Attempting a universal implementation of EasyCache, starting with flux as test; I screwed up the math a bit, but when I set it just right it works.

This commit is contained in:
Jedrzej Kosinski 2025-08-18 22:01:31 -07:00
parent bd2ab73976
commit 26d54b1de5
3 changed files with 125 additions and 0 deletions

View File

@ -6,6 +6,7 @@ import torch
from torch import Tensor, nn
from einops import rearrange, repeat
import comfy.ldm.common_dit
import comfy.patcher_extension
from .layers import (
DoubleStreamBlock,
@ -214,6 +215,13 @@ class Flux(nn.Module):
return img, repeat(img_ids, "h w c -> b (h w) c", b=bs)
def forward(self, x, timestep, context, y=None, guidance=None, ref_latents=None, control=None, transformer_options={}, **kwargs):
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
self._forward,
self,
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
).execute(x, timestep, context, y, guidance, ref_latents, control, transformer_options, **kwargs)
def _forward(self, x, timestep, context, y=None, guidance=None, ref_latents=None, control=None, transformer_options={}, **kwargs):
bs, c, h_orig, w_orig = x.shape
patch_size = self.patch_size

View File

@ -0,0 +1,116 @@
from comfy_api.latest import io, ComfyExtension
import comfy.patcher_extension
import logging
import torch
import comfy.model_patcher
def easycache_sample_wrapper(executor, *args, **kwargs):
try:
guider = executor.class_obj
orig_model_options = guider.model_options
guider.model_options = comfy.model_patcher.create_model_options_clone(orig_model_options)
if "easycache" in orig_model_options["transformer_options"]:
guider.model_options["transformer_options"]["easycache"] = guider.model_options["transformer_options"]["easycache"].clone()
guider.model_options["transformer_options"]["easycache"].dict["start_timestep"] = guider.model_patcher.model.model_sampling.percent_to_sigma(guider.model_options["transformer_options"]["easycache"].dict["start_percent"])
guider.model_options["transformer_options"]["easycache"].dict["end_timestep"] = guider.model_patcher.model.model_sampling.percent_to_sigma(guider.model_options["transformer_options"]["easycache"].dict["end_percent"])
return executor(*args, **kwargs)
finally:
guider.model_options = orig_model_options
def easycache_forward_wrapper(executor, *args, **kwargs):
x: torch.Tensor = args[0]
timestep: torch.Tensor = args[1]
transformer_options = args[-1]
do_easycache = timestep < transformer_options["easycache"].dict["start_timestep"] and timestep > transformer_options["easycache"].dict["end_timestep"]
logging.info(f"easycache_wrapper: do_easycache: {do_easycache}")
x_prev = None
input_change = None
# input_data = x.flatten().abs().mean()
if do_easycache and "easycache" in transformer_options:
if "x_prev" in transformer_options["easycache"].dict:
x_prev = transformer_options["easycache"].dict["x_prev"]
else:
transformer_options["easycache"].dict["x_prev"] = x.clone()
if x_prev is not None:
input_change = (x_prev - x).flatten().abs().mean()
if do_easycache and transformer_options["easycache"].dict.get("change_rate", None) is not None:
change_rate = transformer_options["easycache"].dict["change_rate"]
output_prev = transformer_options["easycache"].dict["output_prev"]
pred_change = change_rate * (input_change / output_prev.flatten().abs().mean())
accumulated_change = transformer_options["easycache"].dict["accumulated_change"] + pred_change
if transformer_options["easycache"].dict["reuse_threshold"] <= accumulated_change:
logging.info(f"easycache_wrapper: skipping step; accumulated_change: {accumulated_change}, reuse_threshold: {transformer_options['easycache'].dict['reuse_threshold']}")
transformer_options["easycache"].dict["accumulated_change"] = 0.0
return x + transformer_options["easycache"].dict["cache_diff"]
else:
transformer_options["easycache"].dict["accumulated_change"] = accumulated_change
logging.info(f"easycache_wrapper: NOT skipping step; accumulated_change: {accumulated_change}, reuse_threshold: {transformer_options['easycache'].dict['reuse_threshold']}")
logging.info(f"easycache_wrapper pred_change: {pred_change}")
output: torch.Tensor = executor(*args, **kwargs)
if x_prev is not None:
# output_data = output.flatten().abs().mean()
output_prev = transformer_options["easycache"].dict["output_prev"]
output_change = (output_prev - output).flatten().abs().mean()
k = output_change / input_change
transformer_options["easycache"].dict["change_rate"] = k
logging.info(f"easycache_wrapper: {input_change} {output_change} {k}")
if do_easycache and "easycache" in transformer_options:
transformer_options["easycache"].dict["output_prev"] = output.clone()
transformer_options["easycache"].dict["cache_diff"] = output - x
if not do_easycache:
transformer_options["easycache"].dict["accumulated_change"] = 0.0
transformer_options["easycache"].dict["change_rate"] = None
transformer_options["easycache"].dict["output_prev"] = None
transformer_options["easycache"].dict["cache_diff"] = None
return output
class EasyCacheStore:
def __init__(self, dict: dict):
self.dict = dict
def clone(self):
return EasyCacheStore(self.dict.copy())
class EasyCacheNode(io.ComfyNode):
@classmethod
def define_schema(cls) -> io.Schema:
return io.Schema(
node_id="EasyCache",
display_name="Easy Cache",
description="Easy Cache",
category="advanced/debug/model",
inputs=[
io.Model.Input("model", tooltip="The model to add EasyCache to."),
io.Float.Input("reuse_threshold", min=0.0, default=0.0, max=100.0, step=0.01, tooltip="The threshold for reusing cached steps."),
io.Float.Input("start_percent", min=0.0, default=0.0, max=1.0, step=0.01, tooltip="The relative sampling step to begin use of EasyCache."),
io.Float.Input("end_percent", min=0.0, default=1.0, max=1.0, step=0.01, tooltip="The relative sampling step to end use of EasyCache."),
],
outputs=[
io.Model.Output(tooltip="The model with EasyCache."),
],
)
@classmethod
def execute(cls, model: io.Model.Type, reuse_threshold: float, start_percent: float, end_percent: float) -> io.NodeOutput:
model = model.clone()
easycache_dict = {
"reuse_threshold": reuse_threshold,
"start_percent": start_percent,
"end_percent": end_percent,
"accumulated_change": 0.0,
}
model.model_options["transformer_options"]["easycache"] = EasyCacheStore(easycache_dict)
model.add_wrapper_with_key(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, "easycache", easycache_forward_wrapper)
model.add_wrapper_with_key(comfy.patcher_extension.WrappersMP.OUTER_SAMPLE, "easycache", easycache_sample_wrapper)
return io.NodeOutput(model)
class EasyCacheExtension(ComfyExtension):
async def get_node_list(self) -> list[type[io.ComfyNode]]:
return [
EasyCacheNode,
]
def comfy_entrypoint():
return EasyCacheExtension()

View File

@ -2321,6 +2321,7 @@ async def init_builtin_extra_nodes():
"nodes_edit_model.py",
"nodes_tcfg.py",
"nodes_context_windows.py",
"nodes_easycache.py",
]
import_failed = []