stable-diffusion-webui-comp.../composable_lora.py

176 lines
7.1 KiB
Python

from typing import List, Dict
import re
import torch
from modules import extra_networks, prompt_parser, shared
re_AND = re.compile(r"\bAND\b")
def load_prompt_loras(prompt: str):
prompt_loras.clear()
subprompts = re_AND.split(prompt)
for i, subprompt in enumerate(subprompts):
loras = {}
_, extra_network_data = extra_networks.parse_prompt(subprompt)
for params in extra_network_data['lora']:
name = params.items[0]
multiplier = float(params.items[1]) if len(params.items) > 1 else 1.0
loras[name] = multiplier
prompt_loras.append(loras)
def reset_counters():
global get_learned_conditioning_prompt_schedules_counter
global text_model_encoder_counter
global diffusion_model_counter
get_learned_conditioning_prompt_schedules_counter = 0
text_model_encoder_counter = 0
diffusion_model_counter = 0
def debug(*values: object, lora_layer_name: str):
if verbose:
if lora_layer_name.startswith("transformer_"):
if lora_layer_name.endswith("_11_mlp_fc2"):
print(*values)
elif lora_layer_name.startswith("diffusion_model_"):
if lora_layer_name.endswith("_11_1_proj_out"):
print(*values)
def lora_forward(compvis_module, input, res):
global get_learned_conditioning_prompt_schedules_counter
global text_model_encoder_counter
global diffusion_model_counter
import lora
if len(lora.loaded_loras) == 0:
return res
lora_layer_name: str | None = getattr(compvis_module, 'lora_layer_name', None)
if lora_layer_name is None:
return res
num_prompts = len(prompt_loras)
# debug(f"lora.forward lora_layer_name={lora_layer_name} in.shape={input.shape} res.shape={res.shape} num_batches={num_batches} num_prompts={num_prompts}", lora_layer_name=lora_layer_name)
for lora in lora.loaded_loras:
module = lora.modules.get(lora_layer_name, None)
if module is None:
continue
if shared.opts.lora_apply_to_outputs and res.shape == input.shape:
patch = module.up(module.down(res))
else:
patch = module.up(module.down(input))
alpha = module.alpha / module.up.weight.shape[1] if module.alpha else 1.0
# debug(f"lora.name={lora.name} lora.mul={lora.multiplier} alpha={alpha} pat.shape={patch.shape}")
if enabled:
if lora_layer_name.startswith("transformer_"): # "transformer_text_model_encoder_"
#
if get_learned_conditioning_prompt_schedules_counter != 1 and 0 <= text_model_encoder_counter < len(prompt_loras):
# c
loras = prompt_loras[text_model_encoder_counter]
multiplier = loras.get(lora.name, 0.0)
if multiplier != 0.0:
debug(f"c #{text_model_encoder_counter} lora.name={lora.name} mul={multiplier}", lora_layer_name=lora_layer_name)
res += multiplier * alpha * patch
else:
# uc
if opt_uc_text_model_encoder and lora.multiplier != 0.0:
debug(f"uc #{text_model_encoder_counter} lora.name={lora.name} lora.mul={lora.multiplier}", lora_layer_name=lora_layer_name)
res += lora.multiplier * alpha * patch
if get_learned_conditioning_prompt_schedules_counter != 1 and lora_layer_name.endswith("_11_mlp_fc2"): # last lora_layer_name of text_model_encoder
text_model_encoder_counter += 1
# c1 c2 c1 c2 ..
if text_model_encoder_counter == len(prompt_loras):
text_model_encoder_counter = 0
elif res.shape[0] == num_batches * num_prompts + num_batches: # "diffusion_model_"
# tensor.shape[1] == uncond.shape[1]
tensor_off = 0
uncond_off = num_batches * num_prompts
for b in range(num_batches):
# c
for p, loras in enumerate(prompt_loras):
multiplier = loras.get(lora.name, 0.0)
if multiplier != 0.0:
debug(f"tensor #{b}.{p} lora.name={lora.name} mul={multiplier}", lora_layer_name=lora_layer_name)
res[tensor_off] += multiplier * alpha * patch[tensor_off]
tensor_off += 1
# uc
if opt_uc_diffusion_model and lora.multiplier != 0.0:
debug(f"uncond lora.name={lora.name} lora.mul={lora.multiplier}", lora_layer_name=lora_layer_name)
res[uncond_off] += lora.multiplier * alpha * patch[uncond_off]
uncond_off += 1
else: # "diffusion_model_"
# tensor.shape[1] != uncond.shape[1]
if 0 <= diffusion_model_counter < len(prompt_loras):
# c
loras = prompt_loras[diffusion_model_counter]
multiplier = loras.get(lora.name, 0.0)
if multiplier != 0.0:
debug(f"c #{diffusion_model_counter} lora.name={lora.name} mul={multiplier}", lora_layer_name=lora_layer_name)
res += multiplier * alpha * patch
else:
# uc
if opt_uc_diffusion_model and lora.multiplier != 0.0:
debug(f"uc {lora_layer_name} lora.name={lora.name} lora.mul={lora.multiplier}", lora_layer_name=lora_layer_name)
res += lora.multiplier * alpha * patch
if lora_layer_name.endswith("_11_1_proj_out"): # last lora_layer_name of diffusion_model
diffusion_model_counter += 1
# c1 c2 .. uc
if diffusion_model_counter == len(prompt_loras) + 1:
diffusion_model_counter = 0
else:
# default
if lora.multiplier != 0.0:
debug(f"DEFAULT {lora_layer_name} lora.name={lora.name} lora.mul={lora.multiplier}", lora_layer_name=lora_layer_name)
res += lora.multiplier * alpha * patch
return res
def lora_Linear_forward(self, input):
return lora_forward(self, input, torch.nn.Linear_forward_before_lora(self, input))
def lora_Conv2d_forward(self, input):
return lora_forward(self, input, torch.nn.Conv2d_forward_before_lora(self, input))
def lora_get_learned_conditioning_prompt_schedules(prompts, steps):
global get_learned_conditioning_prompt_schedules_counter
#
# order: uc c
#
prompt_schedules = prompt_parser.get_learned_conditioning_prompt_schedules_before_lora(prompts, steps)
get_learned_conditioning_prompt_schedules_counter += 1
return prompt_schedules
enabled = False
opt_uc_text_model_encoder = False
opt_uc_diffusion_model = False
verbose = False
num_batches: int = 0
prompt_loras: List[Dict[str, float]] = []
get_learned_conditioning_prompt_schedules_counter: int = 0
text_model_encoder_counter: int = 0
diffusion_model_counter: int = 0