This PR adds the block_1x128_128x128, block_128x128_1x128 and block_1x128_1x128 for scaled_mm_v2 for xpu.
It does the following:
- Adds the scaled_mm v2 frontend API in ScaledBlas.cpp
- Adds the oneDNN backend in QMatMul.cpp
The oneDNN fp8 blockwise supports relies on the oneDNN > v3.11, a dedicated oneDNN upgrade is needed for the functionalit and is not included in this PR. However, it doesn't affect the overall build. The full feature will be enabled only after oneDNN upgrade is done.
PR Stack:
- : Contains the scaled_mm v1 version.
Approved by:↗ Original-Artikel auf github.com lesenVollständiger Original-BerichtAusführliche Details, Code-Beispiele & Hersteller-Stellungnahme auf github.com.
Community-Analysen & Experten-Meinungen 0
Verwandte Story-Cluster & Quellen (Vektor-KI)
Ähnliche Beiträge
Auch interessante Nachrichten trunk/500cfbec122d9a3d5500e96a09817c7005ddb44e: [xpu][feat][1/3] Add blockwise FP8 support for scaled_mm_v2 (#173630)
Thematisch verwandte Begriffe: trunk500cfbec122d9a3d5500e96a09817c7005ddb44e, xpufeat13, blockwise, support · 6 Treffer
Apple accuses OpenAI of destroying evidence as trade-secrets fight intensifies
GPT-6 Astra Release Today? OpenAI’s Next Major AI Model Is Almost Here
Videos werden geladen ...
Beiträge werden geladen ...
Videos werden geladen ...
Beiträge werden geladen ...
Videos werden geladen ...
Beiträge werden geladen ...
Videos werden geladen ...
Beiträge werden geladen ...
Videos werden geladen ...
SOCIAL SHARE CARD GENERATOR