Update models.py

models.py

@@ -96,6 +96,15 @@ class OpenAIModelGPT3_5(BaseTCOModel):
         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")
@@ -108,16 +117,6 @@ class OpenAIModelGPT3_5(BaseTCOModel):
                                            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)
-        
-        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.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, 
@@ -140,6 +139,11 @@ class OpenSourceLlama2Model(BaseTCOModel):
         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.vm = gr.Textbox(value="2x A100 80GB NVLINK", 
                               visible=False,
@@ -148,6 +152,16 @@ class OpenSourceLlama2Model(BaseTCOModel):
         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.")
+        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="($) Price/1K input prompt tokens",
                                            interactive=False
@@ -156,6 +170,7 @@ class OpenSourceLlama2Model(BaseTCOModel):
                                            label="($) 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.source = gr.Markdown("""<span style="font-size: 16px; font-weight: 600; color: #212529;">Source</span>""")
         self.info = gr.Markdown("The cost per input and output tokens values above are from [these benchmark results](https://www.cursor.so/blog/llama-inference#user-content-fn-llama-paper)", 
                                  label="Source",