Zum Hauptinhalt springen
IT Security NachrichtenEY Survey Finds Autonomous AI Implementation Outpaces Oversight(18.09.2026 um 21:23 Uhr)
IT Security NachrichtenUnser erster Eindruck von Apples iPhone Duo - Golem.de(18.09.2026 um 20:47 Uhr)
IT Security NachrichtenAfter your iOS 27 upgrade, do these 5 things first(18.09.2026 um 21:01 Uhr)
AI & KI NachrichtenGitHub Release: openai/codex vrust-v0.156.0-alpha.4 (18.09.2026)(18.09.2026 um 22:09 Uhr)
IT NachrichtenAnthropic Pursues IPO Despite Its A.I. Safety Warnings(18.09.2026 um 22:06 Uhr)
IT Security NachrichtenEY Survey Finds Autonomous AI Implementation Outpaces Oversight(18.09.2026 um 21:23 Uhr)
IT Security NachrichtenUnser erster Eindruck von Apples iPhone Duo - Golem.de(18.09.2026 um 20:47 Uhr)
IT Security NachrichtenAfter your iOS 27 upgrade, do these 5 things first(18.09.2026 um 21:01 Uhr)
AI & KI NachrichtenGitHub Release: openai/codex vrust-v0.156.0-alpha.4 (18.09.2026)(18.09.2026 um 22:09 Uhr)
IT NachrichtenAnthropic Pursues IPO Despite Its A.I. Safety Warnings(18.09.2026 um 22:06 Uhr)
Intelligence View
⚡ tsecurity.de Intelligence

ciflow/torchtitan/183838: [Inductor][HOP] Handle unbacked FlexAttention predicates

EP-overlap graph chunking traces FlexAttention with unbacked symbolic batch and sequence dimensions. These are valid runtime sizes, but FlexAttention predicate sites forced trace-time decisions that should either fall back to the general kernel or remain represented as deferred symbolic assertions.

Flex decoding remains an optional optimized implementation. Its predicates use guard_or_false because they are kernel eligibility checks, not user-visible invariants: when ShapeEnv can prove eligibility, decode can be selected; when an unbacked predicate is not provable, the general FlexAttention kernel is used. Failing to prove an optimization must not impose a runtime shape contract on user inputs.

Backward fake propagation preserves the accepted KV-batch-broadcast metadata contract. It reduces grad_key and grad_value back to key/value batch when key/value is provably batch-broadcasted; otherwise it records the non-broadcast invariant with torch._check(Bq == Bkv) and returns unreduced metadata.

Public BlockMask length validation is expressed as two symbolic torch._check predicates: the mask must not be smaller than the query/key lengths, and it must not be larger than those lengths. Concrete mismatches still fail with the existing guidance, while unbacked symbolic lengths become deferred runtime assertions instead of forcing Python to branch on an unbacked expression.

This PR was authored with assistance from an AI assistant.

Test Plan:

python -m pytest test/inductor/test_flex_attention.py -q -s -k "unbacked_flex_decoding_eligibility_falls_back or backward_fake_symbolic_query_key_batch_non_broadcast or block_mask_vs_sequence_lengths"
python -m pytest test/inductor/test_flex_attention.py -q -s -k "mask_mod_handles_symint_addition or mask_mod_handles_derived_symint_closure or symbol_closure_in_score_mod"
lintrunner -a

stack-info: PR: #183838, branch: sanketpurandare/stack/12

Ähnliche Beiträge
🔍 Verwandte News

Auch interessante Nachrichten ciflow/torchtitan/183838: [Inductor][HOP] Handle unbacked FlexAttention predicates

Thematisch verwandte Begriffe: ciflowtorchtitan183838, InductorHOP, Handle, unbacked · 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 ...

Zum Aktualisieren ziehen
ZERO-DAY CVE-2026-61591 | djust provides Phoenix LiveView-style reactive server-side rendering for…
Advisory →
TTS Reader • tsecurity.de Voice
tsecurity.de Icon
tsecurity.de App
Offline-Lesen, Eilmeldungen & 0ms Ladezeit

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

Nächster Beitrag
Community Radar & Live Chat
Sentinel Bot online • Live-Stream
Dein Cluster: Security Explorer
Match:
lädt…
Verbindung zum Community-Stream wird aufgebaut...
Bearbeitungsmodus — Senden überschreibt deine Nachricht
Community-Puls — was gerade passiert
lädt…
Aktivitäten deiner Analysten
lädt…
Neues Thema oder Eilmeldung einreichen

Reiche interessante Links, Zero-Days oder Debatten ein. Die Community entscheidet per Upvote über die Veröffentlichung.

Heiß diskutierte Einreichungen
🔖 Gespeicherte Artikel
📂 Keine gespeicherten Artikel vorhanden.
Rechts: Artikel Ziehen Links: RSS
Hoch: nächster Artikel Runter: zurück / schließen
News NIS-2 Frühwarnung Tier-1 Intel ⏱️ 3 Min vor 10 Min
Artikeldaten werden geladen...

Rechts: Original Links: RSS-Ansicht
↗ Original-Quelle
Social Reaktionen Stimme abgeben (+5 Karma)
Einstufung & Relevanz-Poll 0 Stimmen
In sozialen Netzwerken teilen 1-Klick