Updated to avoid weird outputs

app.py

@@ -25,16 +25,18 @@ login(token=hf_token)
 
 # Models
 sentence_transformer_model = "all-MiniLM-L6-v2"
-repo_id = "bartowski/Llama-3.2-3B-Instruct-GGUF"  # Switched to 3B; revert to "bartowski/Llama-3.2-1B-Instruct-GGUF" if too heavy
-filename = "Llama-3.2-3B-Instruct-Q4_K_M.gguf"    # Use "Llama-3.2-1B-Instruct-Q4_K_M.gguf" for 1B
+repo_id = "bartowski/Llama-3.2-3B-Instruct-GGUF"
+filename = "Llama-3.2-3B-Instruct-Q4_K_M.gguf"
 
 # Define FAQs
 faqs = [
     {"question": "What is your name?", "answer": "My name is Tim Luka Horstmann."},
     {"question": "Where do you live?", "answer": "I live in Paris, France."},
-    {"question": "What is your education?", "answer": "I am currently pursuing a MSc in Data and AI at Institut Polytechnique de Paris. I also hold an MPhil in Advanced Computer Science from the University of Cambridge and a BSc in Business Informatics from RheinMain University of Applied Sciences."},
+    {"question": "What is your education?", "answer": "I am currenlty pursuing a MSc in Data and AI at Institut Polytechnique de Paris. I have an MPhil in Advanced Computer Science from the University of Cambridge, and a BSc in Business Informatics from RheinMain University of Applied Sciences."},
     {"question": "What are your skills?", "answer": "I am proficient in Python, Java, SQL, Cypher, SPARQL, VBA, JavaScript, HTML/CSS, and Ruby. I also use tools like PyTorch, Hugging Face, Scikit-Learn, NumPy, Pandas, Matplotlib, Jupyter, Git, Bash, IoT, Ansible, QuickSight, and Wordpress."},
-    # Add more from your CV
+    {"question": "How are you?", "answer": "I’m doing great, thanks for asking! I’m enjoying life in Paris and working on some exciting AI projects."},
+    {"question": "What do you do?", "answer": "I’m a Computer Scientist and AI enthusiast, currently pursuing a MSc in Data and AI at Institut Polytechnique de Paris and interning as a Machine Learning Research Engineer at Hi! PARIS."},
+    {"question": "How’s it going?", "answer": "Things are going well, thanks! I’m busy with my studies and research, but I love the challenges and opportunities I get to explore."},
 ]
 
 try:
@@ -105,25 +107,32 @@ def stream_response(query):
             idx = np.argmax(similarities)
             yield f"data: {faqs[idx]['answer']}\n\n"
             yield "data: [DONE]\n\n"
-        else:
-            context = retrieve_context(query)
-            prompt = (
-                f"<|im_start|>system\nYou are Tim Luka Horstmann, a Computer Scientist. Here is your CV:\n{context}\n"
-                f"A user is asking you a question about your CV. Respond as yourself, using the first person, and base your answer strictly on the information provided in the CV. Do not invent or assume any details not mentioned.\n<|im_end>\n"
-                f"<|im_start|>user\n{query}\n<|im_end>\n"
-                f"<|im_start|>assistant\n"
-            )
-            for chunk in generator(
-                prompt,
-                max_tokens=512,
-                stream=True,
-                stop=["<|im_end|>", "[DONE]"],
-                temperature=0.5,      # Lower for factual responses
-                top_p=0.9,
-                repeat_penalty=1.1,   # Reduce repetition/hallucination
-            ):
-                yield f"data: {chunk['choices'][0]['text']}\n\n"
-            yield "data: [DONE]\n\n"
+            return
+
+        context = retrieve_context(query)
+        prompt = (
+            f"<|im_start|>system\nYou are Tim Luka Horstmann, a Computer Scientist. Here is your CV:\n{context}\n"
+            f"A user is asking you a question. Respond as yourself, using the first person, in a friendly and concise manner. For questions about your CV, base your answer strictly on the provided CV information. For casual questions not covered by the CV, respond naturally but do not invent specific details beyond what’s generally true about you (e.g., your current location or field of work). Avoid meta-commentary or critiquing your own response.\n<|im_end>\n"
+            f"<|im_start|>user\n{query}\n<|im_end>\n"
+            f"<|im_start|>assistant\n"
+        )
+        
+        response_text = ""
+        for chunk in generator(
+            prompt,
+            max_tokens=200,
+            stream=True,
+            stop=["<|im_end|>", "[DONE]"],
+            temperature=0.5,     # Slightly higher for friendly tone
+            top_p=0.9,
+            repeat_penalty=1.2,  # Maintain control
+        ):
+            text = chunk['choices'][0]['text']
+            response_text += text
+            yield f"data: {text}\n\n"
+            if "<|im_end>" in response_text or "[DONE]" in response_text:
+                break
+        yield "data: [DONE]\n\n"
     except Exception as e:
         logger.error(f"Error in stream_response: {str(e)}")
         yield f"data: Error: {str(e)}\n\n"