pipeline_cls
stringlengths 23
35
| ckpt_id
stringlengths 14
43
| batch_size
int64 1
1
| num_inference_steps
int64 1
50
| model_cpu_offload
bool 1
class | run_compile
bool 2
classes | time (secs)
stringlengths 12
15
| memory (gbs)
stringlengths 13
15
| actual_gpu_memory (gbs)
float64 22
22
| github_sha
stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|
StableDiffusionImg2ImgPipeline | Lykon/DreamShaper | 1 | 50 | false | false | 2.703 (-1.53%) | 3.193 (+0.28%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionImg2ImgPipeline | Lykon/DreamShaper | 1 | 50 | false | true | 2.027 (-1.70%) | 3.193 (+0.19%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionInpaintPipeline | Lykon/DreamShaper | 1 | 50 | false | false | 3.384 (-1.51%) | 3.193 (+0.41%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionInpaintPipeline | Lykon/DreamShaper | 1 | 50 | false | true | 2.544 (-1.74%) | 3.192 (+0.25%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionPipeline | Lykon/DreamShaper | 1 | 50 | false | false | 3.337 (-0.60%) | 3.184 (-0.22%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionPipeline | Lykon/DreamShaper | 1 | 50 | false | true | 2.452 (-1.01%) | 3.187 (-0.19%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLAdapterPipeline | TencentARC/t2i-adapter-canny-sdxl-1.0 | 1 | 50 | false | false | 19.214 (-0.21%) | 10.645 (-0.04%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLAdapterPipeline | TencentARC/t2i-adapter-canny-sdxl-1.0 | 1 | 50 | false | true | 17.812 (+0.20%) | 10.649 (-0.04%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionAdapterPipeline | TencentARC/t2iadapter_canny_sd14v1 | 1 | 50 | false | false | 3.276 (-1.56%) | 3.34 (-0.27%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionAdapterPipeline | TencentARC/t2iadapter_canny_sd14v1 | 1 | 50 | false | true | 3.07 (-1.22%) | 3.34 (+0.06%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLControlNetPipeline | diffusers/controlnet-canny-sdxl-1.0 | 1 | 50 | false | false | 27.562 (-0.75%) | 12.96 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLControlNetPipeline | diffusers/controlnet-canny-sdxl-1.0 | 1 | 50 | false | true | 23.929 (-1.05%) | 12.858 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
KandinskyV22CombinedPipeline | kandinsky-community/kandinsky-2-2-decoder | 1 | 50 | false | false | 4.089 (-0.97%) | 9.764 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
KandinskyV22CombinedPipeline | kandinsky-community/kandinsky-2-2-decoder | 1 | 50 | false | true | 3.483 (-0.99%) | 9.766 (-0.36%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | latent-consistency/lcm-lora-sdxl | 1 | 4 | false | false | 1.649 (-1.55%) | 10.468 (+0.01%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | latent-consistency/lcm-lora-sdxl | 1 | 4 | false | true | 1.763 (-0.62%) | 10.837 (-0.09%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionControlNetPipeline | lllyasviel/sd-controlnet-canny | 1 | 50 | false | false | 4.501 (-1.85%) | 3.893 (-0.26%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionControlNetPipeline | lllyasviel/sd-controlnet-canny | 1 | 50 | false | true | 3.348 (-1.53%) | 3.878 (+0.28%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | segmind/SSD-1B | 1 | 50 | false | false | 12.212 (-0.87%) | 8.113 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | segmind/SSD-1B | 1 | 50 | false | true | 10.596 (-1.68%) | 8.113 (-0.12%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLImg2ImgPipeline | stabilityai/sdxl-turbo | 1 | 2 | false | false | 0.421 (-0.71%) | 7.653 (-0.04%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLImg2ImgPipeline | stabilityai/sdxl-turbo | 1 | 2 | false | true | 1.625 (+0.18%) | 7.654 (+0.01%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | stabilityai/sdxl-turbo | 1 | 1 | false | false | 0.321 (+1.58%) | 7.655 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | stabilityai/sdxl-turbo | 1 | 1 | false | true | 1.487 (-2.43%) | 7.653 (-0.04%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLInpaintPipeline | stabilityai/stable-diffusion-xl-base-1.0 | 1 | 50 | false | false | 19.675 (-1.01%) | 10.468 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLInpaintPipeline | stabilityai/stable-diffusion-xl-base-1.0 | 1 | 50 | false | true | 17.261 (-2.04%) | 10.467 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | stabilityai/stable-diffusion-xl-base-1.0 | 1 | 50 | false | false | 19.265 (-0.42%) | 10.465 (-0.02%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLPipeline | stabilityai/stable-diffusion-xl-base-1.0 | 1 | 50 | false | true | 16.716 (-0.41%) | 10.472 (+0.05%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLImg2ImgPipeline | stabilityai/stable-diffusion-xl-refiner-1.0 | 1 | 50 | false | false | 7.2 (-1.30%) | 9.613 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
StableDiffusionXLImg2ImgPipeline | stabilityai/stable-diffusion-xl-refiner-1.0 | 1 | 50 | false | true | 6.539 (-0.92%) | 9.622 (-0.01%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
WuerstchenCombinedPipeline | warp-ai/wuerstchen | 1 | 50 | false | false | 4.547 (+5.13%) | 6.075 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
WuerstchenCombinedPipeline | warp-ai/wuerstchen | 1 | 50 | false | true | 4.512 (+4.08%) | 6.072 (0.00%) | 21.951 | e2ead7cdcc00859533e6bec7b0707a6fb0efef0a |
Welcome to 🤗 Diffusers Benchmarks!
This is dataset where we keep track of the inference latency and memory information of the core pipelines in the diffusers library.
Currently, the core pipelines are the following:
- Stable Diffusion and its derivatives such as ControlNet, T2I Adapter, Image-to-Image, Inpainting
- Stable Diffusion XL and its derivatives
- SSD-1B
- Kandinsky
- Würstchen
- LCM
Note that we will continue to extend the list of core pipelines based on their API usage.
We use this GitHub Actions workflow to report the above numbers automatically. This workflow runs on a biweekly cadence.
The benchmarks are run on an A10G GPU.
- Downloads last month
- 4