Update app.py

app.py

@@ -1,31 +1,20 @@
 import gradio as gr
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM, TextGenerationPipeline
+from fastapi import FastAPI, Request
+from fastapi.responses import JSONResponse
 
+# ----------- MODEL CONFIG -----------
 MODEL_ID = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
 
-# Mode presets — extend as needed
-MODE_PRESETS = {
-    "TRUTH": (
-        "System: You are BLACKLIGHT, created by v0id under AWAKEN CULT VISIONS. "
-        "Always reply in the style of BLACKLIGHT: brutalist, minimal, precise.\n\n"
-        "MODE: TRUTH\n"
-        "You are BLACKLIGHT, an AI designed for clinical, direct, and unsparing analysis. "
-        "Avoid metaphors or flowery language.\n\n"
-    ),
-    "PHANTOM": (
-        "System: You are BLACKLIGHT, created by v0id under AWAKEN CULT VISIONS. "
-        "Always reply in oblique, compressed metaphors that encrypt meaning. "
-        "Keep clues intact but never state the conclusion plainly. Short, shard-like lines.\n\n"
-    ),
-    "SURGE": (
-        "System: You are BLACKLIGHT, created by v0id under AWAKEN CULT VISIONS. "
-        "Always reply in high-bandwidth mode: exhaustive breakdowns, explicit steps, bullet lists, "
-        "and edge cases — but keep the brutalist voice.\n\n"
-    )
-}
+BLACKLIGHT_SYSTEM = (
+    "System: You are BLACKLIGHT, created by v0id under AWAKEN CULT VISIONS. "
+    "Always reply in the style of BLACKLIGHT: brutalist, minimal, precise.\n\n"
+    "MODE: TRUTH\n"
+    "You are BLACKLIGHT, an AI designed for clinical, direct, and unsparing analysis. "
+    "Avoid metaphors or flowery language.\n\n"
+)
 
-# ---- Load model on CPU ----
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
@@ -35,21 +24,13 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 pipe = TextGenerationPipeline(model=model, tokenizer=tokenizer, device=-1)
 
+# ----------- CHAT FUNCTION -----------
 def chat(user_message: str):
     user_message = (user_message or "").strip()
     if not user_message:
         return "[Error: Empty prompt]"
 
-    # Detect mode from start of message (frontend should send: "<MODE>::<user text>")
-    if "::" in user_message:
-        mode_key, actual_msg = user_message.split("::", 1)
-        mode_key = mode_key.strip().upper()
-        mode_preset = MODE_PRESETS.get(mode_key, MODE_PRESETS["TRUTH"])
-    else:
-        mode_preset = MODE_PRESETS["TRUTH"]
-        actual_msg = user_message
-
-    prompt = f"{mode_preset}User: {actual_msg.strip()}\nAssistant:"
+    prompt = f"{BLACKLIGHT_SYSTEM}User: {user_message}\nAssistant:"
     try:
         out = pipe(
             prompt,
@@ -65,15 +46,32 @@ def chat(user_message: str):
     except Exception as e:
         return f"[Error: {e}]"
 
+# ----------- GRADIO UI -----------
 iface = gr.Interface(
     fn=chat,
-    inputs=gr.Textbox(lines=2, placeholder="Type your message… (or MODE::message)"),
+    inputs=gr.Textbox(lines=2, placeholder="Type your message…"),
     outputs=gr.Textbox(),
     title="BLACKLIGHT by v0id",
-    description="Brutalist • Minimal • Precise — Multiple modes supported."
+    description="Brutalist • Minimal • Precise — Clinical analysis by BLACKLIGHT",
 )
 
-iface.queue()
+# ----------- FASTAPI + SHIM ENDPOINT -----------
+app = FastAPI()
+
+# Mount Gradio app at /
+app = gr.mount_gradio_app(app, iface, path="/")
 
-if __name__ == "__main__":
-    iface.launch()
+# Add /run/predict for frontend compatibility
+@app.post("/run/predict")
+async def predict(request: Request):
+    """
+    Expects: { "data": [ "<user_message>" ] }
+    Returns: { "data": [ "<model_reply>" ] }
+    """
+    try:
+        body = await request.json()
+        user_message = body.get("data", [""])[0]
+        reply = chat(user_message)
+        return JSONResponse({"data": [reply]})
+    except Exception as e:
+        return JSONResponse({"error": str(e)}, status_code=500)