diffusers-workflow/workflows/models/flux-dev-compile.json at master · dkackman/diffusers-workflow · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
{
    "id": "FluxDevCompile",
    "description": "FLUX.1 dev with regional torch.compile on the transformer, otherwise the plain bf16 model-offload configuration of models/flux-dev. Measured on an RTX 3090 (24GB) at 1024x1024, 28 steps: 52s per image with the pipeline loaded against 55s for models/flux-dev, 64s against 72s cold. The earlier int8 TorchAO recipe this entry carried ran at over a minute per denoising step on the same card, with or without the first-block cache, so it is gone; float8 TorchAO needs an Ada (40-series) card and is unmeasured here. The compile is skipped with a warning on MPS.",
    "summary": "FLUX.1 dev, bf16 model offload plus regional compile: 52s per 1024x1024 image on an RTX 3090, 55s without.",
    "cost": [
        {"device": "cuda", "name": "RTX 3090", "vram_gb": 24, "minutes": 0.9}
    ],
    "configures": "templates/text-to-image",
    "variables": {
        "prompt": "a marmot wearing a tophat, in a forest scene.",
        "num_inference_steps": 28,
        "guidance_scale": 3.5,
        "width": 1024,
        "height": 1024,
        "model_name": "black-forest-labs/FLUX.1-dev"
    },
    "steps": [
        {
            "name": "main",
            "pipeline": {
                "configuration": {
                    "component_type": "FluxPipeline",
                    "offload": "model",
                    "components": {
                        "transformer": {
                            "compile": {
                                "repeated_blocks": true
                            }
                        }
                    }
                },
                "from_pretrained_arguments": {
                    "model_name": "variable:model_name",
                    "torch_dtype": "torch.bfloat16"
                },
                "arguments": {
                    "prompt": "variable:prompt",
                    "num_inference_steps": "variable:num_inference_steps",
                    "guidance_scale": "variable:guidance_scale",
                    "width": "variable:width",
                    "height": "variable:height",
                    "max_sequence_length": 512
                }
            },
            "result": {
                "content_type": "image/jpeg",
                "embed_metadata": true
            }
        }
    ]
}