Zum Hauptinhalt springen
Echtzeit-Radar & Feeds
Alle RSS Feeds ➔
👥 Community & Social
AI & KI NachrichtenWhy faster AI coding can mean harder engineering(29.09.2026 um 11:00 Uhr)
•••••
AI & KI NachrichtenCollege Students Flex Their Power in A.I. Investment Frenzy(29.09.2026 um 11:03 Uhr)
••
AI & KI NachrichtenBill Gates’s Blunt Warning on A.I.(29.09.2026 um 11:03 Uhr)
•
AI & KI NachrichtenEx-Tesla team raises $12.5M to put supply chains on autopilot(29.09.2026 um 11:00 Uhr)
••
AI & KI NachrichtenWhy faster AI coding can mean harder engineering(29.09.2026 um 11:00 Uhr)
•••••
AI & KI NachrichtenCollege Students Flex Their Power in A.I. Investment Frenzy(29.09.2026 um 11:03 Uhr)
••
AI & KI NachrichtenBill Gates’s Blunt Warning on A.I.(29.09.2026 um 11:03 Uhr)
•
AI & KI NachrichtenEx-Tesla team raises $12.5M to put supply chains on autopilot(29.09.2026 um 11:00 Uhr)
••
Intelligence View
⚡ tsecurity.de Intelligence

trunk/9455d3d34559799d09196341ca545d946610eb2e: Teach min-cut about online softmax prep (#185783)

The online softmax preparation pattern computes softmax max and sum as a multi-output prim. When this prim appears in an AOTAutograd joint graph, the min-cut partitioner previously did not know that it was a recomputable/fusible operation…

0
↗ Quelle (github.com)
Reagiere als Erste:r — dein Feedback zählt!

The online softmax preparation pattern computes softmax max and sum as a

multi-output prim. When this prim appears in an AOTAutograd joint graph, the

min-cut partitioner previously did not know that it was a recomputable/fusible

operation like the decomposed amax/sum pattern it replaces. That skewed the cut

and could make the forward save a full softmax activation instead of the much

smaller max/sum tuple outputs.


Add prims.prepare_softmax_online to the partitioner's recomputable op set. The

existing post-grad placement is still the normal path, but this makes the

partitioner robust if the online softmax preparation is introduced before

partitioning.


I considered relying only on the post-grad pattern placement that originally

avoided the regression, but that leaves the partitioner's cost model wrong for

this prim and makes future joint-graph uses fragile.


Benchmark Results:

A focused AOTAutograd min-cut reproducer that simulates joint-graph online

softmax placement saved 212,992 activation bytes before the fix and 149,504

after the fix, a 29.8% reduction. The old cut saved an extra full

[2, 128, 128] fp16 softmax activation; the new cut saves the small

[2, 128, 1] fp32 max/sum tuple outputs instead.


Test Plan:



  • python test/inductor/test_online_softmax.py -k test_prepare_softmax_after_partitioning

  • python test/inductor/test_online_softmax.py -k test_prepare_softmax

  • python ../state/148141/benchmark_partition_saved_bytes.py

  • lintrunner -a torch/_functorch/partitioners.py test/inductor/test_online_softmax.py

  • git diff --check --cached


Fixes #148141

Generated by my agent


Pull Request resolved: #185783

Approved by: https://github.com/shunting314

Ähnliche Beiträge
🔍 Verwandte News

Auch interessante Nachrichten trunk/9455d3d34559799d09196341ca545d946610eb2e: Teach min-cut about online softmax prep (#185783)

Thematisch verwandte Begriffe: trunk9455d3d34559799d09196341ca545d946610eb2e, Teach, mincut, about · 6 Treffer

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

💬 Kommentare werden geladen…
Zum Aktualisieren ziehen
ZERO-DAY CVE-2026-102240 | A vulnerability was found in Netcore NAP930 0.1.241010.141410. This aff…
Advisory →
tsecurity.de Icon
Offline-Lesen, Eilmeldungen & 0ms Ladezeit

Installiere tsecurity.de direkt auf deinen Home-Bildschirm für das ultimative Vollbild-Magazinerlebnis ohne Browser-Leisten.

Nächster Beitrag