Gradio UI: Colab-Ollama-Proxy mit Streaming

app.py

@@ -0,0 +1,229 @@
+import gradio as gr
+import requests
+import json
+import time
+import re
+
+# Default Ollama URL (Colab tunnel - changes on restart)
+DEFAULT_OLLAMA_URL = "https://fairfield-configure-replies-paths.trycloudflare.com"
+DEFAULT_MODEL = "qwen-coder-boosted"
+
+def get_ollama_url(base_url):
+    """Get the Ollama API URL from the base URL."""
+    base = base_url.rstrip("/")
+    # If it's a tunnel URL, Ollama is on port 11434
+    if "trycloudflare.com" in base:
+        return base.replace("trycloudflare.com", "trycloudflare.com") + "/api"
+    return base + "/api"
+
+def list_models(ollama_url):
+    """List available models from Ollama."""
+    try:
+        api_url = ollama_url.rstrip("/")
+        if "/api" not in api_url:
+            api_url += "/api"
+        resp = requests.get(f"{api_url}/tags", timeout=5)
+        if resp.status_code == 200:
+            models = resp.json().get("models", [])
+            return [m["name"] for m in models]
+        return [DEFAULT_MODEL]
+    except:
+        return [DEFAULT_MODEL]
+
+def chat_stream(message, history, ollama_url, model_name, system_prompt, temperature):
+    """Stream chat response from Ollama."""
+    api_url = ollama_url.rstrip("/")
+    if "/api" not in api_url:
+        api_url += "/api"
+
+    # Build messages array
+    messages = []
+    if system_prompt:
+        messages.append({"role": "system", "content": system_prompt})
+
+    for user_msg, assistant_msg in history:
+        if user_msg:
+            messages.append({"role": "user", "content": user_msg})
+        if assistant_msg:
+            messages.append({"role": "assistant", "content": assistant_msg})
+
+    messages.append({"role": "user", "content": message})
+
+    payload = {
+        "model": model_name,
+        "messages": messages,
+        "stream": True,
+        "options": {
+            "temperature": temperature,
+            "num_predict": 32768
+        }
+    }
+
+    try:
+        resp = requests.post(f"{api_url}/chat", json=payload, stream=True, timeout=120)
+        resp.raise_for_status()
+
+        full_response = ""
+        for line in resp.iter_lines():
+            if line:
+                try:
+                    chunk = json.loads(line)
+                    if "message" in chunk and "content" in chunk["message"]:
+                        content = chunk["message"]["content"]
+                        full_response += content
+                        yield full_response
+                    if chunk.get("done", False):
+                        break
+                except json.JSONDecodeError:
+                    continue
+    except Exception as e:
+        yield f"❌ Fehler: {str(e)}\n\n**Tipp:** Prüfe die Ollama-URL oben – der Tunnel ändert sich bei jedem Colab-Neustart!"
+
+def check_connection(ollama_url):
+    """Check if Ollama is reachable."""
+    api_url = ollama_url.rstrip("/")
+    if "/api" not in api_url:
+        api_url += "/api"
+    try:
+        resp = requests.get(f"{api_url}/tags", timeout=5)
+        if resp.status_code == 200:
+            models = resp.json().get("models", [])
+            model_list = "\n".join([f"  • `{m['name']}` ({m.get('size', 0)/1e9:.1f}GB)" for m in models])
+            return f"✅ **Verbunden!** Modelle:\n{model_list}"
+        return f"⚠️ API antwortet mit Status {resp.status_code}"
+    except Exception as e:
+        return f"❌ **Keine Verbindung:** {str(e)}\n\nStarte den Colab-Tunnel neu und aktualisiere die URL oben."
+
+# Build the Gradio UI
+with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="indigo"), title="Qwen Coder Boosted") as demo:
+    gr.Markdown("""
+    # 🚀 Qwen Coder Boosted
+    **Verbunden mit Colab-Ollama-Backend** – Zero GPU auf HF, Rechenpower auf Colab L4
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            ollama_url = gr.Textbox(
+                label="🌐 Ollama API URL",
+                value=DEFAULT_OLLAMA_URL,
+                info="Dein Colab-Tunnel oder localhost:11434",
+                scale=3
+            )
+            check_btn = gr.Button("🔌 Verbindung prüfen", variant="secondary", size="sm")
+            connection_status = gr.Markdown("💡 Gib die URL ein und klicke auf 'Verbindung prüfen'")
+
+        with gr.Column(scale=1):
+            model_name = gr.Dropdown(
+                label="🤖 Modell",
+                choices=[DEFAULT_MODEL],
+                value=DEFAULT_MODEL,
+                interactive=True,
+                scale=2
+            )
+            refresh_models = gr.Button("🔄 Modelle aktualisieren", variant="secondary", size="sm")
+
+    with gr.Accordion("⚙️ Erweiterte Einstellungen", open=False):
+        with gr.Row():
+            system_prompt = gr.Textbox(
+                label="System Prompt",
+                value="You are Qwen Coder Boosted, an expert coding assistant. Generate clean, working code. Always show complete files with [FILE:filename] markers.",
+                lines=3,
+                scale=3
+            )
+            temperature = gr.Slider(
+                label="Temperature",
+                minimum=0.1,
+                maximum=1.5,
+                value=0.6,
+                step=0.05,
+                scale=1
+            )
+
+    chatbot = gr.Chatbot(
+        label="💬 Chat",
+        height=500,
+        bubble_full_width=False,
+        show_copy_button=True
+    )
+
+    with gr.Row():
+        msg = gr.Textbox(
+            label="Deine Nachricht",
+            placeholder="Schreib einen Prompt... z.B. 'Baue eine To-Do-App mit HTML/CSS/JS'",
+            scale=8,
+            container=True
+        )
+        submit = gr.Button("📤 Senden", variant="primary", scale=1)
+        clear = gr.Button("🗑️ Leeren", variant="secondary", scale=1)
+
+    gr.Markdown("""
+    ---
+    ### 💡 Tipps
+    - **Tunnel-URL ändert sich** bei jedem Colab-Neustart – aktualisiere sie oben
+    - **Modelle:** `qwen-coder-boosted` (Custom), `qwen3-coder:30b-a3b-q4_K_M` (Base)
+    - **Max Tokens:** 32K (hardcodiert)
+    - **Temperature:** 0.6 (Standard für Coding)
+    """)
+
+    # Event handlers
+    def update_models(url):
+        api_url = url.rstrip("/")
+        if "/api" not in api_url:
+            api_url += "/api"
+        try:
+            resp = requests.get(f"{api_url}/tags", timeout=5)
+            if resp.status_code == 200:
+                models = [m["name"] for m in resp.json().get("models", [])]
+                return gr.Dropdown(choices=models, value=models[0] if models else DEFAULT_MODEL)
+        except:
+            pass
+        return gr.Dropdown(choices=[DEFAULT_MODEL], value=DEFAULT_MODEL)
+
+    refresh_models.click(
+        fn=update_models,
+        inputs=[ollama_url],
+        outputs=[model_name]
+    )
+
+    check_btn.click(
+        fn=check_connection,
+        inputs=[ollama_url],
+        outputs=[connection_status]
+    )
+
+    def respond(message, chat_history, url, model, sys_prompt, temp):
+        if not message.strip():
+            yield chat_history
+            return
+        chat_history.append((message, ""))
+        full_response = ""
+        for partial in chat_stream(message, chat_history[:-1], url, model, sys_prompt, temp):
+            full_response = partial
+            chat_history[-1] = (message, full_response)
+            yield chat_history
+
+    submit.click(
+        fn=respond,
+        inputs=[msg, chatbot, ollama_url, model_name, system_prompt, temperature],
+        outputs=[chatbot]
+    ).then(
+        fn=lambda: "",
+        outputs=[msg]
+    )
+
+    msg.submit(
+        fn=respond,
+        inputs=[msg, chatbot, ollama_url, model_name, system_prompt, temperature],
+        outputs=[chatbot]
+    ).then(
+        fn=lambda: "",
+        outputs=[msg]
+    )
+
+    clear.click(
+        fn=lambda: None,
+        outputs=[chatbot]
+    )
+
+if __name__ == "__main__":
+    demo.launch()