# Image Classification on ImageNet-1k (Quantization Accuracy Sweep)

Source: [Image Classification on ImageNet-1k (Quantization Accuracy Sweep)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-quantization-accuracy-sweep>)

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

Data updated: 2026-03-20T09:46:09.236Z

Evaluation of Image Classification performance on the ImageNet-1k dataset, focusing on accuracy results across various weight quantization schemes (e.g., FP32, W3A16, W4A16, W6A16, W8A16).

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 | Image Classification |
| Dataset (including any stated split) | ImageNet-1k |
| Metric | Accuracy (FP32) |
| 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

- [Accuracy (FP32)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-quantization-accuracy-sweep.md?metric=Accuracy%20(FP32)>): Higher is better; unit: PERCENT.

- [Accuracy (W3A16)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-quantization-accuracy-sweep.md?metric=Accuracy%20(W3A16)>): Higher is better; unit: PERCENT.

- [Accuracy (W4A16)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-quantization-accuracy-sweep.md?metric=Accuracy%20(W4A16)>): Higher is better; unit: PERCENT.

- [Accuracy (W6A16)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-quantization-accuracy-sweep.md?metric=Accuracy%20(W6A16)>): Higher is better; unit: PERCENT.

- [Accuracy (W8A16)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-quantization-accuracy-sweep.md?metric=Accuracy%20(W8A16)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (FP32) | Result date | Report paper |
| --- | --- | --- | --- |
| AWQ | 68.32 | Not recorded | 68fa7bfd1e209131cea8bade |
| AWQ + RegCache | 68.32 | Not recorded | 68fa7bfd1e209131cea8bade |

### AWQ

Accuracy (FP32): 68.32. Result ID: 69bd174d47f34ce924a797c4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bfd1e209131cea8bade.md>)

Recorded evaluation configuration: {"Model":"CLIP-B/16","Evaluation Protocol":"Zero-shot","Quantization Group Size":"128"}. This record does not assert independent reproduction.

### AWQ + RegCache

Accuracy (FP32): 68.32. Result ID: 69bd174d47f34ce924a797c5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bfd1e209131cea8bade.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bfd1e209131cea8bade.md>)

Recorded evaluation configuration: {"Model":"CLIP-B/16","Evaluation Protocol":"Zero-shot","Quantization Group Size":"128"}. This record does not assert independent reproduction.

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