Creating model and tokenizer and passing it to watermarker init

app.py

@@ -6,6 +6,7 @@ import numpy as np
 from watermark import Watermarker
 import time
 import gradio as gr
+from transformers import AutoModelForCausalLM
 
 device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
 
@@ -20,26 +21,28 @@ USERS = ['Alice', 'Bob', 'Charlie', 'Dan']
 EMBED_METHODS = [ 'aaronson', 'kirchenbauer', 'sampling', 'greedy' ]
 DETECT_METHODS = [ 'aaronson', 'aaronson_simplified', 'aaronson_neyman_pearson', 'kirchenbauer']
 PAYLOAD_BITS = 2
+device = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
 
-watermarker = Watermarker(modelname=args.model, window_size=window_size, payload_bits=PAYLOAD_BITS)
 DEFAULT_SYSTEM_PROMPT = """\
 You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe.  Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
 If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\
 """
+model = AutoModelForCausalLM.from_pretrained(args.model, use_auth_token=hf_token, torch_dtype=torch.float16,
+        device_map='auto').to(device)
+tokenizer = AutoTokenizer.from_pretrained(args.model, use_auth_token=hf_token)
 
 def embed(user, max_length, window_size, method, prompt):
     uid = USERS.index(user)
     
-    
+    watermarker = Watermarker(tokenizer=tokenizer, model=model, window_size=window_size, payload_bits=PAYLOAD_BITS)
     watermarked_texts = watermarker.embed(key=args.key, messages=[ uid ],
-                                          max_length=max_length, method=method, prompt=prompt)
+                                          max_length=max_length, method=method, prompt=prompt, window_size=window_size)
     print("watermarked_texts: ", watermarked_texts)
 
     return watermarked_texts[0]
     
 def detect(attacked_text, window_size, method, prompt):
-    watermarker = Watermarker(modelname=args.model,
-                              window_size=window_size, payload_bits=PAYLOAD_BITS)
+    watermarker = Watermarker(tokenizer=tokenizer, model=model, window_size=window_size, payload_bits=PAYLOAD_BITS)
 
     pvalues, messages = watermarker.detect([ attacked_text ], key=args.key, method=method, prompts=[prompt])
     print("messages: ", messages)

watermark.py

@@ -1,9 +1,6 @@
 import transformers
 from transformers import AutoTokenizer
-from transformers import (
-    AutoTokenizer,
-    AutoModelForCausalLM,
-)
+
 from transformers import pipeline, set_seed, LogitsProcessor
 from transformers.generation.logits_process import TopPLogitsWarper, TopKLogitsWarper
 import torch
@@ -90,10 +87,9 @@ class WatermarkingKirchenbauerLogitsProcessor(WatermarkingLogitsProcessor):
         return scores
 
 class Watermarker(object):
-    def __init__(self, modelname="facebook/opt-350m", window_size = 0, payload_bits = 0, logits_processor = None, *args, **kwargs):
-        self.tokenizer = AutoTokenizer.from_pretrained(modelname, use_auth_token=hf_token)
-        self.model = AutoModelForCausalLM.from_pretrained(modelname, use_auth_token=hf_token, torch_dtype=torch.float16,
-        device_map='auto').to(device)
+    def __init__(self, tokenizer=None, model=None, window_size = 0, payload_bits = 0, logits_processor = None, *args, **kwargs):
+        self.tokenizer = tokenizer
+        self.model = model
         self.model.eval()
         self.window_size = window_size