From 5809c626c9aa4f70c9fc72af45929c8c9089ae80 Mon Sep 17 00:00:00 2001 From: ivanshan_8170 Date: Tue, 19 Aug 2025 10:55:38 +0800 Subject: [PATCH 1/2] fix: mla ring add nzcache --- .../multi_latent_attention_operation.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp b/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp index 02949a59..d5202dc4 100644 --- a/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp +++ b/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp @@ -109,8 +109,8 @@ static bool ParamCheck(const infer::MultiLatentAttentionParam &opParam) if ((opParam.calcType == infer::MultiLatentAttentionParam::CalcType::CALC_TYPE_RING || opParam.calcType == infer::MultiLatentAttentionParam::CalcType::CALC_TYPE_SPEC_AND_RING) && (opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::KROPE_CTKV && - opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::INT8_NZCACHE)) { - ATB_LOG(ERROR) << "CalcType is ring only support krppe ctkv and int8 nzcache"; + opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::INT8_NZCACHE && opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::NZCACHE)) { + ATB_LOG(ERROR) << "CalcType is ring only support krope, ctkv, int8 nzcache and nzcache"; return false; } if ((opParam.cacheMode == infer::MultiLatentAttentionParam::CacheMode::INT8_NZCACHE) && -- Gitee From 71ff7dd5525c5cb689e041c1a079ca6997418e07 Mon Sep 17 00:00:00 2001 From: ivanshan_8170 Date: Tue, 19 Aug 2025 11:04:59 +0800 Subject: [PATCH 2/2] style: cleancode --- .../multi_latent_attention_operation.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp b/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp index d5202dc4..3eaf2b3e 100644 --- a/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp +++ b/src/ops_infer/multi_latent_attention/multi_latent_attention_operation.cpp @@ -109,7 +109,8 @@ static bool ParamCheck(const infer::MultiLatentAttentionParam &opParam) if ((opParam.calcType == infer::MultiLatentAttentionParam::CalcType::CALC_TYPE_RING || opParam.calcType == infer::MultiLatentAttentionParam::CalcType::CALC_TYPE_SPEC_AND_RING) && (opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::KROPE_CTKV && - opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::INT8_NZCACHE && opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::NZCACHE)) { + opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::INT8_NZCACHE && + opParam.cacheMode != infer::MultiLatentAttentionParam::CacheMode::NZCACHE)) { ATB_LOG(ERROR) << "CalcType is ring only support krope, ctkv, int8 nzcache and nzcache"; return false; } -- Gitee