SolitaryThinker
4e331eb7ff
Mechanical conversion of every relative import under v2/ to an absolute v2.* path (from .x / ..x / ...x -> from v2.<pkg>.x) so imports are unambiguous, grep-able, and stable when code is copied/moved between entrypoints (VideoGenerator, CLI, server). Surgical prefix-only rewrite: only the 'from <dots><module>' prefix changed — import names, parentheses, multi-line formatting, comments, and ordering are byte-for-byte preserved (no collapsing, no reorder, no unrelated reformatting). - 471 imports across 125 files; v2/tests/ was already absolute (untouched). - Validated: all 125 files compile, every 'from v2.* import' target resolves to a real module/package, zero relative imports remain (full sweep), CPU mini suite green (202 passed, 2 skipped).
25 lines
1.2 KiB
Python
25 lines
1.2 KiB
Python
"""Tiled-decode program: text_encode → diffusion_denoise → vae_tile (design_v3 §4, §17).
|
|
|
|
The VAE decode is a ``ModelLoopNode`` (the ``vae_tile`` loop), not a one-shot ``ComponentNode`` — so its
|
|
tiles become scheduled ``VAE_TILE`` WorkUnits that interleave with other requests' ``DIFFUSION_STEP``
|
|
units. (Contrast the plain Wan program, where decode is a single in-tick node.)
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
from v2.program import ComponentNode, ModelLoopNode, Program, ProgramKind
|
|
from v2.models.common import text_encode_node_fn as _text_encode
|
|
|
|
|
|
def build_tiled_program() -> Program:
|
|
return Program(
|
|
program_id="wan.t2v.tiled", kind=ProgramKind.INLINE,
|
|
nodes=[
|
|
ComponentNode("text_encode", fn=_text_encode, writes=("text_embeds", "neg_text_embeds")),
|
|
ModelLoopNode("denoise", loop_id="diffusion_denoise", output_slot="denoise_out",
|
|
reads=("text_embeds",), writes=("denoise_out",)),
|
|
ModelLoopNode("vae_tile", loop_id="vae_tile", output_slot="tiled_out",
|
|
reads=("denoise_out",), writes=("tiled_out",)),
|
|
],
|
|
output_artifacts={"video": "tiled_out", "latents": "denoise_out"},
|
|
).validate()
|