skeleton

app.py

@@ -0,0 +1,35 @@
+# app.py
+import gradio as gr
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
+import torch
+
+MODEL_ID = "cahya/gpt2-small-indonesian-522M"  # change if needed
+
+# Load tokenizer + model (CPU)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype=torch.float32, device_map={"": "cpu"})
+
+generator = pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+    device=-1,  # CPU
+)
+
+def generate(prompt):
+    # prompt already instructs JSON-only output; but we keep generation stable
+    out = generator(prompt, max_new_tokens=180, do_sample=True, top_p=0.95, top_k=40, temperature=0.8)
+    text = out[0]["generated_text"]
+    return text
+
+# Simple gradio UI (the same function is exposed as API)
+demo = gr.Interface(
+    fn=generate,
+    inputs=gr.Textbox(lines=6, placeholder="Prompt..."),
+    outputs=gr.Textbox(),
+    title="TikTok Script Generator",
+    description="Generates short Indonesian scripts (JSON)."
+)
+
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)

requirement.txt

@@ -0,0 +1,3 @@
+gradio>=3.0
+transformers>=4.30.0
+torch   # use default - HF will pull a CPU build