From 2201642cb16a5454a045b2f6deef36820bd1ae19 Mon Sep 17 00:00:00 2001
From: sfc-gh-zhwang <flex.wang@snowflake.com>
Date: Wed, 19 Jul 2023 09:36:19 -0700
Subject: [PATCH] abc

---
 CMakeLists.txt              |  2 +-
 src/libfastertransformer.cc | 17 +++++++++++++++++
 2 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index e9c6c5c..d539453 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -110,7 +110,7 @@ if (EXISTS ${FT_DIR})
 else()
   FetchContent_Declare(
     repo-ft
-    GIT_REPOSITORY https://github.com/NVIDIA/FasterTransformer.git 
+    GIT_REPOSITORY https://github.com/neevaco/FasterTransformer.git 
     GIT_TAG main
     GIT_SHALLOW ON
   )
diff --git a/src/libfastertransformer.cc b/src/libfastertransformer.cc
index a870aa0..f2bbc0e 100644
--- a/src/libfastertransformer.cc
+++ b/src/libfastertransformer.cc
@@ -49,6 +49,7 @@
 
 // FT's libraries have dependency with triton's lib
 #include "src/fastertransformer/triton_backend/bert/BertTritonModel.h"
+#include "src/fastertransformer/triton_backend/deberta/DebertaTritonModel.h"
 #include "src/fastertransformer/triton_backend/gptj/GptJTritonModel.h"
 #include "src/fastertransformer/triton_backend/gptj/GptJTritonModelInstance.h"
 #include "src/fastertransformer/triton_backend/gptneox/GptNeoXTritonModel.h"
@@ -327,6 +328,22 @@ std::shared_ptr<AbstractTransformerModel> ModelState::ModelFactory(
     } else if (data_type == "bf16") {
       ft_model = std::make_shared<BertTritonModel<__nv_bfloat16>>(
             tp, pp, custom_ar, model_dir, int8_mode, is_sparse, remove_padding);
+#endif
+    }
+  } else if (model_type == "deberta") {
+    const int is_sparse      = param_get_bool(param,"is_sparse", false);
+    const int remove_padding = param_get_bool(param,"is_remove_padding", false);
+
+    if (data_type == "fp16") {
+      ft_model = std::make_shared<DebertaTritonModel<half>>(
+            tp, pp, custom_ar, model_dir, is_sparse, remove_padding);
+    } else if (data_type == "fp32") {
+      ft_model = std::make_shared<DebertaTritonModel<float>>(
+            tp, pp, custom_ar, model_dir, is_sparse, remove_padding);
+#ifdef ENABLE_BF16
+    } else if (data_type == "bf16") {
+      ft_model = std::make_shared<DebertaTritonModel<__nv_bfloat16>>(
+            tp, pp, custom_ar, model_dir, is_sparse, remove_padding);
 #endif
     }
   } else {