Update app.py

app.py

@@ -1,59 +1,30 @@
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig
+import moondream as md
 from PIL import Image
-import torch
 
-model_id = "vikhyatk/moondream2"
-revision = "2024-08-26"
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
-# 1. Load Tokenizer
-tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
-
-# 2. Load Config dengan trust_remote_code
-config = AutoConfig.from_pretrained(model_id, trust_remote_code=True, revision=revision)
-
-# 3. INTERVENSI KONFIGURASI (Penting!)
-if hasattr(config, "text_config"):
-    # Fix untuk pad_token_id
-    config.text_config.pad_token_id = tokenizer.eos_token_id
-    
-    # Fix untuk KeyError: 'type' pada rope_scaling
-    # Kita paksa formatnya agar sesuai dengan apa yang diminta modeling_phi.py
-    if hasattr(config.text_config, "rope_scaling") and config.text_config.rope_scaling is not None:
-        if isinstance(config.text_config.rope_scaling, dict) and "type" not in config.text_config.rope_scaling:
-            # Jika ada tapi tidak ada key 'type', kita beri default 'linear' atau hapus
-            config.text_config.rope_scaling = None 
-    else:
-        config.text_config.rope_scaling = None
-
-# 4. Load Model
-model = AutoModelForCausalLM.from_pretrained(
-    model_id, 
-    config=config,
-    trust_remote_code=True, 
-    revision=revision,
-    torch_dtype=torch.float16 if device == "cuda" else torch.float32
-).to(device)
+# Load model secara otomatis
+# Library ini akan menangani download dan konfigurasi sendiri
+model = md.vl(model="vikhyatk/moondream2")
 
 def answer_question(image, question):
     if image is None:
         return "No image provided"
     
+    # Konversi ke PIL Image jika perlu
     image = image.convert("RGB")
     
-    with torch.no_grad():
-        enc_image = model.encode_image(image)
-        answer = model.answer_question(enc_image, question, tokenizer)
+    # Proses encoding dan jawaban dalam satu baris simpel
+    encoded_image = model.encode_image(image)
+    answer = model.answer_question(encoded_image, question)
     
     return answer
 
 # Interface Gradio
 interface = gr.Interface(
     fn=answer_question,
-    inputs=[gr.Image(type="pil"), gr.Textbox(label="Question")],
+    inputs=[gr.Image(type="pil"), gr.Textbox(label="Question", value="What is the result of the math expression?")],
     outputs=gr.Text(label="Answer"),
-    title="Moondream Captcha Solver"
+    title="Moondream Solver (Stable Version)"
 )
 
 if __name__ == "__main__":