Llama 3 8B Instruct AlienLM Full

This repository contains the Llama 3 8B Instruct AlienLM Full weights used in the AlienLM experiments. AlienLM is a client-side text obfuscation approach for black-box LLM APIs: it maps natural text into an alienized token space, adapts the model with AAT, and recovers text on the client side.

Links:

• Lab: DSBA Lab
• Paper: arXiv
• Official repo: KimJaehee0725/AlienLM

Model Table

Example

All happy families are alike; each unhappy family is unhappy in its own way.

One unhappyamilies                        
 hike..:              
 happy     
                                                                        happy hodin                                                                                                             waypoints,

[2460, 6380, 8689, 527, 27083, 26, 1855, 43251, 3070, 374, 43251, 304, 1202, 1866, 1648, 13]

[4054, 43251, 60004, 66417, 35331, 114100, 27381, 6380, 39185, 23136, 6380, 109132, 8299, 21649, 82386, 11]

Variant

• Variant: AlienLM full tokenizer-bijection adaptation
• Base model: Llama 3 8B Instruct
• Upload source: /data2/AlienLM/outputs/Llama3-8B-Instruct-AlienLM-50-all-tokenizer-v3-32-qwenv2/checkpoint-9306
• Tokenizer check: The local tokenizer produced different token IDs from the base tokenizer for the test sentence. Base tokenizer ids: [2460, 6380, 8689, 527, 27083, 26, 1855, 43251, 3070, 374, 43251, 304, 1202, 1866, 1648, 13]

Notes

• Served files only: weights, config, tokenizer, and README.
• Training checkpoints and optimizer artifacts are excluded.
• Intended for research evaluation, not production privacy guarantees.

BibTeX

@article{kim2026alienlm,
  title={AlienLM: Alienization of Language for API-Boundary Privacy in Black-Box LLMs},
  author={Kim, Jaehee and Kang, Pilsung},
  journal={arXiv preprint arXiv:2601.22710},
  year={2026}
}