Spaces:
Runtime error
Runtime error
File size: 6,110 Bytes
ff271bf |
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 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 |
import json
import os
from dataclasses import dataclass
import torch
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file as load_sft
from flux.model import Flux, FluxParams
from flux.modules.autoencoder import AutoEncoder, AutoEncoderParams
from flux.modules.conditioner import HFEmbedder
@dataclass
class SamplingOptions:
prompt: str
width: int
height: int
num_steps: int
guidance: float
seed: int
@dataclass
class ModelSpec:
params: FluxParams
ae_params: AutoEncoderParams
ckpt_path: str
ae_path: str
repo_id: str
repo_flow: str
repo_ae: str
configs = {
"flux-dev": ModelSpec(
repo_id="black-forest-labs/FLUX.1-dev",
repo_flow="flux1-dev.safetensors",
repo_ae="ae.safetensors",
ckpt_path='models/flux1-dev.safetensors',
params=FluxParams(
in_channels=64,
vec_in_dim=768,
context_in_dim=4096,
hidden_size=3072,
mlp_ratio=4.0,
num_heads=24,
depth=19,
depth_single_blocks=38,
axes_dim=[16, 56, 56],
theta=10_000,
qkv_bias=True,
guidance_embed=True,
),
ae_path='models/ae.safetensors',
ae_params=AutoEncoderParams(
resolution=256,
in_channels=3,
ch=128,
out_ch=3,
ch_mult=[1, 2, 4, 4],
num_res_blocks=2,
z_channels=16,
scale_factor=0.3611,
shift_factor=0.1159,
),
),
"flux-schnell": ModelSpec(
repo_id="black-forest-labs/FLUX.1-schnell",
repo_flow="flux1-schnell.safetensors",
repo_ae="ae.safetensors",
ckpt_path=os.getenv("FLUX_SCHNELL"),
params=FluxParams(
in_channels=64,
vec_in_dim=768,
context_in_dim=4096,
hidden_size=3072,
mlp_ratio=4.0,
num_heads=24,
depth=19,
depth_single_blocks=38,
axes_dim=[16, 56, 56],
theta=10_000,
qkv_bias=True,
guidance_embed=False,
),
ae_path=os.getenv("AE"),
ae_params=AutoEncoderParams(
resolution=256,
in_channels=3,
ch=128,
out_ch=3,
ch_mult=[1, 2, 4, 4],
num_res_blocks=2,
z_channels=16,
scale_factor=0.3611,
shift_factor=0.1159,
),
),
}
def print_load_warning(missing: list[str], unexpected: list[str]) -> None:
if len(missing) > 0 and len(unexpected) > 0:
print(f"Got {len(missing)} missing keys:\n\t" + "\n\t".join(missing))
print("\n" + "-" * 79 + "\n")
print(f"Got {len(unexpected)} unexpected keys:\n\t" + "\n\t".join(unexpected))
elif len(missing) > 0:
print(f"Got {len(missing)} missing keys:\n\t" + "\n\t".join(missing))
elif len(unexpected) > 0:
print(f"Got {len(unexpected)} unexpected keys:\n\t" + "\n\t".join(unexpected))
def load_flow_model(name: str, device: str = "cuda", hf_download: bool = True):
# Loading Flux
print("Init model")
ckpt_path = configs[name].ckpt_path
if (
not os.path.exists(ckpt_path)
and configs[name].repo_id is not None
and configs[name].repo_flow is not None
and hf_download
):
ckpt_path = hf_hub_download(configs[name].repo_id, configs[name].repo_flow, local_dir='models')
with torch.device(device):
model = Flux(configs[name].params).to(torch.bfloat16)
if ckpt_path is not None:
print("Loading checkpoint")
# load_sft doesn't support torch.device
sd = load_sft(ckpt_path, device=str(device))
missing, unexpected = model.load_state_dict(sd, strict=False)
print_load_warning(missing, unexpected)
return model
# from XLabs-AI https://github.com/XLabs-AI/x-flux/blob/1f8ef54972105ad9062be69fe6b7f841bce02a08/src/flux/util.py#L330
def load_flow_model_quintized(name: str, device: str = "cuda", hf_download: bool = True):
# Loading Flux
print("Init model")
ckpt_path = 'models/flux-dev-fp8.safetensors'
if (
not os.path.exists(ckpt_path)
and hf_download
):
ckpt_path = hf_hub_download("XLabs-AI/flux-dev-fp8", "flux-dev-fp8.safetensors")
json_path = hf_hub_download("XLabs-AI/flux-dev-fp8", 'flux_dev_quantization_map.json')
model = Flux(configs[name].params).to(torch.bfloat16)
print("Loading checkpoint")
# load_sft doesn't support torch.device
sd = load_sft(ckpt_path, device='cpu')
with open(json_path) as f:
quantization_map = json.load(f)
print("Start a quantization process...")
from optimum.quanto import requantize
requantize(model, sd, quantization_map, device=device)
print("Model is quantized!")
return model
def load_t5(device: str = "cuda", max_length: int = 512) -> HFEmbedder:
# max length 64, 128, 256 and 512 should work (if your sequence is short enough)
return HFEmbedder("xlabs-ai/xflux_text_encoders", max_length=max_length, torch_dtype=torch.bfloat16).to(device)
def load_clip(device: str = "cuda") -> HFEmbedder:
return HFEmbedder("openai/clip-vit-large-patch14", max_length=77, torch_dtype=torch.bfloat16).to(device)
def load_ae(name: str, device: str = "cuda", hf_download: bool = True) -> AutoEncoder:
ckpt_path = configs[name].ae_path
if (
not os.path.exists(ckpt_path)
and configs[name].repo_id is not None
and configs[name].repo_ae is not None
and hf_download
):
ckpt_path = hf_hub_download(configs[name].repo_id, configs[name].repo_ae, local_dir='models')
# Loading the autoencoder
print("Init AE")
with torch.device(device):
ae = AutoEncoder(configs[name].ae_params)
if ckpt_path is not None:
sd = load_sft(ckpt_path, device=str(device))
missing, unexpected = ae.load_state_dict(sd, strict=False)
print_load_warning(missing, unexpected)
return ae
|