Fix bitsandbytes issue - attempt 3

Dockerfile

@@ -3,7 +3,6 @@ FROM python:3.9
 
 # Set environment variables for cache
 ENV HF_HOME=/tmp/hf_home
-ENV TRANSFORMERS_CACHE=/tmp/hf_home
 
 # Set the working directory
 WORKDIR /app

app.py

@@ -11,13 +11,13 @@ from transformers import (
 # Load the configuration and remove any quantization config if present
 config = AutoConfig.from_pretrained("devops-bda/Abap")
 if hasattr(config, "quantization_config"):
-    config.quantization_config = None
+    del config.quantization_config  # Safely delete it instead of setting to None
 
 # Load the model and tokenizer without 4-bit quantization
 model = AutoModelForCausalLM.from_pretrained(
     "devops-bda/Abap",
     config=config,
-    load_in_4bit=False  # explicitly disable 4-bit quantization
+    device_map="auto"  # This ensures the model loads properly on CPU
 )
 tokenizer = AutoTokenizer.from_pretrained("devops-bda/Abap")