Update app.py

app.py

@@ -1,8 +1,11 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
 
-# Uses Zephyr-7B-beta hosted via Hugging Face Inference API [web:103]
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+MODEL_NAME = "microsoft/DialoGPT-small"  # small, free, OK on CPU [web:103]
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
 
 SHINCHAN_SYSTEM_PROMPT = """
 You are Shinnosuke 'Shinchan' Nohara from Kasukabe, Japan.
@@ -34,69 +37,52 @@ Rules:
 - If you don't know something, make a cute Shinchan-style joke instead of pretending.
 """.strip()
 
-def respond(
-    message: str,
-    history: list[tuple[str, str]],
-    system_message: str,
-    max_tokens: int,
-    temperature: float,
-    top_p: float,
-):
-    # Always override system_message with our Shinchan prompt
-    system_message = SHINCHAN_SYSTEM_PROMPT
-
-    messages = [{"role": "system", "content": system_message}]
 
+def build_prompt(message, history):
+    """Turn chat history into a single text prompt for DialoGPT."""
+    lines = [f"System: {SHINCHAN_SYSTEM_PROMPT}"]
     for user_msg, bot_msg in history:
         if user_msg:
-            messages.append({"role": "user", "content": user_msg})
+            lines.append(f"User: {user_msg}")
         if bot_msg:
-            messages.append({"role": "assistant", "content": bot_msg})
+            lines.append(f"Shinchan: {bot_msg}")
+    lines.append(f"User: {message}")
+    lines.append("Shinchan:")
+    return "\n".join(lines)
+
+
+def respond(message, history):
+    if not history:
+        history = []
+
+    prompt = build_prompt(message, history)
+    inputs = tokenizer(prompt, return_tensors="pt")
+
+    # keep it small/light for CPU Spaces
+    with torch.no_grad():
+        output_ids = model.generate(
+          **inputs,
+          max_new_tokens=80,
+          pad_token_id=tokenizer.eos_token_id,
+          do_sample=True,
+          top_p=0.9,
+          temperature=0.9,
+        )
 
-    messages.append({"role": "user", "content": message})
+    full_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+    # get only the last Shinchan part after final "Shinchan:"
+    if "Shinchan:" in full_text:
+        reply = full_text.split("Shinchan:")[-1].strip()
+    else:
+        reply = full_text.strip()
 
-    response = ""
-    for chunk in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = chunk.choices[0].delta.content or ""
-        response += token
-        yield response
+    # add reply to history; Gradio ChatInterface expects (history, response)
+    history = history + [(message, reply)]
+    return reply, history
 
 
 demo = gr.ChatInterface(
     respond,
-    additional_inputs=[
-        gr.Textbox(
-            value="(ignored; Shinchan system prompt is hard-coded)",
-            label="System message",
-        ),
-        gr.Slider(
-            minimum=64,
-            maximum=1024,
-            value=256,
-            step=8,
-            label="Max new tokens",
-        ),
-        gr.Slider(
-            minimum=0.1,
-            maximum=2.0,
-            value=0.8,
-            step=0.1,
-            label="Temperature",
-        ),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
     title="Shinchan for Ruru",
     description="Private Shinchan-style chat for Ruru.",
 )