'll have to select your use case. Then, select the two model service options you'd like to compare. Depending on the options you chose, you could be able to customize 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.
""" 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, labor_cost1, labor_cost2, dropdown, dropdown2): r = tco1 / tco2 comparison_result3 = "" if r < 1: comparison_result = f"The cost/request of the {dropdown2} service is {1/r:.5f} times more expensive than the one of the {dropdown} service." if labor_cost1 > labor_cost2: comparison_result2 = f"However, it will cost a lot more to deploy the set-up for the {dropdown} service because the labor cost per month is of ${labor_cost1}, compared to a labor cost per month of ${labor_cost2} for the {dropdown2} solution." meeting_point = (labor_cost2 - labor_cost1) / (tco1 - tco2) comparison_result3 = f"The number of requests you need to achieve in a month to have the labor cost of the {dropdown} service be absorbed and both solution TCOs be equal would be of {meeting_point:.0f}." elif labor_cost1 < labor_cost2: comparison_result2 = f"Additionally, it will cost a lot more to deploy the set-up for the {dropdown2} service because the labor cost per month is of ${labor_cost2}, compared to a labor cost per month of ${labor_cost1} for the {dropdown} solution." else: comparison_result2 = f"Additionally, both services have the same labor cost per month." elif r > 1: comparison_result = f"The cost/request of the {dropdown2} service is {r:.5f} times cheaper than the one of the {dropdown} service." if labor_cost1 > labor_cost2: comparison_result2 = f"Additionally, it will cost a lot more to deploy the set-up for the {dropdown} service because the labor cost per month is of ${labor_cost1}, compared to a labor cost per month of ${labor_cost2} for the {dropdown2} solution." elif labor_cost1 < labor_cost2: comparison_result2 = f"However, it will cost a lot more to deploy the set-up for the {dropdown2} service because the labor cost per month is of ${labor_cost2}, compared to a labor cost per month of ${labor_cost1} for the {dropdown} solution." meeting_point = (labor_cost2 - labor_cost1) / (tco1 - tco2) comparison_result3 = f"The number of requests you need to achieve in a month to have the labor cost of the {dropdown2} service be absorbed and both solution TCOs be equal would be of {meeting_point:.0f}." else: comparison_result2 = f"Additionally, both services have the same labor cost per month." else: comparison_result = f"Both solutions have the same cost/request." if labor_cost1 > labor_cost2: comparison_result2 = f"However, it will cost a lot more to deploy the set-up for the {dropdown} service because the labor cost per month is of ${labor_cost1}, compared to a labor cost per month of ${labor_cost2} for the {dropdown2} solution." elif labor_cost1 < 1: comparison_result2 = f"However, it will cost a lot more to deploy the set-up for the {dropdown2} service because the labor cost per month is of ${labor_cost2}, compared to a labor cost per month of ${labor_cost1} for the {dropdown} solution." else: comparison_result2 = f"Additionally, both services have the same labor cost per month." #return comparison_result + "\n" + "\n" + comparison_result2 + "\n" + "\n" + comparison_result3 return gr.update(value=comparison_result), comparison_result3 def create_table(tco1, tco2, labor_cost1, labor_cost2, dropdown, dropdown2): list_values = [] labor_cost1 = round(labor_cost1, 1) labor_cost2 = round(labor_cost2, 1) first_sol = [tco1, labor_cost1] second_sol = [tco2, labor_cost2] list_values.append(first_sol) list_values.append(second_sol) data = pd.DataFrame(list_values, index=[dropdown, dropdown2], columns=["Cost/request ($) ", "Labor Cost ($/month)"]) formatted_data = data.copy() formatted_data["Cost/request ($) "] = formatted_data["Cost/request ($) "].apply('{:.5f}'.format) formatted_data["Labor Cost ($/month)"] = formatted_data["Labor Cost ($/month)"].apply('{:.1f}'.format) styled_data = formatted_data.style\ .set_properties(**{'background-color': '#050f19', 'color': '#ffffff', 'border-color': '#ffffff', 'border-width': '1px', 'border-style': 'solid'})\ .to_html() return gr.update(value=styled_data) def update_plot(tco1, tco2, dropdown, dropdown2, labour_cost1, labour_cost2): request_ranges = [100, 200, 300, 400, 500, 1000, 10000] costs_tco1 = [(tco1 * req + labour_cost1) for req in request_ranges] costs_tco2 = [(tco2 * req + labour_cost2) for req in request_ranges] data = pd.DataFrame({ "Number of requests": request_ranges * 2, "Cost ($)": 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"]) light_grey = Color( name="light_grey", c50="#e0e0e0", c100="#e0e0e0", c200="#e0e0e0", c300="#e0e0e0", c400="#e0e0e0", c500="#e0e0e0", c600="#e0e0e0", c700="#e0e0e0", c800="#e0e0e0", c900="#e0e0e0", c950="#e0e0e0", ) class Style(Base): def __init__( self, *, primary_hue: colors.Color | str = light_grey, secondary_hue: colors.Color | str = light_grey, neutral_hue: colors.Color | str = light_grey, spacing_size: sizes.Size | str = sizes.spacing_md, radius_size: sizes.Size | str = sizes.radius_md, text_size: sizes.Size | str = sizes.text_md, font: fonts.Font | str | Iterable[fonts.Font | str] = (fonts.GoogleFont("Sora")), font_mono: fonts.Font | str | Iterable[fonts.Font | str] = (fonts.GoogleFont("Sora")), ): super().__init__( primary_hue=primary_hue, secondary_hue=secondary_hue, neutral_hue=neutral_hue, spacing_size=spacing_size, radius_size=radius_size, text_size=text_size, font=font, font_mono=font_mono, ) super().set( background_fill_primary="#050f19", #The color of the background of blocks background_fill_secondary="#050f19", block_background_fill="#050f19", #The color of the background of blocks block_background_fill_dark="#050f19", border_color_primary="#050f19", #The border of a block such as dropdown border_color_primary_dark="#050f19", link_text_color="#f0ba2d", #The color for links link_text_color_dark="#f0ba2d", block_info_text_color="ffffff", block_info_text_color_dark="ffffff", block_border_color="#050f19", #The border color around an item (e.g. Accordion) block_border_color_dark="#050f19", block_shadow="*shadow_drop_lg", #form_gap_width="*spacing_md", #The border gap between form elements, (e.g. consecutive textboxes) input_background_fill="#081527", #The background of an input field input_background_fill_dark="#081527", input_border_color="#050f19", input_border_color_dark="#050f19", input_border_width="2px", block_label_background_fill="#f0ba2d", block_label_background_fill_dark="#f0ba2d", block_label_border_color=None, block_label_border_color_dark=None, block_label_text_color="#050f19", block_label_text_color_dark="#050f19", button_primary_background_fill="#ffffff", button_primary_border_color="#ffffff", button_primary_text_color="#050f19", button_shadow="*shadow_drop_lg", block_title_background_fill="#f0ba2d", #The background of the title of a form element (e.g. textbox). block_title_background_fill_dark="#f0ba2d", block_title_radius="*radius_sm",#The corner radius of the title of a form element (e.g. textbox). block_title_text_color="#050f19", #The text color of the title of a form element (e.g. textbox). block_title_text_color_dark="#050f19", block_title_text_size="*text_lg", block_title_border_width="2px", #The border around the title of a form element (e.g. textbox) block_title_border_width_dark="2px", block_title_border_color="#f0ba2d", block_title_border_color_dark="#f0ba2d", body_background_fill="#050f19", body_background_fill_dark="#050f19", body_text_color="#ffffff", #The default text color. body_text_color_dark="#ffffff", body_text_color_subdued="#ffffff", body_text_color_subdued_dark="#ffffff", slider_color="*secondary_300", slider_color_dark="*secondary_600", ) style = Style() with gr.Blocks(theme=style) 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(): # gr.Markdown(value=text0) with gr.Row(): use_case = gr.Dropdown(["Summarize", "Question-Answering", "Classification"], value="Question-Answering", label=" Describe your use case ") with gr.Accordion("Click here if you want to customize the number of input and output tokens per request", open=False): with gr.Row(): input_tokens = gr.Slider(minimum=1, maximum=1000, value=300, step=1, label=" Input tokens per request", info="We suggest a value that we believe 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=" Output tokens per request", info="We suggest a value that we believe 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(): #gr.Markdown(value=text1) page1 = models.ModelPage(Models) dropdown = gr.Dropdown(model_names, interactive=True, label=" First AI service option ") with gr.Accordion("Click here for more information on the computation parameters for your first AI service option", open=False): page1.render() with gr.Column(): #gr.Markdown(value=text2) page2 = models.ModelPage(Models) dropdown2 = gr.Dropdown(model_names, interactive=True, label=" Second AI service option ") with gr.Accordion("Click here 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()) #gr.Markdown(value=text3) compute_tco_btn = gr.Button("Compute & Compare", size="lg", variant="primary", scale=1) tco1 = gr.State() tco2 = gr.State() labor_cost1 = gr.State() labor_cost2 = gr.State() with gr.Row(): with gr.Column(): tco_output = gr.Text("Cost/request 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 the AI model service deployment TCO.") latency_info = gr.Markdown() with gr.Accordion("Click here to see the formula", open=False): tco_formula = gr.Markdown() with gr.Column(): tco_output2 = gr.Text("Cost/request 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 the AI model service deployment TCO.") latency_info2 = gr.Markdown() with gr.Accordion("Click here to see the formula", open=False): tco_formula2 = gr.Markdown() with gr.Row(): with gr.Column(scale=1): info = gr.Text("Click on Compute & Compare to get the results", label=" Comparison of cost/request and TCOs") table = gr.Markdown() ratio = gr.Markdown() with gr.Column(scale=2): 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, labor_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, labor_cost2]).then(create_table, inputs=[tco1, tco2, labor_cost1, labor_cost2, dropdown, dropdown2], outputs=table).then(compare, inputs=[tco1, tco2, labor_cost1, labor_cost2, dropdown, dropdown2], outputs=[info, ratio]).then(update_plot, inputs=[tco1, tco2, dropdown, dropdown2, labor_cost1, labor_cost2], outputs=plot) demo.launch(debug=True)