made a module

password_entropy.py

@@ -0,0 +1,89 @@
+from datetime import timedelta
+from math import log2
+from transformers import GPT2LMHeadModel, RobertaTokenizerFast
+import torch
+import torch.nn.functional as F
+
+NUM_GENERATIONS = 1
+MIN_CHARS = 6
+MAX_CHARS = 10
+
+def tier(ttd: timedelta):
+    tiers = ["Pathetic", "Low", "Medium", "High", "Extreme"]
+    for i, tier in enumerate(tiers[:-1]):
+        threshold = timedelta(days=365*pow(10, i))
+
+        if ttd < threshold:
+            return tier
+
+    return tiers[-1]
+
+def calculate_password_entropy(predefined_password):
+    assert len(predefined_password) >= MIN_CHARS, "Too short password"
+    assert len(predefined_password) <= MAX_CHARS, "Too long password"
+
+    tokenizer = RobertaTokenizerFast.from_pretrained(
+        "javirandor/passgpt-10characters",
+        model_max_length=MAX_CHARS + 2,  # Max length + start and end tokens
+        padding="max_length",
+        truncation=True,
+        do_lower_case=False,
+        strip_accents=False,
+        mask_token="<mask>",
+        unk_token="<unk>",
+        pad_token="<pad>",
+        truncation_side="right",
+    )
+
+    predefined_tokens = tokenizer(predefined_password, return_tensors="pt")["input_ids"][0][1:]
+
+    model = GPT2LMHeadModel.from_pretrained("javirandor/passgpt-10characters").eval()
+
+    # Manually generate the sequence step-by-step based on predefined password
+    generated = torch.tensor([[tokenizer.bos_token_id]])
+    entropy = 0
+
+    for idx, token in enumerate(predefined_tokens):
+        outputs = model(generated)
+        next_token_logits = outputs.logits[:, -1, :]
+
+        # Calculate probabilities
+        probabilities = F.softmax(next_token_logits, dim=-1)
+
+        # Choose the next token from the predefined password
+        next_token = token.unsqueeze(0).unsqueeze(0)
+
+        # Print the probability of the chosen token
+        chosen_token_prob = probabilities[0, token].item()
+        chosen_token_str = tokenizer.convert_ids_to_tokens([token.item()])[0]
+
+        # Add to entropy
+        try:
+            entropy += -log2(chosen_token_prob)
+        except ValueError:
+            entropy += 10e10
+
+        # Append the chosen token to the generated sequence
+        generated = torch.cat((generated, next_token), dim=1)
+
+    return entropy
+
+def main():
+    # Input a predefined password from the user
+    predefined_password = input("Enter a password: ")
+    entropy = calculate_password_entropy(predefined_password)
+    print("Password entropy:", entropy)
+
+    ttd = timedelta(seconds=10e-9 * pow(2, entropy))
+    years = ttd.days // 365
+    days = ttd.days % 365
+    hours, remainder = divmod(ttd.seconds, 3600)
+    minutes, seconds = divmod(remainder, 60)
+
+    print("Time to decode with 1Gh/s:",
+          f"{years} years {days} days {hours} hours {minutes} minutes {seconds} seconds")
+
+    print("Tier:", tier(ttd))
+
+if __name__ == "__main__":
+    main()