# Lipschitz Ball Verification on Qwen 7B-Instruct LoRA 2.5 (train)

Source: [Lipschitz Ball Verification on Qwen 7B-Instruct LoRA 2.5 (train)](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train>)

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

Data updated: 2026-04-02T08:33:02.335Z

Evaluation of model stability and convergence properties using Lipschitz Ball Verification techniques for the Qwen 7B-Instruct model fine-tuned via LoRA (version 2.5) during the training phase.

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 | Lipschitz Ball Verification |
| Dataset (including any stated split) | Qwen 7B-Instruct LoRA 2.5 (train) |
| Metric | Ball Acceptance Rate |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 2 of 2 |
| 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

- [Ball Acceptance Rate](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Ball%20Acceptance%20Rate>): Higher is better; unit: PERCENT.

- [Oracle Calls Required](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Oracle%20Calls%20Required>): Lower is better; unit: SCALAR.

- [Oracle Rejections](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Oracle%20Rejections>): Lower is better; unit: SCALAR.

- [Chain Transitions](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Chain%20Transitions>): Lower is better; unit: SCALAR.

- [Loss Reduction (%)](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Loss%20Reduction%20(%25)>): Higher is better; unit: PERCENT.

- [Sum of Lk](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Sum%20of%20Lk>): Lower is better; unit: SCALAR.

- [Ball Radius (r)](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Ball%20Radius%20(r)>): Lower is better; unit: OTHER.

- [Total Displacement](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Total%20Displacement>): Lower is better; unit: OTHER.

- [Model Safe Status](<https://www.sota2.com/research/sota/lipschitz-ball-verification-on-qwen-7b-instruct-lora-2-5-train.md?metric=Model%20Safe%20Status>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Ball Acceptance Rate | Result date | Report paper |
| --- | --- | --- | --- |
| Lipschitz ball verifier | 79 | Not recorded | 69cdf7e80636e5bf4afcc91b |
| Lipschitz ball verifier | 75.5 | Not recorded | 69cdf7e80636e5bf4afcc91b |

### Lipschitz ball verifier

Ball Acceptance Rate: 79. Result ID: 69ce299347f34ce924b74e7b.

[Reporting paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

Recorded evaluation configuration: {"Backbone model":"Qwen2.5-7B-Instruct","Fine-tuning method":"LoRA","LoRA rank":"4","Target modules":"Wq and Wv","Number of trainable parameters":"1,261,568","Oracle Evaluation":"10-prompt"}. This record does not assert independent reproduction.

### Lipschitz ball verifier

Ball Acceptance Rate: 75.5. Result ID: 69ce299347f34ce924b74e7c.

[Reporting paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cdf7e80636e5bf4afcc91b.md>)

Recorded evaluation configuration: {"Backbone model":"Qwen2.5-7B-Instruct","Fine-tuning method":"LoRA","LoRA rank":"4","Target modules":"Wq and Wv","Number of trainable parameters":"1,261,568","Oracle Evaluation":"50-prompt"}. This record does not assert independent reproduction.

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