Update app.py

app.py

@@ -26,7 +26,7 @@ ADAPTER_ID = "Reubencf/gemma3-goan-finetuned"
 
 # Base model - MUST match what you used for fine-tuning!
 # Check your adapter's config.json for "base_model_name_or_path"
-BASE_MODEL_ID = "google/gemma-2b-it"  # Change this to your actual base model
+BASE_MODEL_ID = "google/gemma-3-4b-it"  # Change this to your actual base model
 # Common options:
 # - "google/gemma-2b-it" (2B parameters, easier on memory)
 # - "unsloth/gemma-2-2b-it-bnb-4bit" (quantized version)
@@ -181,7 +181,7 @@ def load_model_and_tokenizer():
             print("[Fallback] Trying with gemma-2b-it...")
             try:
                 base_model = AutoModelForCausalLM.from_pretrained(
-                    "google/gemma-2b-it",
+                    "google/gemma-3-4b-it",
                     token=HF_TOKEN,
                     trust_remote_code=True,
                     low_cpu_mem_usage=True,
@@ -190,7 +190,7 @@ def load_model_and_tokenizer():
                 ).to("cpu")
                 
                 tokenizer = AutoTokenizer.from_pretrained(
-                    "google/gemma-2b-it",
+                    "google/gemma-3-4b-it",
                     token=HF_TOKEN,
                     trust_remote_code=True,
                 )