Add DictaLM 2.0 instruct model 000

app.py

@@ -52,7 +52,7 @@ def predict():
     formatted_text = f"<s>[INST] {input_text} [/INST]"
 
     # Tokenize the input
-    inputs = tokenizer(formatted_text, return_tensors='pt', padding=True, truncation=True)
+    inputs = tokenizer(formatted_text, return_tensors='pt', padding=True, truncation=True, max_length=1024)
 
     # Generate the output
     outputs = model.generate(
@@ -68,10 +68,10 @@ def predict():
     # Decode the output
     prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).replace(formatted_text, '').strip()
     
-    # Remove user input part from the response
+    # Remove the user input part from the response
     if "[/INST]" in prediction:
         prediction = prediction.split("[/INST]", 1)[-1].strip()
-
+    
     return jsonify({"prediction": prediction})
 
 def run_flask():
@@ -97,6 +97,9 @@ def model_chat(query: Optional[str], history: Optional[History]) -> Tuple[Histor
     else:
         return history, "Error: Unable to get a response from the model."
 
+def respond(query: str, history: History) -> Tuple[History, str]:
+    history, response = model_chat(query, history)
+    return history, response  # Return history and response to show the model's response
 
 
 with gr.Blocks(css='''
@@ -161,14 +164,10 @@ with gr.Blocks(css='''
     chatbot = gr.Chatbot()
     query = gr.Textbox(placeholder="הכנס שאלה בעברית (או באנגלית!)", rtl=True)
     clear_btn = gr.Button("נקה שיחה")
-    
-    def respond(query: str, history: History) -> Tuple[History, str, History]:
-        history, response = model_chat(query, history)
-        return history, response, history  # Return history, response, and state
 
     demo_state = gr.State([])
 
     query.submit(respond, [query, demo_state], [chatbot, query, demo_state])
-    clear_btn.click(clear_session, [], [chatbot, demo_state])
+    clear_btn.click(clear_session, [], demo_state, chatbot)
 
 demo.queue(api_open=False).launch(max_threads=20, share=False, allowed_paths=['logo_am.png'])