Update: set fmha, and context_qk_half_accumulation as optional

NVIDIA · Feb 6, 2023 · 7921074 · 7921074
1 parent 90cc5a7
commit 7921074
Show file tree

Hide file tree

Showing 95 changed files with 1,440,710 additions and 1,342,718 deletions.
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_128_sm89.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_144_sm89.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm70.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm70.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm72.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm72.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm75.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm75.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm80.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm80.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm86.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm86.cubin.cpp
diff --git a/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm89.cubin.cpp b/...ty/trt_fused_multihead_attention/fmha_v2_flash_attention_fp16_Causal_0_160_sm89.cubin.cpp