Spaces:
Running
Running
jadehardouin
commited on
Commit
•
0ad933c
1
Parent(s):
52a72f5
Update models.py
Browse files
models.py
CHANGED
@@ -59,10 +59,8 @@ class OpenAIModel(BaseTCOModel):
|
|
59 |
def on_model_change(model):
|
60 |
|
61 |
if model == "GPT-4":
|
62 |
-
print("GPT4")
|
63 |
return gr.Dropdown.update(choices=["8K", "32K"])
|
64 |
else:
|
65 |
-
print("GPT3.5")
|
66 |
return gr.Dropdown.update(choices=["4K", "16K"])
|
67 |
|
68 |
self.model = gr.Dropdown(["GPT-4", "GPT-3.5 Turbo"], value="GPT-4",
|
@@ -89,7 +87,7 @@ class OpenAIModel(BaseTCOModel):
|
|
89 |
else:
|
90 |
cost_per_1k_input_tokens = 0.003
|
91 |
|
92 |
-
cost_per_output_token = cost_per_1k_input_tokens *
|
93 |
|
94 |
return cost_per_output_token
|
95 |
|
@@ -156,7 +154,63 @@ class OpenSourceLlama2Model(BaseTCOModel):
|
|
156 |
def compute_cost_per_token(self, vm_cost_per_hour, tokens_per_second, maxed_out, used):
|
157 |
cost_per_token = vm_cost_per_hour / (tokens_per_second * 3600 * maxed_out * used)
|
158 |
return cost_per_token
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
159 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
160 |
class ModelPage:
|
161 |
|
162 |
def __init__(self, Models: BaseTCOModel):
|
|
|
59 |
def on_model_change(model):
|
60 |
|
61 |
if model == "GPT-4":
|
|
|
62 |
return gr.Dropdown.update(choices=["8K", "32K"])
|
63 |
else:
|
|
|
64 |
return gr.Dropdown.update(choices=["4K", "16K"])
|
65 |
|
66 |
self.model = gr.Dropdown(["GPT-4", "GPT-3.5 Turbo"], value="GPT-4",
|
|
|
87 |
else:
|
88 |
cost_per_1k_input_tokens = 0.003
|
89 |
|
90 |
+
cost_per_output_token = cost_per_1k_input_tokens * input_length / 1000
|
91 |
|
92 |
return cost_per_output_token
|
93 |
|
|
|
154 |
def compute_cost_per_token(self, vm_cost_per_hour, tokens_per_second, maxed_out, used):
|
155 |
cost_per_token = vm_cost_per_hour / (tokens_per_second * 3600 * maxed_out * used)
|
156 |
return cost_per_token
|
157 |
+
|
158 |
+
class CohereModel(BaseTCOModel):
|
159 |
+
|
160 |
+
def __init__(self):
|
161 |
+
self.set_name("(SaaS) Cohere")
|
162 |
+
self.set_formula(r"""$CT = \frac{CT\_1K \times 1000}{L}$ <br>
|
163 |
+
with: <br>
|
164 |
+
CT = Cost per output Token <br>
|
165 |
+
CT_1M = Cost per one million Tokens (from Cohere's pricing web page) <br>
|
166 |
+
L = Input Length
|
167 |
+
""")
|
168 |
+
super().__init__()
|
169 |
|
170 |
+
def render(self):
|
171 |
+
def on_use_case_change(use_case):
|
172 |
+
if use_case == "Summarize":
|
173 |
+
return gr.Dropdown.update(choices=["Default"])
|
174 |
+
else:
|
175 |
+
return gr.Dropdown.update(choices=["Default", "Custom"])
|
176 |
+
|
177 |
+
self.use_case = gr.Dropdown(["Embed", "Generate", "Classify", "Summarize"], value="Generate",
|
178 |
+
label="Use case",
|
179 |
+
interactive=True, visible=False)
|
180 |
+
self.model = gr.Dropdown(["Default", "Custom"], value="Default",
|
181 |
+
label="Model",
|
182 |
+
interactive=True, visible=False)
|
183 |
+
self.use_case.change(on_use_case_change, inputs=self.use_case, outputs=self.model)
|
184 |
+
self.input_length = gr.Number(350, label="Average number of input tokens",
|
185 |
+
interactive=True, visible=False)
|
186 |
+
|
187 |
+
def compute_cost_per_token(self, use_case, model, input_length):
|
188 |
+
"""Cost per token = """
|
189 |
+
use_case = use_case[0]
|
190 |
+
model = model[0]
|
191 |
+
|
192 |
+
if use_case == "Embed":
|
193 |
+
if model == "Default":
|
194 |
+
cost_per_1M_input_tokens = 0.4
|
195 |
+
else:
|
196 |
+
cost_per_1M_input_tokens = 0.8
|
197 |
+
elif use_case == "Generate":
|
198 |
+
if model == "Default":
|
199 |
+
cost_per_1M_input_tokens = 15
|
200 |
+
else:
|
201 |
+
cost_per_1M_input_tokens = 30
|
202 |
+
elif use_case == "Classify":
|
203 |
+
if model == "Default":
|
204 |
+
cost_per_1M_input_tokens = 200
|
205 |
+
else:
|
206 |
+
cost_per_1M_input_tokens = 200
|
207 |
+
else:
|
208 |
+
cost_per_1M_input_tokens = 15
|
209 |
+
|
210 |
+
cost_per_output_token = cost_per_1M_input_tokens * input_length / 1000000
|
211 |
+
|
212 |
+
return cost_per_output_token
|
213 |
+
|
214 |
class ModelPage:
|
215 |
|
216 |
def __init__(self, Models: BaseTCOModel):
|