feat(hunyuan-image3): add TI2I and T2I hybrid parallel entry - #1355
Open
Chernobyllight wants to merge 2 commits into
Open
feat(hunyuan-image3): add TI2I and T2I hybrid parallel entry#1355Chernobyllight wants to merge 2 commits into
Chernobyllight wants to merge 2 commits into
Conversation
Run autoregressive decoding with TP4 and denoising with TP2+SP2 over one TP2-loaded ABAB weight layout. Add a Hopper BF16 grouped multi-micro MoE operator that reuses routing across two zero-copy micro-shard views, avoids per-micro FlashInfer calls during denoising, and preserves a single TP2 reduction without phase-time weight movement. Include strict topology validation, a production launcher, and regression coverage for weight views, phase dispatch, and four-rank NCCL groups.
Add a four-GPU TI2I launch path that loads the official checkpoint once in a TP2 ABAB layout, runs autoregressive generation with TP4, and switches denoising to TP2+SP2 through the multi-micro fused MoE path without moving weights between phases. Preserve reference-image VAE and vision conditions across AR and denoising, use serial CFG with KV-cache reuse, and add focused regression coverage for condition replication, branch slicing, and first-step cache injection.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds a four-GPU TI2I and T2I hybrid-parallel path that loads the official checkpoint once in a TP2 ABAB layout, runs autoregressive generation with TP4, and switches denoising to TP2+SP2 through the multi-micro fused MoE path without moving weights between phases. Includes regression coverage for condition reuse, serial CFG slicing, and KV-cache injection.