Create app.py

app.py

@@ -0,0 +1,45 @@
+import gradio as gr
+from llama_cpp import Llama
+from huggingface_hub import hf_hub_download
+REPO_ID = "i04n4/math-solver-llama3-Q8_0-GGUF"
+FILENAME = "model.gguf"
+
+model_path = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)
+
+print("🚀 Încarc motorul Llama...")
+llm = Llama(model_path=model_path, n_ctx=2048, n_threads=2)
+
+def generate_response(message, history):
+    system_prompt = """You are a specialized mathematical assistant. Your goal is to be precise and efficient.
+
+RULES:
+- RESPONSE TYPE A (Basic/Fact-based): If the question is simple arithmetic (e.g., 2+2) or a basic fact, provide ONLY the final result. Be as concise as possible.
+- RESPONSE TYPE B (Complex/Logic): If the question requires multiple steps, algebra, or reasoning, provide a clear 'Step-by-step reasoning' followed by 'Final Answer'.
+- FORMATTING: Never use internal scratchpad tags like <<...>>. Use plain text only. Do not use your inner thougths, only the explanation. Do not use any symbols that are not needed
+- LANGUAGE: Always answer in English."""
+    
+    full_prompt = f"<|start_header_id|>system<|end_header_id|>\n\n{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>\n\n{message}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
+    
+    response = ""
+    output = llm(
+        full_prompt, 
+        max_tokens=512, 
+        stop=["<|eot_id|>", "User:"], 
+        echo=False, 
+        stream=True
+    )
+    
+    for chunk in output:
+        delta = chunk['choices'][0]['text']
+        response += delta
+        yield response
+
+demo = gr.ChatInterface(
+    generate_response, 
+    title="Super Awesome Math Tutor",
+    description="Ask me a math question!!!!!!",
+    examples=["Calculate 25 * 14", "Solve 2x + 5 = 15", "Derivative of x^2"]
+)
+
+if __name__ == "__main__":
+    demo.launch()