Skip to content

Revert "perf(deepseek): reduce fused MTP decode host transfers" - #186

Open
ndleslx wants to merge 1 commit into
hw-native-sys:mainfrom
ndleslx:revert/pr144-fused-mtp-decode
Open

Revert "perf(deepseek): reduce fused MTP decode host transfers"#186
ndleslx wants to merge 1 commit into
hw-native-sys:mainfrom
ndleslx:revert/pr144-fused-mtp-decode

Conversation

@ndleslx

@ndleslx ndleslx commented Aug 24, 2026

Copy link
Copy Markdown
Collaborator

Reverts #144.

#144 (perf(deepseek): reduce fused MTP decode host transfers, commit
0ae4fd85) broke the DeepSeek V4 accuracy guard on main:

  • Its merge run
    (32468071303)
    failed unit-tests on
    tests/test_deepseek_v4_http_completion_matches_expected_text[k1-prefix-cache]
    — an accuracy mismatch, the generated text diverging from the expected
    completion mid-generation.
  • The previous merge on main (feat(skill): add aligned DSV4 GBS32 profiler #170,
    32462006878)
    was green 80 minutes earlier, and nothing landed in between.
  • Both runs resolved the identical pypto revision — pypto under test: 5b15048e836c574238acde135466740c10bf5239 appears in each log — so ptoas and
    the pto-isa commit (read from that same checkout) were the same as well. The
    toolchain is held constant; the only delta between green and red is perf(deepseek): reduce fused MTP decode host transfers #144
    itself.

That is the path #144 changed: the fused MTP decode in
npu_runner.py / task_args.py now keeps outputs device-resident, and the
k1-prefix-cache variant is exactly the configuration that exercises it with a
cached prefix.

Verified locally: the reverted tree passes all 82 tests in
tests/unit/model/deepseek/test_model_components.py, plus the repo's lint
checks. The full unit-tests job — including that DeepSeek guard — runs on this
PR and should confirm main goes green again.

The perf idea is worth relanding once it can pass the accuracy guard.

@coderabbitai

coderabbitai Bot commented Aug 24, 2026

Copy link
Copy Markdown

Important

  • 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 912a5caa-4bdd-4ae2-b745-f3b8a63f492b


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

This reverts commit 0ae4fd85e00ea4f89f6eed52ea37cbc3b0eb0f168,
re-resolved against current main so hw-native-sys#132's chunked-prefill and
K=1/K>1 MTP ABI changes are kept while hw-native-sys#144's device-residency is
removed: TaskArgs slots are host-shared again, the decode-metadata
residency machinery (_materialize_decode_device_metadata and friends)
is gone, and hw-native-sys#144's two device-residency tests are dropped.

The change broke the DeepSeek V4 accuracy guard on main: the merge run
of hw-native-sys#144 failed test_deepseek_v4_http_completion_matches_expected_text
[k1-prefix-cache] with a generated-text mismatch, while the previous
merge run on main (hw-native-sys#170, the same morning) was green — both runs on the
identical pypto revision. Restore the pre-hw-native-sys#144 decode path until the
transfer reduction lands with the guard passing.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant