124 lines
5.3 KiB
Python
124 lines
5.3 KiB
Python
from inspect import cleandoc
|
|
from comfy_api.latest import io
|
|
from .helper_functions import commonLazy
|
|
from .modelLikeCommon import calculate_patches
|
|
from antlr4 import InputStream, CommonTokenStream
|
|
from .Parser.MathExprLexer import MathExprLexer
|
|
from .Parser.MathExprParser import MathExprParser
|
|
import re
|
|
|
|
|
|
class ModelMathNode(io.ComfyNode):
|
|
"""
|
|
This node enables the use of math expressions on Model weights (state_dict) using Autogrow inputs.
|
|
Functionally acts as a custom model merge.
|
|
"""
|
|
|
|
@classmethod
|
|
def define_schema(cls) -> io.Schema:
|
|
return io.Schema(
|
|
node_id="mrmth_ag_ModelMathNode",
|
|
display_name="Model Math",
|
|
category="More math",
|
|
inputs=[
|
|
io.Autogrow.Input(id="V",template=io.Autogrow.TemplatePrefix(io.Model.Input("values"), prefix="V", min=1, max=50)),
|
|
io.Autogrow.Input(id="F", template=io.Autogrow.TemplatePrefix(io.Float.Input("float", default=0.0, optional=True, lazy=True, force_input=True), prefix="F", min=1, max=50)),
|
|
io.String.Input(id="Expression", default="I0*(1-F0)+I1*F0", tooltip="Expression to apply on weights"),
|
|
io.Combo.Input(
|
|
id="length_mismatch",
|
|
options=["error", "passthrough", "pad"],
|
|
default="error",
|
|
tooltip="How to handle mismatched layer counts. For models, this usually defaults to broadcast (zero for missing layers)."
|
|
)
|
|
],
|
|
outputs=[
|
|
io.Model.Output(),
|
|
],
|
|
)
|
|
|
|
tooltip = cleandoc(__doc__)
|
|
|
|
@classmethod
|
|
def check_lazy_status(cls, Expression, V, F, length_mismatch="tile"):
|
|
|
|
input_stream = InputStream(Expression)
|
|
lexer = MathExprLexer(input_stream)
|
|
stream = CommonTokenStream(lexer)
|
|
stream.fill()
|
|
|
|
# Support aliases
|
|
aliases_img = {"a": "V0", "b": "V1", "c": "V2", "d": "V3"}
|
|
aliases_flt = {"w": "F0", "x": "F1", "y": "F2", "z": "F3"}
|
|
|
|
needed = []
|
|
needed1 = []
|
|
for token in filter(lambda t: t.type == MathExprParser.VARIABLE, stream.tokens):
|
|
var_name = token.text
|
|
|
|
if re.match(r"[VF][0-9]+", var_name):
|
|
needed.append(var_name)
|
|
elif var_name in aliases_img:
|
|
needed.append(aliases_img[var_name])
|
|
elif var_name in aliases_flt:
|
|
needed.append(aliases_flt[var_name])
|
|
for v in needed:
|
|
if v.startswith("V"):
|
|
if v not in V or V[v] is None:
|
|
needed1.append(v)
|
|
elif v.startswith("F"):
|
|
if v not in F or F[v] is None:
|
|
needed1.append(v)
|
|
return needed1
|
|
|
|
@classmethod
|
|
def execute(cls, V, F, Expression, length_mismatch="tile") -> io.NodeOutput:
|
|
# Determine reference model for cloning
|
|
a = V.get("V0")
|
|
if a is None:
|
|
# Try finding first valid model
|
|
for m in V.values():
|
|
if m is not None:
|
|
a = m
|
|
break
|
|
if a is None:
|
|
raise ValueError("At least one input model is required.")
|
|
|
|
# Prepare variables
|
|
# V0..V3 map to a..d for backward compatibility in calculate_patches
|
|
b = V.get("V1")
|
|
c = V.get("V2")
|
|
d = V.get("V3")
|
|
w = F.get("F0", 0.0) if F.get("F0") is not None else 0.0
|
|
x = F.get("F1", 0.0) if F.get("F1") is not None else 0.0
|
|
y = F.get("F2", 0.0) if F.get("F2") is not None else 0.0
|
|
z = F.get("F3", 0.0) if F.get("F3") is not None else 0.0
|
|
|
|
# Note: calculate_patches usually takes specific args. We might need to update it to support dynamic V/F or just pass everything.
|
|
# Looking at Step 34, calculate_patches signature: (Model, a, b, c, d, w, x, y, z)
|
|
# We need to verify if calculate_patches handles V/F. It probably doesn't.
|
|
# We should check 'modelLikeCommon.py' to see if update is needed.
|
|
|
|
# Assume for now we pass a,b,c,d,w,x,y,z as standard.
|
|
# But for full autogrow support (more than 4 inputs), calculate_patches needs update.
|
|
# The prompt didn't explicitly ask to update modelLikeCommon, but "switch to Autogrow" implies full functionality.
|
|
# I'll check modelLikeCommon.py after this block.
|
|
# For now, I will pass V and F to calculate_patches if I modify it, or I will stick to legacy args if I don't modify it.
|
|
# However, to support V4+, I MUST modify calculate_patches.
|
|
|
|
# Let's pass the V and F dicts to a modified calculate_patches, or overload it.
|
|
# I will update modelLikeCommon.py as part of this task.
|
|
|
|
from .modelLikeCommon import calculate_patches_autogrow
|
|
|
|
# Map inputs to patchers if needed (Model.Input gives Model wrapper, need state_dict source?)
|
|
# ModelMathNode inputs are Model wrappers (comfy.model_patcher.ModelPatcher).
|
|
# So V items are ready to be used.
|
|
|
|
aliases = {"a": "V0", "b": "V1", "c": "V2", "d": "V3", "w": "F0", "x": "F1", "y": "F2", "z": "F3"}
|
|
patches = calculate_patches_autogrow(Expression, V=V, F=F, mapping=aliases)
|
|
|
|
out_model = a.clone()
|
|
if patches:
|
|
out_model.add_patches(patches, 1.0, 1.0)
|
|
return (out_model,)
|