merged adapter and base model for XL

humanize.py

@@ -26,23 +26,16 @@ batch_size = 64
 model_config = {
     "Base Model": "polygraf-ai/poly-humanizer-base",
     "Large Model": "polygraf-ai/poly-humanizer-large",
-    "XL Model": "polygraf-ai/poly-humanizer-XL-adapter-v2",
+    "XL Model": "polygraf-ai/poly-humanizer-XL-merged-v2",
 }
 
 # cache the base models, tokenizers, and adapters
 # initialize model and tokenizer
 models, tokenizers = {}, {}
 for name, path in model_config.items():
-    if name == "XL Model":
-        model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-xl", torch_dtype=torch.bfloat16).to(device)
-        model = PeftModel.from_pretrained(model, path, torch_dtype=torch.bfloat16, is_trainable=False)
-        model = model.merge_and_unload()
-        models[name] = model
-        tokenizers[name] = T5Tokenizer.from_pretrained("google/flan-t5-xl")
-    else:
-        model = T5ForConditionalGeneration.from_pretrained(path, torch_dtype=torch.bfloat16).to(device)
-        models[name] = model
-        tokenizers[name] = T5Tokenizer.from_pretrained(path)
+    model = T5ForConditionalGeneration.from_pretrained(path, torch_dtype=torch.bfloat16).to(device)
+    tokenizers[name] = T5Tokenizer.from_pretrained(path)
+    models[name] = model
     print(f"Loaded model: {name}, Num. params: {model.num_parameters()}")