解决bf16编译问题

e2d98ddc · zhouxiang · d7117b95 · e2d98ddc
Commit e2d98ddc authored Mar 23, 2024 by zhouxiang
Hide whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

src/turbomind/triton_backend/llama/LlamaTritonModel.cc src/turbomind/triton_backend/llama/LlamaTritonModel.cc +2 -0

No files found.
--- a/src/turbomind/triton_backend/llama/LlamaTritonModel.cc
+++ b/src/turbomind/triton_backend/llama/LlamaTritonModel.cc
@@ -47,6 +47,7 @@ std::shared_ptr<AbstractTransformerModel> AbstractTransformerModel::createLlamaM
            reader.GetInteger("ft_instance_hyperparameter", "enable_custom_all_reduce", 0),
            model_dir);
    }
+#ifdef ENABLE_BF16
    else if (data_type == "bf16") {
 #ifdef ENABLE_BF16
        return std::make_shared<LlamaTritonModel<__nv_bfloat16>>(
@@ -59,6 +60,7 @@ std::shared_ptr<AbstractTransformerModel> AbstractTransformerModel::createLlamaM
        ft::FT_CHECK(false);
 #endif
    }
+#endif
    else {
        return std::make_shared<LlamaTritonModel<float>>(
            reader.GetInteger("ft_instance_hyperparameter", "tensor_para_size"),