Add 5Hz LM planner (CoT-only) and generation-daemon integration#3
Open
Add 5Hz LM planner (CoT-only) and generation-daemon integration#3
Conversation
Wraps acestep-5Hz-lm-1.7B as a Qwen3 causal LM that runs before the DiT. Generates until </think>, parses the YAML metadata block, and returns structured BPM, key/scale, time signature, language, and a rewritten caption to use as DiT conditioning instead of the raw user input.
Add --use-lm flag to load the LM planner on startup and run it before each generation request. The LM output fills in BPM, key/scale, time signature, and language; user-specified duration_s always takes priority over the LM suggestion. The VRAM offload threshold is automatically lowered when the LM is resident to avoid spurious CPU fallback. Also bind the socket before model loading so the file appears within ~1s of startup rather than after the full 30s load time.
3e8c6e4 to
e3b3e1e
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Loads
acestep-5Hz-lm-1.7Bbefore the DiT to expand a raw user caption into structured metadata — BPM, key/scale, time signature, language, and a rewritten caption — which the DiT text encoder then conditions on instead of the raw input.The LM runs in CoT-only mode (stops at
</think>). Phase 2 audio code generation is tracked in #2.The generation daemon gets a
--use-lmflag. The Unix socket is now bound before model loading so it appears within ~1s of startup.