---
library_name: transformers
pipeline_tag: text-generation
tags:
- alienlm
- alien-adaptation-training
- tokenizer-bijection
- instruction-tuned
datasets:
- Magpie-Align/Magpie-Pro-300K-Filtered
- Magpie-Align/Magpie-Reasoning-V1-150K
model-index:
- name: Gemma 2 9B IT Random Bijection
  results: []
license: gemma
base_model: google/gemma-2-9b-it
---

# Gemma 2 9B IT Random Bijection

This repository contains the `Gemma 2 9B IT Random Bijection` 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](https://huggingface.co/dsba-lab)
- Paper: [arXiv](https://arxiv.org/abs/2601.22710)
- Official repo: [KimJaehee0725/AlienLM](https://github.com/KimJaehee0725/AlienLM)

## Model Table

| Uploaded model | Base model | Description | HF Models |
|---|---|---|---|
| Gemma 2 9B IT AlienLM Full | Gemma 2 9B IT | Full AlienLM adaptation | [dsba-lab/gemma2-9b-it-alienlm-full](https://huggingface.co/dsba-lab/gemma2-9b-it-alienlm-full) |
| Gemma 2 9B IT Random Bijection | Gemma 2 9B IT | Random bijection baseline | [dsba-lab/gemma2-9b-it-random-bijection](https://huggingface.co/dsba-lab/gemma2-9b-it-random-bijection) |
| Llama 3 8B Instruct AlienLM Full | Llama 3 8B Instruct | Full AlienLM adaptation | [dsba-lab/llama3-8b-instruct-alienlm-full](https://huggingface.co/dsba-lab/llama3-8b-instruct-alienlm-full) |
| Llama 3 8B Instruct AlienLM Ratio 20 | Llama 3 8B Instruct | Partial alienization ratio 20 | [dsba-lab/llama3-8b-instruct-alienlm-ratio-20](https://huggingface.co/dsba-lab/llama3-8b-instruct-alienlm-ratio-20) |
| Llama 3 8B Instruct AlienLM Ratio 40 | Llama 3 8B Instruct | Partial alienization ratio 40 | [dsba-lab/llama3-8b-instruct-alienlm-ratio-40](https://huggingface.co/dsba-lab/llama3-8b-instruct-alienlm-ratio-40) |
| Llama 3 8B Instruct AlienLM Ratio 60 | Llama 3 8B Instruct | Partial alienization ratio 60 | [dsba-lab/llama3-8b-instruct-alienlm-ratio-60](https://huggingface.co/dsba-lab/llama3-8b-instruct-alienlm-ratio-60) |
| Llama 3 8B Instruct AlienLM Ratio 80 | Llama 3 8B Instruct | Partial alienization ratio 80 | [dsba-lab/llama3-8b-instruct-alienlm-ratio-80](https://huggingface.co/dsba-lab/llama3-8b-instruct-alienlm-ratio-80) |
| Llama 3 8B Instruct Random Bijection | Llama 3 8B Instruct | Random bijection baseline | [dsba-lab/llama3-8b-instruct-random-bijection](https://huggingface.co/dsba-lab/llama3-8b-instruct-random-bijection) |
| Qwen 2.5 14B Instruct AlienLM Full | Qwen2.5 14B Instruct | Full AlienLM adaptation | [dsba-lab/qwen25-14b-instruct-alienlm-full](https://huggingface.co/dsba-lab/qwen25-14b-instruct-alienlm-full) |
| Qwen 2.5 14B Instruct Random Bijection | Qwen2.5 14B Instruct | Random bijection baseline | [dsba-lab/qwen25-14b-instruct-random-bijection](https://huggingface.co/dsba-lab/qwen25-14b-instruct-random-bijection) |
| Qwen 2.5 7B Instruct AlienLM Full | Qwen2.5 7B Instruct | Full AlienLM adaptation | [dsba-lab/qwen25-7b-instruct-alienlm-full](https://huggingface.co/dsba-lab/qwen25-7b-instruct-alienlm-full) |
| Qwen 2.5 7B Instruct Random Bijection | Qwen2.5 7B Instruct | Random bijection baseline | [dsba-lab/qwen25-7b-instruct-random-bijection](https://huggingface.co/dsba-lab/qwen25-7b-instruct-random-bijection) |

## Example


<table>
  <tr>
    <th>Natural text</th>
    <th>Alien text</th>
  </tr>
  <tr>
    <td><pre>All happy families are alike; each unhappy family is unhappy in its own way.</pre></td>
    <td><pre> Dhaka בגCLS patriot Dude ブラウンanova 교neti estufa 교ಟ FestivalsDocumentation bekanntenroquia</pre></td>
  </tr>
  <tr>
    <th>Original token IDs</th>
    <th>Alien token IDs</th>
  </tr>
  <tr>
    <td><pre>[2430, 4915, 9160, 708, 28368, 235289, 1853, 42056, 2730, 603, 42056, 575, 1277, 1997, 1703, 235265]</pre></td>
    <td><pre>[118082, 85241, 174135, 184646, 114599, 58746, 48064, 71689, 147487, 81724, 71689, 163116, 23867, 77693, 75944, 217666]</pre></td>
  </tr>
</table>


## Variant

- Variant: Random bijection baseline, seed 42
- Base model: `Gemma 2 9B IT`
- Upload source: `/data2/AlienLM/outputs/Gemma2-9b-it-random42`
- Tokenizer check: The local tokenizer produced different token IDs from the base tokenizer for the test sentence. Base tokenizer ids: [2430, 4915, 9160, 708, 28368, 235289, 1853, 42056, 2730, 603, 42056, 575, 1277, 1997, 1703, 235265]

## 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

```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}
}
```