[ATOM-vLLM][Qwen3.5][Qwen3Next][GDN] add packed decode fast path for GDN#508
Closed
zejunchen-zejun wants to merge 2 commits intomainfrom
Closed
[ATOM-vLLM][Qwen3.5][Qwen3Next][GDN] add packed decode fast path for GDN#508zejunchen-zejun wants to merge 2 commits intomainfrom
zejunchen-zejun wants to merge 2 commits intomainfrom
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds a packed recurrent decode “fast path” for the GDN attention backend in the vLLM OOT plugin, and enables it for the pure non-speculative decode case (targeting Qwen3.5/Qwen3Next on vLLM 0.19.0).
Changes:
- Introduces a non-spec decode helper (
_forward_core_decode_non_spec) that usesfused_recurrent_gated_delta_rule_packed_decode. - Gates the packed-decode path behind
vllm_envs.VLLM_ENABLE_FLA_PACKED_RECURRENT_DECODEand runtime metadata conditions (no spec masks, decode-only). - Updates KV cache access to use
compilation_config[layer_name].kv_cachedirectly (removingvirtual_engineindexing).
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
zejunchen-zejun
changed the title
fast path, enable the pure non-spec decode GDN fast path into the OOT plugin backend for Qwen3.5/Qwen3Next Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
zejunchen-zejun
force-pushed
the
zejun/vLLM_19_qwen3.5_fast_path_0407
branch
from
799dc44 to
9583056
Compare
zejunchen-zejun
changed the title
zejunchen-zejun
changed the title
Contributor
Author
|
No performance benefit |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
have a packed recurrent attention triton kernel for fast path
Qwen3-Next-80B-A3B-Instruct-FP8 TP4 1K/8K Concurrency 4 TTPS + 1.6%
Qwen3-Next-80B-A3B-Instruct-FP8 TP4 1K/8K Concurrency 32 TTPS + 1.13%