Update app.py

app.py

@@ -1,18 +1,21 @@
-from transformers import AutoTokenizer, AutoModelForCausalLM
-import torch
+from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
 
 MODEL_ID = "savan360/Qwen_prompt_creator"
-SUBFOLDER = "merged"  # or "adapter" if you’re using the LoRA route
 
-tokenizer = AutoTokenizer.from_pretrained(
-    MODEL_ID, subfolder=SUBFOLDER, trust_remote_code=True
-)
+cfg = AutoConfig.from_pretrained(MODEL_ID, trust_remote_code=True)
+
+# ---- PATCH: remove/normalize bad quantization_config ----
+if hasattr(cfg, "quantization_config"):
+    qc = getattr(cfg, "quantization_config")
+    if qc is None:
+        # remove the attribute entirely; from_pretrained doesn't need it if you're not quantizing via HF APIs
+        delattr(cfg, "quantization_config")
 
+tok = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
-    subfolder=SUBFOLDER,
-    dtype=torch.float16,            # <- replaces deprecated torch_dtype
-    device_map="auto",
+    config=cfg,
     trust_remote_code=True,
-    quantization_config=None        # <- THIS LINE disables auto 4-bit path
+    torch_dtype="auto",
+    device_map="auto",
 )