Repository navigation
feat: cache_dit warm-up steps flag and per-request low-noise guidance - #58
Open
Johan-de-R wants to merge 1 commit into
Open
Johan-de-R wants to merge 1 commit into
Johan-de-R wants to merge 1 commit into
Conversation
--cache-dit-max-warmup-steps sets how many leading steps cache_dit always computes before reusing (on Wan 2.2 A14B, the high-noise expert's). Under Wan's official 40-step/shift-12 schedule the default 4 lets the cache reuse layout-deciding steps (extra objects, broken limbs in review); 8 restores uncached-level fidelity at 193 s/clip on 1xB200. The video handler now takes guidance_scale_2 from the request's nvext, which the frontend already carries, falling back to the config default as before. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Johan-de-R
had a problem deploying
to
external_collaborator
October 23, 2026 01:14 — with 
GitHub Actions
Failure
GitHub Actions
Failure
This branch had an error being deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
--cache-dit-max-warmup-steps(envDYN_TRTLLM_CACHE_DIT_MAX_WARMUP_STEPS): passed toCacheDiTConfig(max_warmup_steps=…), the leading steps cache_dit always computes before it starts reusing. On Wan 2.2 A14B this is the high-noise expert's warm-up (the low-noise expert keeps VisualGen's own). Unset keeps VisualGen's default (4).guidance_scale_2: the video handler now reads it from the request'snvext(the frontend already carries the field) and falls back to--default-guidance-scale-2as before.guidance_scalealready worked this way.Why
Under Wan's official A14B settings (40 steps, shift 12) the high-noise expert gets 26 of 40 steps, and with the default warm-up of 4 cache_dit reuses the steps that decide layout. In review that produced an extra milk jug, extra hands and impossible legs. LPIPS against the uncached render, 1xB200, 720p/81 frames:
8 is the chosen setting; reviewed by eye. It holds at 27 steps too (160 s/clip, 0.15 vs uncached 27 steps). One server alternating 27/40/50-step requests ran each at its steady-state time, so the step count does not trigger recompiles.
Per-request dual guidance lets deepapi expose both guidance values (e.g. Artificial Analysis sends 3.5 / 4).
Tests
TestCacheDitOverridescovers the new override;test_request_guidance_overrides_config_defaultscovers the handler. 74/74 pass in the…-swenc-ea1a31baimage, including the async handler tests, which needpytest-asyncio(the image doesn't ship it, so these tests didn't run before). Mutation-checked: removing the override line, or restoring the config-onlyguidance_scale_2, each fails exactly its test.Not enabled by this PR; the Wan config gets
--cache-dit-max-warmup-steps=8with the rollout.🤖 Generated with Claude Code