Datasets:

derekl35
/

diffusers-quantization-benchmarks

Dataset card Data Studio Files Files and versions Community

Split (1)

train · 26 rows

Library stringclasses 5 values	Precision stringlengths 3 18	CPU Offload bool 2 classes	Memory After Loading (GB) float64 0 31.4 ⌀	Peak Memory GB float64 9.16 36.2 ⌀	Inference Time (Seconds) float64 12 109 ⌀
Diffusers	BF16	false	31.447	36.166	12
Diffusers	BF16	true	0	23.271	20
bitsandbytes	4-bit	false	12.584	17.281	12
bitsandbytes	4-bit	true	12.383	12.383	17
bitsandbytes	8-bit	false	19.273	24.432	27
bitsandbytes	8-bit	true	19.182	23.428	27
torchao	int4_weight_only	false	10.635	14.654	109
torchao	int4_weight_only	true	null	null	null
torchao	int8_weight_only	false	17.02	21.482	15
torchao	int8_weight_only	true	0	12.189	26
torchao	float8_weight_only	false	17.016	21.488	15
torchao	float8_weight_only	true	0	12.348	27
quanto	INT4	false	12.254	16.139	109
quanto	INT4	true	null	null	null
quanto	INT8	false	17.33	21.814	15
quanto	INT8	true	0	12.166	25
quanto	FP8	false	16.395	20.898	16
quanto	FP8	true	0	12.332	24
GGUF	Q4_1	false	16.838	21.326	23
GGUF	Q4_1	true	0	9.164	28
GGUF	Q8_0	false	21.502	25.973	15
GGUF	Q8_0	true	0	12.965	22
GGUF	Q2_k	false	13.264	17.752	26
GGUF	Q2_k	true	0	9.164	31
Diffusers	FP8 (e4m3)	false	23.682	28.451	13
Diffusers	FP8 (e4m3)	true	9.119	15.621	22

Subsets and Splits

Low Memory and Fast Inference Models

The query filters models with peak memory usage under 24 GB and inference time less than 30 seconds, providing basic information but limited analytical insight.