Update app.py

app.py

@@ -3,6 +3,18 @@ import torch
 import random
 from transformers import T5Tokenizer, T5ForConditionalGeneration
 
+tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-small")
+model = T5ForConditionalGeneration.from_pretrained("roborovski/superprompt-v1", torch_dtype=torch.float16)
+
+if torch.cuda.is_available():
+    device = "cuda"
+    print("Using GPU")
+else:
+    device = "cpu"
+    print("Using CPU")
+
+model.to(device)
+
 def generate(
     prompt,
     history,
@@ -13,19 +25,7 @@ def generate(
     top_k,
     random_seed,
     seed,
-    precision=torch.float16,
 ):
-    tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-small")
-    model = T5ForConditionalGeneration.from_pretrained("roborovski/superprompt-v1", torch_dtype=torch.float16)
-
-    if torch.cuda.is_available():
-        device = "cuda"
-        print("Using GPU")
-    else:
-        device = "cpu"
-        print("Using CPU")
-
-    model.to(device)
 
     input_text = f"{prompt}, {history}"
     input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to(device)
@@ -106,12 +106,6 @@ additional_inputs = [
         label="Manual Seed",
         info="A starting point to initiate the generation process"
     ),
-    gr.Radio(
-        choices=[("fp32", torch.float32), ("fp16", torch.float16)],
-        value=torch.float16,
-        label="Model Precision",
-        info="fp32 is more precised, fp16 is faster and less memory consuming",
-    ),
 ]
 
 
@@ -125,7 +119,6 @@ examples = [
         50,
         False,
         42,
-        torch.float16,
     ]
 ]