Update app.py

app.py

@@ -2,10 +2,18 @@ from huggingface_hub import InferenceClient
 import gradio as gr
 import random
 import prompts
-client = InferenceClient(
-    "mistralai/Mixtral-8x7B-Instruct-v0.1"
-)
-
+clients = [
+    {'type':'image','name':'black-forest-labs/FLUX.1-dev','rank':'op','max_tokens':16384,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'deepseek-ai/DeepSeek-V2.5-1210','rank':'op','max_tokens':16384,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'Qwen/Qwen2.5-Coder-32B-Instruct','rank':'op','max_tokens':32768,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'meta-llama/Meta-Llama-3-8B','rank':'op','max_tokens':32768,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'Snowflake/snowflake-arctic-embed-l-v2.0','rank':'op','max_tokens':4096,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'Snowflake/snowflake-arctic-embed-m-v2.0','rank':'op','max_tokens':4096,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'HuggingFaceTB/SmolLM2-1.7B-Instruct','rank':'op','max_tokens':4096,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'Qwen/QwQ-32B-Preview','rank':'op','max_tokens':16384,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'meta-llama/Llama-3.3-70B-Instruct','rank':'pro','max_tokens':16384,'schema':{'bos':'<|im_start|>','eos':'<|im_end|>'}},
+    {'type':'text','name':'mistralai/Mixtral-8x7B-Instruct-v0.1','rank':'op','max_tokens':40000,'schema':{'bos':'<s>','eos':'</s>'}},
+]
 def format_prompt(message, history):
   prompt = "<s>"
   for user_prompt, bot_response in history:
@@ -27,10 +35,10 @@ agents =[
 
 ]
 def generate(
-        prompt, history, agent_name=agents[0], sys_prompt="", temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0,
+        prompt, history, mod, agent_name=agents[0], sys_prompt="", temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0,
 ):
     seed = random.randint(1,1111111111111111)
-
+    client=InferenceClient(clients[int(mod)]['name'])
     agent=prompts.WEB_DEV
     if agent_name == "WEB_DEV":
         agent = prompts.WEB_DEV_SYSTEM_PROMPT
@@ -78,6 +86,13 @@ def generate(
 
 
 additional_inputs=[
+    gr.Dropdown(
+        label="Model",
+        choices=[sn['name'] for sn in clients],
+        value=clients[2]['name'],
+        interactive=True,
+        type='index',
+        ),
     gr.Dropdown(
         label="Agents",
         choices=[s for s in agents],