Transformer类接口
【免费下载链接】ops-transformer本项目是CANN提供的transformer类大模型算子库,实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-transformer
- Transformer类aclnn接口列表
- aclnnAllGatherAdd
- aclnnAllGatherMatmul
- aclnnAllGatherMatmulV2
- aclnnAlltoAllAllGatherBatchMatMul
- aclnnAlltoAllMatmul
- aclnnAlltoAllMatmulV2
- aclnnAlltoAllQuantMatmul
- aclnnAlltoAllQuantMatmulV2
- aclnnAlltoAllvGroupedMatMul
- aclnnAlltoAllvGroupedMatMulV2
- aclnnAlltoAllvQuantGroupedMatMul
- aclnnAlltoAllvQuantGroupedMatMulV2
- aclnnApplyRotaryPosEmb
- aclnnApplyRotaryPosEmbV2
- aclnnAttentionToFFN
- aclnnAttentionUpdate
- aclnnAttentionWorkerCombine
- aclnnBSASelectBlockMask
- aclnnBatchMatMulReduceScatterAlltoAll
- aclnnBlockSparseAttention
- aclnnBlockSparseAttentionGrad
- aclnnBlockSparseAttentionV2
- aclnnCausalConv1dFn
- aclnnCausalConv1dUpdate
- aclnnChunkGatedDeltaRule
- aclnnCompressor
- aclnnDenseLightningIndexerGradKLLoss
- aclnnDenseLightningIndexerSoftmaxLse
- aclnnDequantRopeQuantKvcache
- aclnnDistributeBarrier
- aclnnDistributeBarrierV2
- aclnnFFN
- aclnnFFNToAttention
- aclnnFFNV2
- aclnnFFNV3
- aclnnFfnWorkerBatching
- aclnnFlashAttentionScore
- aclnnFlashAttentionScoreGrad
- aclnnFlashAttentionScoreGradV2
- aclnnFlashAttentionScoreGradV3
- aclnnFlashAttentionScoreGradV4
- aclnnFlashAttentionScoreV2
- aclnnFlashAttentionScoreV3
- aclnnFlashAttentionScoreV4
- aclnnFlashAttentionUnpaddingScoreGrad
- aclnnFlashAttentionUnpaddingScoreGradV2
- aclnnFlashAttentionUnpaddingScoreGradV3
- aclnnFlashAttentionUnpaddingScoreGradV4
- aclnnFlashAttentionUnpaddingScoreGradV5
- aclnnFlashAttentionVarLenScore
- aclnnFlashAttentionVarLenScoreV2
- aclnnFlashAttentionVarLenScoreV3
- aclnnFlashAttentionVarLenScoreV4
- aclnnFlashAttentionVarLenScoreV5
- aclnnFusedCausalConv1d
- aclnnFusedFloydAttention
- aclnnFusedFloydAttentionGrad
- aclnnFusedInferAttentionScore
- aclnnFusedInferAttentionScoreV2
- aclnnFusedInferAttentionScoreV3
- aclnnFusedInferAttentionScoreV4
- aclnnFusedInferAttentionScoreV5
- aclnnGatherPaKvCache
- aclnnGroupedMatMulAllReduce
- aclnnGroupedMatMulAlltoAllv
- aclnnGroupedMatMulAlltoAllvV2
- aclnnGroupedMatmul
- aclnnGroupedMatmulActivationQuantWeightNz
- aclnnGroupedMatmulAdd
- aclnnGroupedMatmulAddV2
- aclnnGroupedMatmulFinalizeRouting
- aclnnGroupedMatmulFinalizeRoutingV2
- aclnnGroupedMatmulFinalizeRoutingV3
- aclnnGroupedMatmulFinalizeRoutingWeightNz
- aclnnGroupedMatmulFinalizeRoutingWeightNzV2
- aclnnGroupedMatmulSwigluQuant
- aclnnGroupedMatmulSwigluQuantV2
- aclnnGroupedMatmulSwigluQuantWeightNZ
- aclnnGroupedMatmulSwigluQuantWeightNzV2
- aclnnGroupedMatmulV2
- aclnnGroupedMatmulV3
- aclnnGroupedMatmulV4
- aclnnGroupedMatmulV5
- aclnnGroupedMatmulWeightNz
- aclnnIncreFlashAttention
- aclnnIncreFlashAttentionV2
- aclnnIncreFlashAttentionV3
- aclnnIncreFlashAttentionV4
- aclnnIndexerQuantCache
- aclnnInplaceAttentionWorkerScheduler
- aclnnInplaceFfnWorkerScheduler
- aclnnInplaceFusedCausalConv1d
- aclnnInplaceMatmulAllReduceAddRmsNorm
- aclnnInplacePartialRotaryMul
- aclnnInplacePartialRotaryMulGrad
- aclnnInplaceQuantMatmulAllReduceAddRmsNorm
- aclnnInplaceWeightQuantMatmulAllReduceAddRmsNorm
- aclnnInterleaveRope
- aclnnKvCompressEpilog
- aclnnKvRmsNormRopeCache
- aclnnKvRmsNormRopeCacheV2
- aclnnLightningIndexer
- aclnnLightningIndexerGrad
- aclnnLightningIndexerV2
- aclnnLightningIndexerV2Metadata
- aclnnMaskedCausalConv1d
- aclnnMaskedCausalConv1dBackward
- aclnnMatmulAllReduce
- aclnnMatmulAllReduceAddRmsNorm
- aclnnMatmulAllReduceV2
- aclnnMatmulAllReduceV3
- aclnnMatmulAlltoAll
- aclnnMatmulAlltoAllV2
- aclnnMatmulReduceScatter
- aclnnMatmulReduceScatterV2
- aclnnMhcPost
- aclnnMhcPostBackward
- aclnnMhcPre
- aclnnMhcPreBackward
- aclnnMhcPreSinkhorn
- aclnnMhcPreSinkhornBackward
- aclnnMhcSinkhorn
- aclnnMhcSinkhornBackward
- aclnnMixedQuantSparseFlashMla
- aclnnMixedQuantSparseFlashMlaMetadata
- aclnnMlaPreprocess
- aclnnMlaPreprocessV2
- aclnnMlaProlog
- aclnnMlaPrologV2WeightNz
- aclnnMlaPrologV3WeightNz
- aclnnMoeComputeExpertTokens
- aclnnMoeDistributeCombine
- aclnnMoeDistributeCombineAddRmsNorm
- aclnnMoeDistributeCombineAddRmsNormV2
- aclnnMoeDistributeCombineSetup
- aclnnMoeDistributeCombineTeardown
- aclnnMoeDistributeCombineV2
- aclnnMoeDistributeCombineV3
- aclnnMoeDistributeCombineV4
- aclnnMoeDistributeDispatch
- aclnnMoeDistributeDispatchSetup
- aclnnMoeDistributeDispatchTeardown
- aclnnMoeDistributeDispatchV2
- aclnnMoeDistributeDispatchV3
- aclnnMoeDistributeDispatchV4
- aclnnMoeFinalizeRouting
- aclnnMoeFinalizeRoutingV2
- aclnnMoeFinalizeRoutingV2Grad
- aclnnMoeFinalizeRoutingV3
- aclnnMoeFinalizeRoutingV4
- aclnnMoeFusedTopk
- aclnnMoeGatingTopK
- aclnnMoeGatingTopKBackward
- aclnnMoeGatingTopKSoftmax
- aclnnMoeGatingTopKSoftmaxV2
- aclnnMoeGatingTopKV2
- aclnnMoeInitRouting
- aclnnMoeInitRoutingQuant
- aclnnMoeInitRoutingQuantV2
- aclnnMoeInitRoutingV2
- aclnnMoeInitRoutingV2Grad
- aclnnMoeInitRoutingV3
- aclnnMoeTokenPermute
- aclnnMoeTokenPermuteGrad
- aclnnMoeTokenPermuteV2
- aclnnMoeTokenPermuteWithEp
- aclnnMoeTokenPermuteWithEpGrad
- aclnnMoeTokenPermuteWithRoutingMap
- aclnnMoeTokenPermuteWithRoutingMapGrad
- aclnnMoeTokenUnpermute
- aclnnMoeTokenUnpermuteGrad
- aclnnMoeTokenUnpermuteWithEp
- aclnnMoeTokenUnpermuteWithEpGrad
- aclnnMoeTokenUnpermuteWithRoutingMap
- aclnnMoeTokenUnpermuteWithRoutingMapGrad
- aclnnMoeUpdateExpert
- aclnnNormRopeConcat
- aclnnNormRopeConcatBackward
- aclnnNsaCompress
- aclnnNsaCompressAttention
- aclnnNsaCompressAttentionInfer
- aclnnNsaCompressGrad
- aclnnNsaCompressWithCache
- aclnnNsaSelectedAttention
- aclnnNsaSelectedAttentionGrad
- aclnnNsaSelectedAttentionInfer
- aclnnPromptFlashAttention
- aclnnPromptFlashAttentionV2
- aclnnPromptFlashAttentionV3
- aclnnQkvRmsNormRopeCache
- aclnnQkvRmsNormRopeCacheWithKScale
- aclnnQuantAllReduce
- aclnnQuantFlashAttentionScore
- aclnnQuantGroupedMatMulAlltoAllv
- aclnnQuantGroupedMatMulAlltoAllvV2
- aclnnQuantGroupedMatmulDequant
- aclnnQuantGroupedMatmulDequantWeightNZ
- aclnnQuantGroupedMatmulInplaceAdd
- aclnnQuantLightningIndexer
- aclnnQuantLightningIndexerV2Metadata
- aclnnQuantMatmulAllReduce
- aclnnQuantMatmulAllReduceAddRmsNorm
- aclnnQuantMatmulAllReduceV2
- aclnnQuantMatmulAllReduceV3
- aclnnQuantMatmulAllReduceV4
- aclnnQuantMatmulAllReduceV5
- aclnnQuantMatmulAlltoAll
- aclnnQuantMatmulAlltoAllV2
- aclnnQuantReduceScatter
- aclnnRainFusionAttention
- aclnnRecurrentGatedDeltaRule
- aclnnRingAttentionUpdate
- aclnnRingAttentionUpdateV2
- aclnnRopeWithSinCosCache
- aclnnRopeWithSinCosCacheV2
- aclnnRotaryPositionEmbedding
- aclnnRotaryPositionEmbeddingGrad
- aclnnRotaryPositionEmbeddingV2
- aclnnScatterPaCache
- aclnnScatterPaKvCache
- aclnnScatterPaKvCacheWithKScale
- aclnnSparseFlashAttention
- aclnnSparseFlashAttentionGrad
- aclnnSparseFlashMla
- aclnnSparseFlashMlaGrad
- aclnnSparseFlashMlaGradMetadata
- aclnnSparseFlashMlaMetadata
- aclnnSparseLightningIndexerGradKLLoss
- aclnnSparseLightningIndexerKLLossGrad
- aclnnSparseLightningIndexerKLLossGradMetadata
- aclnnSwinAttentionScoreQuant
- aclnnSwinTransformerLnQkvQuant
- aclnnWeightQuantMatmulAllReduce
- aclnnWeightQuantMatmulAllReduceAddRmsNorm
- aclnnWeightQuantMatmulAllReduceV2
- 编译与运行样例
- aclnn返回码
【免费下载链接】ops-transformer本项目是CANN提供的transformer类大模型算子库,实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-transformer
创作声明:本文部分内容由AI辅助生成(AIGC),仅供参考