Repository navigation
Conversation
Replace tools/anima_comfy.py, which ran ComfyUI (GPL) to produce the Anima test tensors, with tools/anima_reference.py, which runs diffusers' own AnimaTextConditioner, CosmosTransformer3DModel and AutoencoderKLQwenImage (Apache-2.0) on the published weights with strict loading. The exporter's -comfyui flag is gone; -test_output runs from a venv built from the new tools/requirements-anima.txt (diffusers 0.41, transformers 5.19). Every ComfyUI citation, quoted comment and code snippet is removed from docs/anima.md, the add-anima-model skill, tools/requirements.txt and the comments in waifu/src/anima/ and waifu/src/sdxl/sampler.rs. The Rust Anima code was measured against diffusers on Anima Turbo v1.1: text encoder 0.0039, adapter 0.0025, one DiT step 0.012, VAE decode 0.00069 relative RMSE. diffusers clamps a decode to [-1, 1] and the runtime does not, so the decode test now clamps the runtime's picture first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ling0322
changed the title
ling0322
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Removes all code and text in the repository that references ComfyUI (GPL), and rewrites what used it.
tools/anima_comfy.py, which ran ComfyUI to compute Anima's test tensors, is deleted and replaced bytools/anima_reference.py, a new MIT script that computes the same tensors with Apache-2.0 code.anima_exporter.pyloses-comfyui, and the ComfyUI citations, quoted comments and code snippets indocs/anima.md, theadd-anima-modelskill,tools/requirements.txtand the comments inwaifu/are rewritten.Why
libwaifu is MIT and must not depend on or port from copyleft code. Anima's test tensors were computed by importing ComfyUI at run time, and
docs/anima.mdquoted its source. A case-insensitive search for "comfy" across the tree now finds nothing.What changed
tools/anima_reference.pyloads the published weights with strict name checking, so a wrong tensor name is an error rather than a randomly initialized layer. It needs its own venv,tools/requirements-anima.txt.waifu/tests/anima.rs: the decode test clamps the runtime's picture to [-1, 1] before comparing, since the new test tensors are clamped and the runtime leaves that to pixel conversion.waifu/src/sdxl/sampler.rs: the test named after ComfyUI is nowthe_schedule_is_the_floor_of_the_steps_over_the_strength, with unchanged behaviour.Measured
Relative RMSE of the Rust runtime against the new test tensors, Anima Turbo v1.1, on the code before this change (
de01833) and after:anima/anima_pipeline/anima_tokenizerA plain export from the old and new exporter is byte-identical: weights, manifest and both tokenizers.
./build/unittestandcargo testpass.Still to know
anima-turbo-v11_test.safetensorsfixture must be regenerated from.venv-anima(see the skill).🤖 Generated with Claude Code