neural

neural

Character-level neural password generation (optional [neural] extra).

Syntax

 __          _______ _    _         
 \ \        / /  ____| |  | |        
  \ \  /\  / /| |__  | |__| |       
   \ \/  \/ / |  __| |  __  |       
    \  /\  /  | |    | |  | |       
     \/  \/   |_|    |_|  |_|       

  WordlistXPL-Forge  v1.2.0
  Author: André Henrique (@mrhenrike)
  Unified wordlist generation for pentest & red team

usage: wlf.py neural [-h] [--wordlist FILE [FILE ...]] [--model PATH]
                     [--epochs EPOCHS] [--batch-size BATCH_SIZE]
                     [--embed EMBED] [--hidden HIDDEN] [--layers LAYERS]
                     [--max-lines MAX_LINES] [--count COUNT]
                     [--temperature TEMPERATURE] [--prefix PREFIX]
                     [--mask MASK] [--max-len MAX_LEN] [--adapt FILE]
                     [--dpg-steps DPG_STEPS] [--adapter {passgpt,hf}]
                     [--no-dedupe] [--seed SEED] [-o OUTPUT]
                     [{train,generate}]

Character-level neural password generation (FLA/PassGPT style).
Requires the optional extra: pip install wordlistxpl-forge[neural].
Supports temperature sampling, guided generation with a prefix or
mask, Dynamic Password Guessing, and external PassGPT adapters.

Examples:
  wlf.py neural train --wordlist rockyou.txt --epochs 5
  wlf.py --limit 100000 neural generate --temperature 0.9 -o out.lst
  wlf.py --limit 5000 neural generate --prefix admin --mask '?u?l?l?l?d?d'
  wlf.py --limit 100000 neural generate --adapt cracked.txt
  wlf.py --limit 5000 neural generate --adapter passgpt --model javirandor/passgpt-10characters

positional arguments:
  {train,generate}      Action (default: generate)

options:
  -h, --help            show this help message and exit
  --wordlist FILE [FILE ...]
                        Training file(s) (train mode)
  --model PATH          Model path (.pt) or adapter model name
  --epochs EPOCHS       Training epochs
  --batch-size BATCH_SIZE
                        Training batch size
  --embed EMBED         Embedding dim
  --hidden HIDDEN       LSTM hidden size
  --layers LAYERS       LSTM layers
  --max-lines MAX_LINES
                        Max training lines (0 = all)
  --count COUNT         Candidates to generate (generate mode)
  --temperature TEMPERATURE
                        Sampling temperature (default: 1.0)
  --prefix PREFIX       Guided prefix
  --mask MASK           Guided hashcat-style mask
  --max-len MAX_LEN     Max candidate length (default: 32)
  --adapt FILE          Recovered passwords for Dynamic Password Guessing
  --dpg-steps DPG_STEPS
                        DPG adaptation steps (default: 200)
  --adapter {passgpt,hf}
                        Use an external HuggingFace causal LM adapter
  --no-dedupe           Allow duplicate candidates
  --seed SEED           RNG seed (0 = nondeterministic)
  -o, --output OUTPUT   Output file

Flags

Flag / arg	Default	Required	Description
neural_action	'generate'	no	Action (default: generate)
--wordlist	-	no	Training file(s) (train mode)
--model	'.model/neural_lstm.pt'	no	Model path (.pt) or adapter model name
--epochs	5	no	Training epochs
--batch-size	256	no	Training batch size
--embed	64	no	Embedding dim
--hidden	256	no	LSTM hidden size
--layers	2	no	LSTM layers
--max-lines	0	no	Max training lines (0 = all)
--count	10000	no	Candidates to generate (generate mode)
--temperature	1.0	no	Sampling temperature (default: 1.0)
--prefix	''	no	Guided prefix
--mask	''	no	Guided hashcat-style mask
--max-len	32	no	Max candidate length (default: 32)
--adapt	-	no	Recovered passwords for Dynamic Password Guessing
--dpg-steps	200	no	DPG adaptation steps (default: 200)
--adapter	-	no	Use an external HuggingFace causal LM adapter
--no-dedupe	False	no	Allow duplicate candidates
--seed	0	no	RNG seed (0 = nondeterministic)
-o, --output	-	no	Output file

Input combinations

• Minimum: wlf neural --help
• Typical: see examples in the syntax block.
• Globals: --min-len, --max-len, --limit, --threads when the command generates lists.

Output

Stdout and/or -o / --output when the parser exposes it. Exit 0 on success; non-zero on invalid input.

Smoke

• Valid: exit 0
• Invalid: exit 2

Notes

Examples use generic labels (BrandX, acme.example). This repo does not ship wlist_brasil.