Update models.py

models.py

@@ -99,7 +99,7 @@ class OpenSourceLlama2Model(BaseTCOModel):
                          with: <br>
                          CT = Cost per Token <br>
                          VM_CH = VM Cost per Hour <br>
-                         TS = Tokens per Second (for an input length of 233 tokens) <br>
+                         TS = Tokens per Second <br>
                          MO = Maxed Out <br>
                          U = Used
                          """)
@@ -107,15 +107,16 @@ class OpenSourceLlama2Model(BaseTCOModel):
     
     def render(self):
         vm_choices = ["1x Nvidia A100 (Azure NC24ads A100 v4)",
-                      "2x Nvidia A100 (Azure NC48ads A100 v4)"]
+                      "2x Nvidia A100 (Azure NC48ads A100 v4)", 
+                      "4x Nvidia A100 (Azure NC48ads A100 v4)"]
         
         def on_model_change(model):
             if model == "Llama 2 7B":
-                return gr.Dropdown.update(choices=vm_choices)
+                return [gr.Dropdown.update(choices=vm_choices), gr.Markdown.update(visible=True), gr.Markdown.update(visible=False)]
             else:
-                not_supported_vm = ["1x Nvidia A100 (Azure NC24ads A100 v4)"]
+                not_supported_vm = ["1x Nvidia A100 (Azure NC24ads A100 v4)", "2x Nvidia A100 (Azure NC48ads A100 v4)"]
                 choices = [x for x in vm_choices if x not in not_supported_vm]
-                return gr.Dropdown.update(choices=choices)
+                return [gr.Dropdown.update(choices=choices), gr.Markdown.update(visible=False), gr.Markdown.update(visible=True)]
 
         def on_vm_change(model, vm):
             # TO DO: load info from CSV
@@ -123,6 +124,10 @@ class OpenSourceLlama2Model(BaseTCOModel):
                 return [gr.Number.update(value=3.6730), gr.Number.update(value=694.38)]
             elif model == "Llama 2 7B" and vm == "2x Nvidia A100 (Azure NC48ads A100 v4)":
                 return [gr.Number.update(value=7.346), gr.Number.update(value=1388.76)]
+            elif model == "Llama 2 7B" and vm == "4x Nvidia A100 (Azure NC48ads A100 v4)":
+                return [gr.Number.update(value=14.692), gr.Number.update(value=2777.52)]
+            elif model == "Llama 2 70B" and vm == "4x Nvidia A100 (Azure NC48ads A100 v4)":
+                return [gr.Number.update(value=14.692), gr.Number.update(value=18.6)]
         
         self.model = gr.Dropdown(["Llama 2 7B", "Llama 2 70B"], value="Llama 2 7B", label="OpenSource models", visible=False)
         self.vm = gr.Dropdown(vm_choices,
@@ -132,22 +137,67 @@ class OpenSourceLlama2Model(BaseTCOModel):
                               info="Your options for this choice depend on the model you previously chose"
                               )
         self.vm_cost_per_hour = gr.Number(3.6730, label="VM instance cost per hour", 
-                                      interactive=True, visible=False)
+                                      interactive=False, visible=False)
         self.tokens_per_second = gr.Number(694.38, visible=False,
                                            label="Number of tokens per second for this specific model and VM instance",
                                            interactive=False
                                            )
-        self.input_length = gr.Number(350, label="Average number of input tokens", 
-                                      interactive=True, visible=False)
+        self.input_length = gr.Number(233, label="Average number of input tokens", info="This is the number of input tokens used when the model was benchmarked to get the number of tokens/second it processes",
+                                      interactive=False, visible=False)
+        self.info_7B = gr.Markdown("To see the script used to benchmark the Llama2-7B model, [click here](https://example.com/script)", interactive=False, visible=False)
+        self.info_70B = gr.Markdown("To see the benchmark results used for the Llama2-70B model, [click here](https://www.cursor.so/blog/llama-inference#user-content-fn-llama-paper)", interactive=False, visible=False)
         
-        self.model.change(on_model_change, inputs=self.model, outputs=self.vm)
+        self.model.change(on_model_change, inputs=self.model, outputs=[self.vm, self.info_7B, self.info_70B])
         self.vm.change(on_vm_change, inputs=[self.model, self.vm], outputs=[self.vm_cost_per_hour, self.tokens_per_second])
         self.maxed_out = gr.Slider(minimum=0.01, value=50., step=0.01, label="% maxed out", 
-                                   info="How much the GPU is fully used.",
+                                   info="How much the GPU is fully used",
                                    interactive=True,
                                    visible=False)
         self.used = gr.Slider(minimum=0.01, value=50., step=0.01, label="% used", 
-                                   info="Percentage of time the GPU is used.",
+                                   info="Percentage of time the GPU is used",
+                                   interactive=True,
+                                   visible=False)
+
+    def compute_cost_per_token(self, vm_cost_per_hour, tokens_per_second, maxed_out, used):
+        cost_per_token = vm_cost_per_hour / (tokens_per_second * 3600 * maxed_out * used)
+        return cost_per_token 
+
+class OpenSourceDIY(BaseTCOModel):
+    
+    def __init__(self):
+        self.set_name("(Open source) DIY")
+        self.set_formula(r"""$CT = \frac{VM\_CH}{TS \times 3600 \times MO \times U}$<br>
+                         with: <br>
+                         CT = Cost per Token <br>
+                         VM_CH = VM Cost per Hour <br>
+                         TS = Tokens per Second <br>
+                         MO = Maxed Out <br>
+                         U = Used
+                         """)
+        super().__init__()
+    
+    def render(self):   
+        self.info = gr.Markdown("Compute the cost/token based on our formula below, using your own parameters", visible=False)
+        self.display_formula = gr.Markdown(r"""$CT = \frac{VM\_CH}{TS \times 3600 \times MO \times U}$<br>
+                         with: <br>
+                         CT = Cost per Token <br>
+                         VM_CH = VM Cost per Hour <br>
+                         TS = Tokens per Second <br>
+                         MO = Maxed Out <br>
+                         U = Used
+                         """, visible=False)
+        self.vm_cost_per_hour = gr.Number(3.5, label="VM instance cost per hour", 
+                                      interactive=True, visible=False)
+        self.tokens_per_second = gr.Number(700, visible=False,
+                                           label="Number of tokens per second for this specific model and VM instance",
+                                           interactive=True
+                                           )
+        self.maxed_out = gr.Slider(minimum=0.01, value=50., step=0.01, label="% maxed out", 
+                                   info="How much the GPU is fully used",
+                                   interactive=True,
+                                   visible=False)
+        self.used = gr.Slider(minimum=0.01, value=50., step=0.01, label="% used", 
+                                   info="Percentage of time the GPU is used",
                                    interactive=True,
                                    visible=False)
 
@@ -174,8 +224,8 @@ class CohereModel(BaseTCOModel):
             else:
                 return gr.Dropdown.update(choices=["Default", "Custom"])
             
-        self.use_case = gr.Dropdown(["Embed", "Generate", "Classify", "Summarize"], value="Generate",
-                                 label="Use case",
+        self.use_case = gr.Dropdown(["Generate", "Summarize"], value="Generate",
+                                 label="API",
                                  interactive=True, visible=False)
         self.model = gr.Dropdown(["Default", "Custom"], value="Default",
                                  label="Model",
@@ -189,21 +239,11 @@ class CohereModel(BaseTCOModel):
         use_case = use_case[0]        
         model = model[0]
 
-        if use_case == "Embed":
-            if model == "Default":
-                cost_per_1M_input_tokens = 0.4
-            else: 
-                cost_per_1M_input_tokens = 0.8
-        elif use_case == "Generate":
+        if use_case == "Generate":
             if model == "Default":
                 cost_per_1M_input_tokens = 15
             else: 
                 cost_per_1M_input_tokens = 30
-        elif use_case == "Classify":
-            if model == "Default":
-                cost_per_1M_input_tokens = 200
-            else: 
-                cost_per_1M_input_tokens = 200
         else:
             cost_per_1M_input_tokens = 15