Support BF16 for low-latency kernels

This commit is contained in:
Chenggang Zhao
2025-03-10 17:24:41 +08:00
parent 1fc40d50f3
commit ed7487c15e
8 changed files with 138 additions and 111 deletions

View File

@@ -128,7 +128,7 @@ struct LowLatencyLayout {
// Message sizes
EP_HOST_ASSERT(num_scales * sizeof(float) <= hidden);
size_t num_bytes_per_dispatch_msg = hidden + num_scales * sizeof(float) + sizeof(int4);
size_t num_bytes_per_dispatch_msg = sizeof(int4) + std::max(hidden * sizeof(nv_bfloat16), hidden + num_scales * sizeof(float));
size_t num_bytes_per_combine_msg = sizeof(int4) + hidden * sizeof(nv_bfloat16);
// Send buffer