Update app.py

app.py

@@ -1,7 +1,8 @@
 import spaces
 import gradio as gr
 
-from unsloth import FastLanguageModel
+from peft import AutoPeftModelForCausalLM
+from transformers import AutoTokenizer
 
 max_seq_length = 2048
 dtype = (
@@ -9,21 +10,12 @@ dtype = (
 )
 load_in_4bit = True
 
-model = None
-
-@spaces.GPU
-def load_model():
-    if model is None:
-        model, tokenizer = FastLanguageModel.from_pretrained(
-            model_name="ua-l/gemma-2-9b-legal-uk",
-            max_seq_length=max_seq_length,
-            dtype=dtype,
-            load_in_4bit=load_in_4bit,
-        )
-
-load_model()
+model = AutoPeftModelForCausalLM.from_pretrained(
+    "ua-l/gemma-2-9b-legal-uk",
+    load_in_4bit = load_in_4bit,
+)
+tokenizer = AutoTokenizer.from_pretrained("ua-l/gemma-2-9b-legal-uk")
 
-FastLanguageModel.for_inference(model)
 
 
 @spaces.GPU
@@ -35,7 +27,7 @@ def predict(question):
     ### Answer:
     '''], return_tensors = "pt").to("cuda")
 
-    outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)
+    outputs = model.generate(**inputs, max_new_tokens = 128)
     
     results = tokenizer.batch_decode(outputs, skip_special_tokens=True)
 
@@ -43,7 +35,7 @@ def predict(question):
 
 inputs = gr.Textbox(lines=2, label="Enter a question", value="Як отримати виплати ВПО?")
 
-outputs = gr.JSON(label="Answer")
+outputs = gr.Textbox(label="Answer")
 
 demo = gr.Interface(fn=predict, inputs=inputs, outputs=outputs)
 demo.launch()