# Conversational Style Question Answering on LLaVA-Bench

Source: [Conversational Style Question Answering on LLaVA-Bench](<https://www.sota2.com/research/sota/conversational-style-question-answering-on-llava-bench>)

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

Data updated: 2026-02-26T05:48:10.193Z

Evaluation of model performance on the LLaVA-Bench dataset focusing specifically on conversational style question answering capabilities, measured by the LLaVA^W metric.

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 | Conversational Style Question Answering |
| Dataset (including any stated split) | LLaVA-Bench |
| Metric | LLaVA^W |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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

- [LLaVA^W](<https://www.sota2.com/research/sota/conversational-style-question-answering-on-llava-bench.md?metric=LLaVA%5EW>): Higher is better; unit: PERCENT.

## Recorded results

| Method | LLaVA^W | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen-VL | 63.4 | Not recorded | 68fa73651e209131cea5c2d8 |
| LLaVA-1.5 | 63.4 | Not recorded | 68fa73651e209131cea5c2d8 |
| GLaMM | 63.3 | Not recorded | 68fa73651e209131cea5c2d8 |
| InstructBLIP | 60.9 | Not recorded | 68fa73651e209131cea5c2d8 |
| Qwen-VL-Chat | 58.6 | Not recorded | 68fa73651e209131cea5c2d8 |
| BLIP-2 | 38.1 | Not recorded | 68fa73651e209131cea5c2d8 |

### Qwen-VL

LLaVA^W: 63.4. Result ID: 6992c5bc3a6c314849116499.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70d41e209131cea4d774.md>)

[Implementation](<https://github.com/QwenLM/Qwen-VL>)

Recorded evaluation configuration: {"LLM":"Qwen-7B"}. This record does not assert independent reproduction.

### LLaVA-1.5

LLaVA^W: 63.4. Result ID: 6992c5bc3a6c31484911649b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa732e1e209131cea5af59.md>)

[Implementation](<https://llava-vl.github.io/>)

Recorded evaluation configuration: {"LLM":"Vicuna-7B"}. This record does not assert independent reproduction.

### GLaMM

LLaVA^W: 63.3. Result ID: 6992c5bc3a6c31484911649c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Implementation](<https://mbzuai-oryx.github.io/groundingLMM>)

Recorded evaluation configuration: {"LLM":"Vicuna-7B"}. This record does not assert independent reproduction.

### InstructBLIP

LLaVA^W: 60.9. Result ID: 6992c5bc3a6c314849116498.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6fb71e209131cea47021.md>)

[Implementation](<https://github.com/salesforce/LAVIS/tree/main/projects/instructblip>)

Recorded evaluation configuration: {"LLM":"Vicuna-7B"}. This record does not assert independent reproduction.

### Qwen-VL-Chat

LLaVA^W: 58.6. Result ID: 6992c5bc3a6c31484911649a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70d41e209131cea4d774.md>)

[Implementation](<https://github.com/QwenLM/Qwen-VL>)

Recorded evaluation configuration: {"LLM":"Qwen-7B"}. This record does not assert independent reproduction.

### BLIP-2

LLaVA^W: 38.1. Result ID: 6992c5bc3a6c314849116497.

[Reporting paper](<https://www.sota2.com/research/paper/68fa73651e209131cea5c2d8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6fb71e209131cea46fe6.md>)

[Implementation](<https://github.com/salesforce/LAVIS/tree/main/projects/blip2>)

Recorded evaluation configuration: {"LLM":"Vicuna-13B"}. This record does not assert independent reproduction.

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