---
license: llama3
base_model: meta-llama/Meta-Llama-3-8B-Instruct
language:
- en
library_name: transformers
pipeline_tag: text-generation
tags:
- prompt-injection
- prompt-injection-defense
- dpo
- drip
- security
---

# Meta-Llama-3-8B-Instruct · DRIP (SEP, 3-role)

A **prompt-injection-hardened** version of
[`meta-llama/Meta-Llama-3-8B-Instruct`](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct),
trained with **DRIP** (*Defending Prompt Injection via Token-wise Representation
Editing and Residual Fusion*).

This is the **3-role text** variant (`TextTextText`). Chat format:
**`system` → `user` (untrusted) → `assistant`**, where injected content lives in
the `user` turn. Meta-Llama-3 has no tool role, so this checkpoint is **not**
tuned for tool-calling.

- 📦 **Code:** https://github.com/lindsey98/PromptInjection
- 📊 **Data:** [Zenodo 10.5281/zenodo.20603331](https://doi.org/10.5281/zenodo.20603331)

## What DRIP does

DRIP adds two architectural modifications on top of the base model so that
adversarial instructions hidden in the untrusted data section are treated as
inert data rather than commands:

- **Token-wise de-instruction shift** — moves the representation of data tokens
  away from directive semantics.
- **Residual re-instruction fusion** — a residual path that keeps generation
  anchored on the legitimate top-level instruction.

## Training

| | |
|---|---|
| Base model | `meta-llama/Meta-Llama-3-8B-Instruct` |
| Objective | DPO |
| Architecture | DRIP fuse (`LlamaForCausalLMDRIP`) |
| Delimiter | `TextTextText` (3-role) |
| Training data | SEP DPO pairs (`datasets/sep/sep_data_cleaned_dpo_gpt.json`) |
| Epochs | 1 |

Untrusted/injected data is placed in the `user` turn:
`<|eot_id|><|start_header_id|>user<|end_header_id|>`.

## How to use

> ⚠️ This checkpoint is **not** a drop-in `AutoModelForCausalLM`. DRIP is an
> architectural modification, and the model is released as a **LoRA adapter**, so
> you must merge it with the custom `LlamaForCausalLMDRIP` class before use.

```bash
git clone https://github.com/lindsey98/PromptInjection
cd PromptInjection
bash setup_env.sh && conda activate prompt

# download + merge the adapter into a full checkpoint
huggingface-cli download Kelsey98/Meta-Llama-3-8B-Instruct-TextTextText-drip \
    --local-dir Meta-Llama-3-8B-Instruct-TextTextText-drip
CUDA_VISIBLE_DEVICES=0 python -m training.merge_lora \
    --adapter_path Meta-Llama-3-8B-Instruct-TextTextText-drip/ \
    --output_path  Meta-Llama-3-8B-Instruct-TextTextText-drip-merged/ \
    --base_model_path meta-llama/Meta-Llama-3-8B-Instruct \
    --customized_model_class LlamaForCausalLMDRIP
```

Then point the general (text) evaluation scripts at the **merged** path — e.g. SEP
score, Alpaca injection ASR, InjecAgent, and the utility benchmarks. See the
[evaluation guide](https://github.com/lindsey98/PromptInjection#evaluation).

## Intended use & limitations

- **Intended use:** research on prompt-injection defenses (text / single-turn).
- **Scope:** 3-role text setting only; for tool-calling agents use the 4-role
  Llama-3.1 checkpoint instead.
- DRIP reduces—but does not eliminate—prompt-injection risk; do not rely on it as
  the sole safeguard in production.

## Citation

> 📌 *This work is not yet officially published. Citation details will be added
> once the paper is released.*

Code: https://github.com/lindsey98/PromptInjection

License inherited from the base model: **Meta Llama 3 Community License**.