Update app.py

app.py

@@ -18,7 +18,7 @@ import threading
 
 # --- Login to Hugging Face using secret ---
 # Make sure HF_TOKEN is set in your Hugging Face Space > Settings > Repository secrets
-hf_token = os.environ.get("hugface") # Using "HF_TOKEN" is the standard on Spaces
+hf_token = os.environ.get("hugface")
 if not hf_token:
     raise ValueError("HF_TOKEN not found. Please set it in Hugging Face Space repository secrets.")
 login(token=hf_token)
@@ -26,7 +26,7 @@ print("Successfully logged into Hugging Face Hub!")
 
 # --- Configuration ---
 STT_MODEL_ID = "EYEDOL/SALAMA_C3"
-LLM_MODEL_ID = "google/gemma-1.1-2b-it"
+LLM_MODEL_ID = "meta-llama/Llama-3.2-1B-Instruct" # <-- FIX: Switched to Llama-3.2
 TTS_TOKENIZER_ID = "facebook/mms-tts-swh"
 TTS_ONNX_MODEL_PATH = "swahili_tts.onnx"
 
@@ -116,20 +116,12 @@ class WeeboAssistant:
         return output_path
 
     def get_llm_response(self, chat_history):
-        # <-- START OF FIX: Rebuild message history without a 'system' role -->
-        messages = []
+        # <-- FIX: Reverted to using a 'system' role, which is correct for Llama 3 -->
+        messages = [{'role': 'system', 'content': self.SYSTEM_PROMPT}]
         for user_msg, assistant_msg in chat_history:
-            # Add the user's message
             messages.append({"role": "user", "content": user_msg})
-            # Add the assistant's message if it exists
             if assistant_msg:
                 messages.append({"role": "assistant", "content": assistant_msg})
-
-        # Prepend the system prompt to the content of the very first user message.
-        # This is the correct way to use a system prompt with Gemma models.
-        if messages:
-            messages[0]["content"] = f"{self.SYSTEM_PROMPT}\n\n{messages[0]['content']}"
-        # <-- END OF FIX -->
         
         prompt = self.llm_pipeline.tokenizer.apply_chat_template(
             messages, tokenize=False, add_generation_prompt=True
@@ -229,52 +221,4 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Msaidizi wa Kiswahili") as demo:
                     gr.Markdown("### Utengenezaji wa Sauti (Speech Synthesis)")
                     tool_t2s_text_in = gr.Textbox(label="Maandishi ya Kuingiza (Input Text)", placeholder="Andika Kiswahili hapa...")
                     tool_t2s_audio_out = gr.Audio(type="filepath", label="Sauti Iliyotengenezwa (Synthesized Audio)", autoplay=False)
-                    tool_t2s_btn = gr.Button("Tengeneza Sauti (Synthesize)")
-
-    s2s_submit_btn.click(
-        fn=s2s_pipeline,
-        inputs=[s2s_audio_in, s2s_chatbot],
-        outputs=[s2s_chatbot, s2s_audio_out, s2s_text_out],
-        queue=True
-    ).then(
-        fn=lambda: gr.Audio(value=None),
-        inputs=None,
-        outputs=s2s_audio_in
-    )
-
-    t2t_submit_btn.click(
-        fn=t2t_pipeline,
-        inputs=[t2t_text_in, t2t_chatbot],
-        outputs=[t2t_chatbot],
-        queue=True
-    ).then(
-        fn=clear_textbox,
-        inputs=None,
-        outputs=t2t_text_in
-    )
-    
-    t2t_text_in.submit(
-        fn=t2t_pipeline,
-        inputs=[t2t_text_in, t2t_chatbot],
-        outputs=[t2t_chatbot],
-        queue=True
-    ).then(
-        fn=clear_textbox,
-        inputs=None,
-        outputs=t2t_text_in
-    )
-
-    tool_s2t_btn.click(
-        fn=assistant.transcribe_audio,
-        inputs=tool_s2t_audio_in,
-        outputs=tool_s2t_text_out,
-        queue=True
-    )
-    tool_t2s_btn.click(
-        fn=assistant.generate_speech,
-        inputs=tool_t2s_text_in,
-        outputs=tool_t2s_audio_out,
-        queue=True
-    )
-
-demo.queue().launch(debug=True)
+                    tool_t2s_btn = gr.Button("Tengeneza Sauti