Update app.py

app.py

@@ -1,20 +1,17 @@
 from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
-from peft import PeftModel  # <-- important
+from peft import PeftModel
 
-BASE = "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"           # your true base
-ADAPTER = "savan360/Qwen_prompt_creator"     # your LoRA repo (adapter-only)
+BASE = "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B"   # ← your base
+REPO = "savan360/Qwen_prompt_creator"              # your repo (root)
+SUBFOLDER = "adapter"                               # where adapter files live
 
 tok = AutoTokenizer.from_pretrained(BASE, trust_remote_code=True)
-if tok.pad_token is None and hasattr(tok, "eos_token"):
-    tok.pad_token = tok.eos_token
+if tok.pad_token is None and hasattr(tok, "eos_token"): tok.pad_token = tok.eos_token
 
-# (Optional) 4-bit to save VRAM
-bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype="float16")
+bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
+                         bnb_4bit_compute_dtype="float16", bnb_4bit_use_double_quant=True)
 
 base_model = AutoModelForCausalLM.from_pretrained(
-    BASE, trust_remote_code=True, quantization_config=bnb, device_map="auto"
+    BASE, trust_remote_code=True, device_map="auto", quantization_config=bnb, dtype="auto"
 )
-
-model = PeftModel.from_pretrained(base_model, ADAPTER)  # attach LoRA
-
-# ...your generation code...
+model = PeftModel.from_pretrained(base_model, REPO, subfolder=SUBFOLDER)