Update app.py

Added the support for mistral codestral throught simple, inference client.
(Note : Still in development)

app.py

@@ -4,7 +4,7 @@ import os
 
 API_URL = {
     "Mistral" : "https://api-inference.huggingface.co/models/mistralai/Mixtral-8x7B-Instruct-v0.1",
-    "Codestral" : ""
+    "Codestral" : "mistralai/Codestral-22B-v0.1"
 }
 
 HF_TOKEN = os.environ['HF_TOKEN']
@@ -16,7 +16,7 @@ mistralClient = InferenceClient(
 )
 
 codestralClient = InferenceClient(
-    API_URL["Codestral"],
+    model = API_URL["Codestral"],
     headers = {"Authorization" : f"Bearer {HF_TOKEN}"},
 )
 
@@ -32,11 +32,13 @@ def format_prompt(message, history, enable_hinglish=False):
   prompt += f"[INST] {message} [/INST]"
   return prompt
 
-def generate(prompt, history, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0, enable_hinglish=False):
-    temperature = float(temperature)
+def generate(prompt, history, model = "Mistral", enable_hinglish=False, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0):
+    temperature = float(temperature) # Generation arguments
     if temperature < 1e-2:
         temperature = 1e-2
+        
     top_p = float(top_p)
+    
     generate_kwargs = dict(
         temperature=temperature,
         max_new_tokens=max_new_tokens,
@@ -45,6 +47,9 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=256, top_p=0.95, r
         do_sample=True,
         seed=42,
     )
+
+    # Selecting model to be used
+    client = mistralClient if(model == "Mistral") else codestralClient
     
     formatted_prompt = format_prompt(prompt, history, enable_hinglish)
     stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
@@ -55,6 +60,19 @@ def generate(prompt, history, temperature=0.9, max_new_tokens=256, top_p=0.95, r
     return output
 
 additional_inputs=[
+    gr.Dropdown(
+        choices = ["Mistral","Codestral"],
+        value = "Mistral",
+        label = "Model to be used",
+        interactive=True,
+        info = "Mistral for general-purpose chatting and codestral for code related task (Supports 80+ languages)"
+    ),
+    gr.Checkbox(
+        label="Hinglish",
+        value=False,
+        interactive=True,
+        info="Enables the MistralTalk to talk in Hinglish (Combination of Hindi and English)",
+    ),
     gr.Slider(
         label="Temperature",
         value=0.9,
@@ -91,12 +109,7 @@ additional_inputs=[
         interactive=True,
         info="Penalize repeated tokens",
     ),
-    gr.Checkbox(
-        label="Hinglish",
-        value=False,
-        interactive=True,
-        info="Enables the MistralTalk to talk in Hinglish (Combination of Hindi and English)",
-    )
+    
 ]
 
 css = """
@@ -115,7 +128,7 @@ with gr.Blocks(css=css) as demo:
         generate,
         additional_inputs=additional_inputs,
         theme = gr.themes.Soft(),
-        examples=[["What is the secret to life?"], ["How the universe works?"],["What can you do?"],["What is quantum mechanics?"],["Do you belive in after life?"]]
+        examples=[["What is the secret to life?"], ["How the universe works?"],["What can you do?"],["What is quantum mechanics?"],["Do you belive in after life?"], ["Java function to check if URL is valid or not."]]
     )
 
 demo.queue(max_size=100).launch(debug=True)