ostris/ai-toolkit
1
0
Fork 0
mirror of https://github.com/ostris/ai-toolkit.git synced 2026-04-29 18:51:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fuse flux schnell assistant adapter in pieces when doing lowvram to drastically speed ip up from minutes to seconds.

Browse Source
This commit is contained in:
Jaret Burkett
2024-08-18 09:09:11 -06:00
parent 81899310f8
commit f944eeaa4d
1 changed files with 55 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
toolkit/stable_diffusion_model.py
View File

@@ -523,11 +523,7 @@ class StableDiffusion:
if self.model_config.lora_path is not None: if self.model_config.lora_path is not None:
print("Fusing in LoRA") print("Fusing in LoRA")
# if doing low vram, do this on the gpu, painfully slow otherwise # need the pipe for peft
if self.low_vram:
print(" - this process is painfully slow with 'low_vram' enabled. Disable it if possible.")
# need the pipe to do this unfortunately for now
# we have to fuse in the weights before quantizing
pipe: FluxPipeline = FluxPipeline( pipe: FluxPipeline = FluxPipeline(
scheduler=None, scheduler=None,
text_encoder=None, text_encoder=None,
@@ -537,10 +533,60 @@ class StableDiffusion:
vae=None, vae=None,
transformer=transformer, transformer=transformer,
) )
pipe.load_lora_weights(self.model_config.lora_path, adapter_name="lora1") if self.low_vram:
pipe.fuse_lora() # we cannot fuse the loras all at once without ooming in lowvram mode, so we have to do it in parts
# unfortunately, not an easier way with peft # we can do it on the cpu but it takes about 5-10 mins vs seconds on the gpu
pipe.unload_lora_weights() # we are going to separate it into the two transformer blocks one at a time
lora_state_dict = load_file(self.model_config.lora_path)
single_transformer_lora = {}
single_block_key = "transformer.single_transformer_blocks."
double_transformer_lora = {}
double_block_key = "transformer.transformer_blocks."
for key, value in lora_state_dict.items():
if single_block_key in key:
new_key = key.replace(single_block_key, "")
single_transformer_lora[new_key] = value
elif double_block_key in key:
new_key = key.replace(double_block_key, "")
double_transformer_lora[new_key] = value
else:
raise ValueError(f"Unknown lora key: {key}. Cannot load this lora in low vram mode")
# double blocks
transformer.transformer_blocks = transformer.transformer_blocks.to(
torch.device(self.quantize_device), dtype=dtype
)
pipe.load_lora_weights(double_transformer_lora, adapter_name=f"lora1_double")
pipe.fuse_lora()
transformer.transformer_blocks = transformer.transformer_blocks.to(
'cpu', dtype=dtype
)
# single blocks
transformer.single_transformer_blocks = transformer.single_transformer_blocks.to(
torch.device(self.quantize_device), dtype=dtype
)
pipe.load_lora_weights(single_transformer_lora, adapter_name=f"lora1_single")
pipe.fuse_lora()
transformer.single_transformer_blocks = transformer.single_transformer_blocks.to(
'cpu', dtype=dtype
)
# cleanup
del single_transformer_lora
del double_transformer_lora
del lora_state_dict
flush()
else:
# need the pipe to do this unfortunately for now
# we have to fuse in the weights before quantizing
pipe.load_lora_weights(self.model_config.lora_path, adapter_name="lora1")
pipe.fuse_lora()
# unfortunately, not an easier way with peft
pipe.unload_lora_weights()
flush()
if self.model_config.quantize: if self.model_config.quantize:
quantization_type = qfloat8 quantization_type = qfloat8