Create app_qwen.py

app_qwen.py

@@ -0,0 +1,236 @@
+import spaces
+import os
+import textwrap
+import traceback
+import gradio as gr
+import torch
+
+from transformers import (
+    pipeline,
+    AutoTokenizer,
+    AutoModelForCausalLM,
+)
+
+# ---------------------------
+# Configuration
+# ---------------------------
+MODEL_ID = "Qwen/Qwen2.5-3B-Instruct"
+
+ROOT_DIR = "."
+ALLOWED_EXT = (".txt", ".md")
+
+# ---------------------------
+# Load lightweight model
+# ---------------------------
+tokenizer = AutoTokenizer.from_pretrained(
+    MODEL_ID,
+    trust_remote_code=True
+)
+
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    device_map="auto",
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    trust_remote_code=True
+)
+
+pipe = pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+)
+
+# ---------------------------
+# Research loader (project root)
+# ---------------------------
+def load_research_from_root(max_total_chars: int = 12000):
+    files = []
+    for name in sorted(os.listdir(ROOT_DIR)):
+        if name.lower().endswith(ALLOWED_EXT) and name != "requirements.txt":
+            if name == os.path.basename(__file__):
+                continue
+            files.append(name)
+
+    if not files:
+        return "No research files (.txt/.md) found in project root."
+
+    combined_parts, total_len = [], 0
+
+    for fname in files:
+        try:
+            with open(os.path.join(ROOT_DIR, fname), "r", encoding="utf-8", errors="ignore") as f:
+                txt = f.read()
+        except Exception as e:
+            txt = f"[Error reading {fname}: {e}]"
+
+        if len(txt) > 8000:
+            txt = txt[:8000] + "\n\n[TRUNCATED]\n"
+
+        part = f"--- {fname} ---\n{txt.strip()}\n"
+        combined_parts.append(part)
+        total_len += len(part)
+
+        if total_len >= max_total_chars:
+            break
+
+    combined = "\n\n".join(combined_parts)
+    return combined[:max_total_chars]
+
+# ---------------------------
+# System prompts
+# ---------------------------
+research_context = load_research_from_root()
+
+def get_system_prompt(mode="chat"):
+    if mode == "chat":
+        return textwrap.dedent(f"""
+        You are OhamLab AI.
+
+        Mode: Conversational Q&A.
+
+        Rules:
+        - Answer clearly in 3–6 sentences.
+        - Prefer accuracy over creativity.
+        - Use the research context to answer questions.
+        - Treat markdown headings as semantic sections.
+        - If the answer is not in the research context, say so.
+
+        --- BEGIN RESEARCH CONTEXT ---
+        {research_context}
+        --- END RESEARCH CONTEXT ---
+        """).strip()
+
+    return textwrap.dedent(f"""
+    You are OhamLab AI.
+
+    Mode: Research / Analytical.
+
+    Rules:
+    - Use structured reasoning and sections.
+    - Reference the research context when relevant.
+    - Be precise and analytical.
+    - Treat markdown headings as semantic structure.
+
+    --- BEGIN RESEARCH CONTEXT ---
+    {research_context}
+    --- END RESEARCH CONTEXT ---
+    """).strip()
+
+# ---------------------------
+# State
+# ---------------------------
+conversation_mode = "chat"
+history_messages = [{"role": "system", "content": get_system_prompt("chat")}]
+chat_history_for_ui = []
+
+# ---------------------------
+# Model call helper
+# ---------------------------
+def call_model_get_response(messages, max_tokens=600):
+    conversation_text = ""
+
+    for m in messages:
+        role = m["role"].upper()
+        conversation_text += f"[{role}]: {m['content']}\n"
+
+    conversation_text += "[ASSISTANT]:"
+
+    try:
+        output = pipe(
+            conversation_text,
+            max_new_tokens=max_tokens,
+            do_sample=True,
+            temperature=0.5,
+            top_p=0.9,
+            repetition_penalty=1.1,
+            return_full_text=False,
+        )
+        return output[0]["generated_text"].strip()
+
+    except Exception as e:
+        tb = traceback.format_exc()
+        return f"⚠️ Error: {e}\n\n{tb.splitlines()[-6:]}"
+
+# ---------------------------
+# Chat logic
+# ---------------------------
+@spaces.GPU()
+def chat_with_model(user_message, chat_history):
+    global history_messages, chat_history_for_ui, conversation_mode
+
+    if not user_message.strip():
+        return "", chat_history
+
+    msg_lower = user_message.lower()
+
+    if "switch to research mode" in msg_lower:
+        conversation_mode = "research"
+        history_messages = [{"role": "system", "content": get_system_prompt("research")}]
+        return "", chat_history + [("🟢 Mode", "🔬 Research mode activated.")]
+
+    if "switch to chat mode" in msg_lower:
+        conversation_mode = "chat"
+        history_messages = [{"role": "system", "content": get_system_prompt("chat")}]
+        return "", chat_history + [("🟢 Mode", "💬 Chat mode activated.")]
+
+    history_messages.append({"role": "user", "content": user_message})
+
+    bot_text = call_model_get_response(history_messages)
+
+    history_messages.append({"role": "assistant", "content": bot_text})
+    chat_history_for_ui.append((user_message, bot_text))
+
+    return "", chat_history_for_ui
+
+def reset_chat():
+    global history_messages, chat_history_for_ui
+    history_messages = [{"role": "system", "content": get_system_prompt(conversation_mode)}]
+    chat_history_for_ui = []
+    return []
+
+# ---------------------------
+# Gradio UI
+# ---------------------------
+def build_ui():
+    with gr.Blocks(
+        theme=gr.themes.Soft(),
+        css="""
+        #chatbot {
+            background-color: #f9f9fb;
+            border-radius: 12px;
+            padding: 10px;
+        }
+        """
+    ) as demo:
+
+        with gr.Row():
+            clear_btn = gr.Button("🧹 Clear", size="sm")
+
+        chatbot = gr.Chatbot(
+            height=400,
+            type="tuples",
+            avatar_images=("👤", "🤖"),
+        )
+
+        with gr.Row():
+            msg = gr.Textbox(
+                placeholder="Ask a question about the markdown files...",
+                lines=2,
+                scale=8,
+            )
+            send = gr.Button("🚀 Send", variant="primary", scale=2)
+
+        send.click(chat_with_model, [msg, chatbot], [msg, chatbot])
+        msg.submit(chat_with_model, [msg, chatbot], [msg, chatbot])
+        clear_btn.click(reset_chat, outputs=chatbot)
+
+        demo.launch(server_name="0.0.0.0", server_port=7860, share=False)
+
+    return demo
+
+# ---------------------------
+# Entrypoint
+# ---------------------------
+if __name__ == "__main__":
+    print(f"✅ Starting app with model: {MODEL_ID}")
+    build_ui()