SVDQuant: Benchmarking Dataset for Quantized Diffusion Models

SVDQuant datasets are used for benchmarking the SVDQuant method, which absorbs outliers by low-rank components for 4-bit diffusion models. The dataset is hosted by MIT-HAN-Lab and was last updated on July 21, 2025. It is associated with a research paper and code repository for the SVDQuant technique.

Use Cases

• Benchmarking 4-bit quantization performance based on the SVDQuant method described in the paper.
• Evaluating the impact of low-rank component absorption on diffusion model output quality.
• Comparing diffusion model efficiency and accuracy before and after quantization.

Strengths

• Dataset is directly linked to a peer-reviewed research paper (Li et al., 2024).
• Hosted by a recognized research lab (MIT-HAN-Lab).
• Last update timestamp is provided (2025-07-21 05:00:56).

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

mit-han-lab

Collection Method

Created for benchmarking the SVDQuant method as described in the associated research paper.

Freshness

Last updated 2025-07-21 05:00:56; freshness should be verified.