# Word Deletion Robustness on ELI5 prompts 32-bit payload Llama3.1-8B (test)

Source: [Word Deletion Robustness on ELI5 prompts 32-bit payload Llama3.1-8B (test)](<https://www.sota2.com/research/sota/word-deletion-robustness-on-eli5-prompts-32-bit-payload-llama3-1-8b-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-13T09:14:07.813Z

Evaluation of model robustness against word deletion perturbations on text generated by Llama3.1-8B using the ELI5 32-bit payload dataset. Performance is measured by Bit Accuracy across various levels of deletion/corruption (10% to 50%).

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Word Deletion Robustness |
| Dataset (including any stated split) | ELI5 prompts 32-bit payload Llama3.1-8B (test) |
| Metric | Bit Accuracy (10% Deletion) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Bit Accuracy (10% Deletion)](<https://www.sota2.com/research/sota/word-deletion-robustness-on-eli5-prompts-32-bit-payload-llama3-1-8b-test.md?metric=Bit%20Accuracy%20(10%25%20Deletion)>): Higher is better; unit: PERCENT.

- [Bit Accuracy (20% Deletion)](<https://www.sota2.com/research/sota/word-deletion-robustness-on-eli5-prompts-32-bit-payload-llama3-1-8b-test.md?metric=Bit%20Accuracy%20(20%25%20Deletion)>): Higher is better; unit: PERCENT.

- [Bit Accuracy (30% Deletion)](<https://www.sota2.com/research/sota/word-deletion-robustness-on-eli5-prompts-32-bit-payload-llama3-1-8b-test.md?metric=Bit%20Accuracy%20(30%25%20Deletion)>): Higher is better; unit: PERCENT.

- [Bit Accuracy (40% Deletion)](<https://www.sota2.com/research/sota/word-deletion-robustness-on-eli5-prompts-32-bit-payload-llama3-1-8b-test.md?metric=Bit%20Accuracy%20(40%25%20Deletion)>): Higher is better; unit: PERCENT.

- [Bit Accuracy (50% Deletion)](<https://www.sota2.com/research/sota/word-deletion-robustness-on-eli5-prompts-32-bit-payload-llama3-1-8b-test.md?metric=Bit%20Accuracy%20(50%25%20Deletion)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Bit Accuracy (10% Deletion) | Result date | Report paper |
| --- | --- | --- | --- |
| Ours+ | 70.7 | Not recorded | 6a0405c4dbc836b87d8277a1 |
| BiMark | 66.4 | Not recorded | 6a0405c4dbc836b87d8277a1 |
| MC2Mark | 66.4 | Not recorded | 6a0405c4dbc836b87d8277a1 |
| MPAC | 60.9 | Not recorded | 6a0405c4dbc836b87d8277a1 |
| StealthInk | 60.6 | Not recorded | 6a0405c4dbc836b87d8277a1 |
| MirrorMark | 53.9 | Not recorded | 6a0405c4dbc836b87d8277a1 |
| RSBH | 51.1 | Not recorded | 6a0405c4dbc836b87d8277a1 |

### Ours+

Bit Accuracy (10% Deletion): 70.7. Result ID: 6a04400d47f34ce924ea53e5.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B","Type":"Stateful Encoder"}. This record does not assert independent reproduction.

### BiMark

Bit Accuracy (10% Deletion): 66.4. Result ID: 6a04400d47f34ce924ea53df.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7acf1e209131cea85383.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B"}. This record does not assert independent reproduction.

### MC2Mark

Bit Accuracy (10% Deletion): 66.4. Result ID: 6a04400d47f34ce924ea53e0.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6993f6829ebc499c97409e52.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B"}. This record does not assert independent reproduction.

### MPAC

Bit Accuracy (10% Deletion): 60.9. Result ID: 6a04400d47f34ce924ea53e1.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72911e209131cea57239.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B"}. This record does not assert independent reproduction.

### StealthInk

Bit Accuracy (10% Deletion): 60.6. Result ID: 6a04400d47f34ce924ea53e4.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa792b1e209131cea7c607.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B"}. This record does not assert independent reproduction.

### MirrorMark

Bit Accuracy (10% Deletion): 53.9. Result ID: 6a04400d47f34ce924ea53e2.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802faea8b01e76cc8e1680.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B"}. This record does not assert independent reproduction.

### RSBH

Bit Accuracy (10% Deletion): 51.1. Result ID: 6a04400d47f34ce924ea53e3.

[Reporting paper](<https://www.sota2.com/research/paper/6a0405c4dbc836b87d8277a1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b51e209131cea6e998.md>)

Recorded evaluation configuration: {"Payload":"32-bit","Backbone":"LLAMA3.1-8B"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
