From 40274aac39f2542483906d92ec3b8014faf62912 Mon Sep 17 00:00:00 2001
From: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
Date: Thu, 26 Sep 2024 10:26:16 +0800
Subject: [PATCH] Bump version to `0.14.0.dev2024092401` (#2258)

---
 .../batch_manager/aarch64-linux-gnu/version.txt             | 2 +-
 .../x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.a | 4 ++--
 .../libtensorrt_llm_batch_manager_static.pre_cxx11.a        | 4 ++--
 cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/version.txt | 6 +++---
 .../tensorrt_llm_batch_manager_static.lib                   | 4 ++--
 .../batch_manager/x86_64-windows-msvc/version.txt           | 4 ++--
 .../aarch64-linux-gnu/libtensorrt_llm_executor_static.a     | 2 +-
 .../libtensorrt_llm_executor_static.pre_cxx11.a             | 2 +-
 cpp/tensorrt_llm/executor/aarch64-linux-gnu/version.txt     | 6 +++---
 .../x86_64-linux-gnu/libtensorrt_llm_executor_static.a      | 4 ++--
 .../libtensorrt_llm_executor_static.pre_cxx11.a             | 4 ++--
 cpp/tensorrt_llm/executor/x86_64-linux-gnu/version.txt      | 6 +++---
 .../x86_64-windows-msvc/tensorrt_llm_executor_static.lib    | 4 ++--
 cpp/tensorrt_llm/executor/x86_64-windows-msvc/version.txt   | 4 ++--
 .../nvrtcWrapper/aarch64-linux-gnu/version.txt              | 2 +-
 .../nvrtcWrapper/x86_64-linux-gnu/version.txt               | 2 +-
 .../x86_64-windows-msvc/tensorrt_llm_nvrtc_wrapper.dll      | 2 +-
 .../nvrtcWrapper/x86_64-windows-msvc/version.txt            | 4 ++--
 .../libtensorrt_llm_internal_cutlass_kernels_static.a       | 2 +-
 ...tensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a | 2 +-
 .../internal_cutlass_kernels/aarch64-linux-gnu/version.txt  | 6 +++---
 .../libtensorrt_llm_internal_cutlass_kernels_static.a       | 2 +-
 ...tensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a | 2 +-
 .../internal_cutlass_kernels/x86_64-linux-gnu/version.txt   | 6 +++---
 .../tensorrt_llm_internal_cutlass_kernels_static.lib        | 2 +-
 .../x86_64-windows-msvc/version.txt                         | 4 ++--
 examples/baichuan/requirements.txt                          | 2 +-
 examples/bloom/requirements.txt                             | 2 +-
 examples/chatglm/requirements.txt                           | 2 +-
 examples/dbrx/requirements.txt                              | 2 +-
 examples/falcon/requirements.txt                            | 2 +-
 examples/gemma/requirements.txt                             | 2 +-
 examples/gpt/requirements.txt                               | 2 +-
 examples/gptj/requirements.txt                              | 2 +-
 examples/gptneox/requirements.txt                           | 2 +-
 examples/grok/requirements.txt                              | 2 +-
 examples/internlm/requirements.txt                          | 2 +-
 examples/jais/requirements.txt                              | 2 +-
 examples/llama/requirements.txt                             | 2 +-
 examples/llm-api/requirements.txt                           | 2 +-
 examples/mamba/requirements.txt                             | 2 +-
 examples/medusa/requirements.txt                            | 2 +-
 examples/mixtral/requirements.txt                           | 2 +-
 examples/mpt/requirements.txt                               | 2 +-
 examples/nemotron/requirements.txt                          | 2 +-
 examples/opt/requirements.txt                               | 2 +-
 examples/phi/requirements.txt                               | 2 +-
 examples/quantization/requirements.txt                      | 2 +-
 examples/qwen/requirements.txt                              | 2 +-
 examples/qwenvl/requirements.txt                            | 2 +-
 examples/recurrentgemma/requirements.txt                    | 2 +-
 examples/redrafter/requirements.txt                         | 2 +-
 examples/skywork/requirements.txt                           | 2 +-
 examples/smaug/requirements.txt                             | 2 +-
 examples/whisper/requirements.txt                           | 2 +-
 tensorrt_llm/version.py                                     | 2 +-
 56 files changed, 76 insertions(+), 76 deletions(-)

diff --git a/cpp/tensorrt_llm/batch_manager/aarch64-linux-gnu/version.txt b/cpp/tensorrt_llm/batch_manager/aarch64-linux-gnu/version.txt
index e9d96e44d2..9759954600 100644
--- a/cpp/tensorrt_llm/batch_manager/aarch64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/batch_manager/aarch64-linux-gnu/version.txt
@@ -1,3 +1,3 @@
 f229593e4699180b52e38f99c8ac31dc libtensorrt_llm_batch_manager_static.a
 440b3ae47982d88fc8517c5f01f67b3c libtensorrt_llm_batch_manager_static.pre_cxx11.a
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.a b/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.a
index 3ba1a6dd91..2b9c3f0037 100644
--- a/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.a
+++ b/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a71c70d555673ce9a5086c27cbd27642f940d2439effb72a75f1302725a2513
-size 4522988
+oid sha256:cb9d3d05ef4b08df0fc02f39c053a4435b58f9431d1ce269439b2c1f0a055b21
+size 4523116
diff --git a/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.pre_cxx11.a b/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.pre_cxx11.a
index a947d701ee..54354fcf89 100644
--- a/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.pre_cxx11.a
+++ b/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/libtensorrt_llm_batch_manager_static.pre_cxx11.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93c436037b41d06d735acfbf065ccef4ea50085052920cb6a54fb3f84c59fb12
-size 4486958
+oid sha256:6597b35faffe93244d89595dc369ece59729c984871ad5aab531d714d39c8e49
+size 4487214
diff --git a/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/version.txt b/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/version.txt
index 6ef2144a91..1b565a5faf 100644
--- a/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/batch_manager/x86_64-linux-gnu/version.txt
@@ -1,3 +1,3 @@
-f402983564a358853384650313301b32 libtensorrt_llm_batch_manager_static.a
-44558f89a86de69191f18a2704cff505 libtensorrt_llm_batch_manager_static.pre_cxx11.a
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+c015186a31c789891a27e44f5a9ab9ec libtensorrt_llm_batch_manager_static.a
+cac21708838abf82b18e1846c40b5c79 libtensorrt_llm_batch_manager_static.pre_cxx11.a
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/tensorrt_llm_batch_manager_static.lib b/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/tensorrt_llm_batch_manager_static.lib
index 308f79d544..c234b864ef 100644
--- a/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/tensorrt_llm_batch_manager_static.lib
+++ b/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/tensorrt_llm_batch_manager_static.lib
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f8f3d6e22edead45c5bde864b541311a4b9a28f1916cd7b5bbf1292746c06c5
-size 28211626
+oid sha256:85f18d38a66cd8b15c7e447be16171b9db854f2b2fe9dc49daa4f93fae9bc125
+size 30145896
diff --git a/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/version.txt b/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/version.txt
index 0970c75a55..063e584b15 100644
--- a/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/version.txt
+++ b/cpp/tensorrt_llm/batch_manager/x86_64-windows-msvc/version.txt
@@ -1,2 +1,2 @@
-7b8f52e42d11765115c185d7d13e40a3 tensorrt_llm_batch_manager_static.lib
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+8ebb7d383e97bcd738cc24b00d58a2d0 tensorrt_llm_batch_manager_static.lib
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.a b/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.a
index 59f5c5af24..8726ff8ae0 100644
--- a/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.a
+++ b/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:954f77299e1d61a038c90bc578936ec06da842909ace2e8ba978fd0c0da0cc1f
+oid sha256:41eae80923f8634a635f2fce84fdbe33101ee6cf86c0a98ed4ce30a7f4cea350
 size 1782460
diff --git a/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a b/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a
index 7bd0e4b905..4823445883 100644
--- a/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a
+++ b/cpp/tensorrt_llm/executor/aarch64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ac77ca2662830a5990dbd06ee4d664f8ac97dc342206f5c51ca9f9ca6cb6ce1
+oid sha256:dd4fb660b1c3e664a012e26cea949dcded855e133aa6aadd01157a15df3e0d44
 size 1808956
diff --git a/cpp/tensorrt_llm/executor/aarch64-linux-gnu/version.txt b/cpp/tensorrt_llm/executor/aarch64-linux-gnu/version.txt
index 1b6f3a6393..af0c5cd814 100644
--- a/cpp/tensorrt_llm/executor/aarch64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/executor/aarch64-linux-gnu/version.txt
@@ -1,3 +1,3 @@
-425c9af7b0ae82e622771fc3ef7e3f01 libtensorrt_llm_executor_static.a
-efa5708f62775822591ebd50974ccfd8 libtensorrt_llm_executor_static.pre_cxx11.a
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+2744c4784cfd34c7311148c7f7614757 libtensorrt_llm_executor_static.a
+d56af9e74a9d49e32860d89dcca024d0 libtensorrt_llm_executor_static.pre_cxx11.a
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.a b/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.a
index 1003bc3cd0..a9670e0090 100644
--- a/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.a
+++ b/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d74551056bd413556a9485bdb7b084e14264e78f12ae6037472878371f2b3b62
-size 1846866
+oid sha256:e9a5c64c347400297bc8b6907b4feaa890305aaf5c1b45ce57aca8fcae3e881f
+size 1846898
diff --git a/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a b/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a
index b78f3799ab..0edd3b394c 100644
--- a/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a
+++ b/cpp/tensorrt_llm/executor/x86_64-linux-gnu/libtensorrt_llm_executor_static.pre_cxx11.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:393724263d6f08b831b8d8b56a3e0677f51df80183ba0d4b1fa7f40c2f8611ca
-size 1757514
+oid sha256:e4b5912ce9a1c13554f4b16d29deb6f2ad51477c56810b758ba488212f8e5dc9
+size 1757522
diff --git a/cpp/tensorrt_llm/executor/x86_64-linux-gnu/version.txt b/cpp/tensorrt_llm/executor/x86_64-linux-gnu/version.txt
index f3cf6aec16..6717291c81 100644
--- a/cpp/tensorrt_llm/executor/x86_64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/executor/x86_64-linux-gnu/version.txt
@@ -1,3 +1,3 @@
-12fd54cc3b5ec8462a9db96d646ea578 libtensorrt_llm_executor_static.a
-75fcebc1eae90d05bd3e2d3321a50041 libtensorrt_llm_executor_static.pre_cxx11.a
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+69dc85fe48625b6f8f684487f2048458 libtensorrt_llm_executor_static.a
+d46f0be3543e24c4df51ae287086ca52 libtensorrt_llm_executor_static.pre_cxx11.a
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/executor/x86_64-windows-msvc/tensorrt_llm_executor_static.lib b/cpp/tensorrt_llm/executor/x86_64-windows-msvc/tensorrt_llm_executor_static.lib
index 542b209953..cc641f7063 100644
--- a/cpp/tensorrt_llm/executor/x86_64-windows-msvc/tensorrt_llm_executor_static.lib
+++ b/cpp/tensorrt_llm/executor/x86_64-windows-msvc/tensorrt_llm_executor_static.lib
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98dcfc32cb6a6fcbd84625f0232f9f6d8305b38a7e8380b7e228be5f820c0dd4
-size 19615228
+oid sha256:fc94de246db88ca06e5a7c20d04f78057cc3c2928b3fa3e79f49e8b9d90b76da
+size 19683718
diff --git a/cpp/tensorrt_llm/executor/x86_64-windows-msvc/version.txt b/cpp/tensorrt_llm/executor/x86_64-windows-msvc/version.txt
index 09fdbe3a80..e88bce29be 100644
--- a/cpp/tensorrt_llm/executor/x86_64-windows-msvc/version.txt
+++ b/cpp/tensorrt_llm/executor/x86_64-windows-msvc/version.txt
@@ -1,2 +1,2 @@
-422bcd24325bf7ec0b26a9e4a23cce63 tensorrt_llm_executor_static.lib
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+0e361ba639fa897f489f6d0f48cfe13f tensorrt_llm_executor_static.lib
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/aarch64-linux-gnu/version.txt b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/aarch64-linux-gnu/version.txt
index 5fad36bf51..49e66bdce8 100644
--- a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/aarch64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/aarch64-linux-gnu/version.txt
@@ -1,2 +1,2 @@
 88c30973b9b3452baa3f063d34d08169 libtensorrt_llm_nvrtc_wrapper.so
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-linux-gnu/version.txt b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-linux-gnu/version.txt
index afa7b2b973..f5da2f14a1 100644
--- a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-linux-gnu/version.txt
@@ -1,2 +1,2 @@
 95e9f87610383348e444d2d0b8396f2d libtensorrt_llm_nvrtc_wrapper.so
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/tensorrt_llm_nvrtc_wrapper.dll b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/tensorrt_llm_nvrtc_wrapper.dll
index 6ec7429e8f..fadecd1af3 100644
--- a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/tensorrt_llm_nvrtc_wrapper.dll
+++ b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/tensorrt_llm_nvrtc_wrapper.dll
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9a2ccc0462e815aae0e7fd458c0423f76b3c0bb63ecc7a8902b94194803c4bc
+oid sha256:6fc6f35c712d83404e40a7840a0c9d1f5157df61df91a7207c4e4131783f4676
 size 1128448
diff --git a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/version.txt b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/version.txt
index 3be268796a..b9555dc077 100644
--- a/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/version.txt
+++ b/cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/decoderXQAImplJIT/nvrtcWrapper/x86_64-windows-msvc/version.txt
@@ -1,3 +1,3 @@
 700fc148d9a0f939e0088bf69e899360 tensorrt_llm_nvrtc_wrapper.lib
-de95527e8d24da458cf9e3c30b21abea tensorrt_llm_nvrtc_wrapper.dll
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+6ea6ac6dff8793afbd79dd5768daae85 tensorrt_llm_nvrtc_wrapper.dll
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a
index c6cf8e43c9..a6ff47b69a 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e29b1c9c454f90b68feffb65f270bba757a649c4cfa26134d21a5c71ecee9d17
+oid sha256:f4cc1e6f0b6d1e7bc875284275b591d34c707471e636019b4c2904f30798dbc9
 size 25364090
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
index a59e2258e5..cc7772d707 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:616fd0c5db05f9ba95fec4180907a890e470a42fe0d361a212b161c915d61a7b
+oid sha256:658a3f0bc5b9877e5ad447437287908dc9b7df87ae0e86f5338aaf81e26f723e
 size 25768990
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/version.txt b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/version.txt
index 963a242e0f..b8d933d49d 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/aarch64-linux-gnu/version.txt
@@ -1,3 +1,3 @@
-dc54f44211570918194805b138a0c5eb libtensorrt_llm_internal_cutlass_kernels_static.a
-7b74b643b98590d9540d62e22764a45d libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+979f3165fbc7a68528df6e343cc54e3f libtensorrt_llm_internal_cutlass_kernels_static.a
+68e84c294a658734a8b26d7270540e1d libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a
index fb440ce7ba..61b711b17c 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04908e6011b2cda19555da0ca257e0f562719a32da76ba288cd35ae17b19c762
+oid sha256:24358d8eb15a5e802cbee6d2d27735033eedb33091e9355d199229c3ba7b6447
 size 44173632
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
index 22c05c35e0..0f981d9d40 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de44f2b89ef65a7016bd074d6a69610f81752e1cebd6e86b2314c4348cc48cc9
+oid sha256:4b54efb78e98bf6580f73a3b6b689823d7c2eb851c0bab5f36906f4ebbfc44fc
 size 43561142
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/version.txt b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/version.txt
index 538702469a..b6f3278333 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/version.txt
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-linux-gnu/version.txt
@@ -1,3 +1,3 @@
-52fd02865207f1e48afb7e2090c750c3 libtensorrt_llm_internal_cutlass_kernels_static.a
-6d1e31066fce1e3c70ec0c56c4b0abb5 libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+0b49d88b8b5e83c8c6997c725a37f373 libtensorrt_llm_internal_cutlass_kernels_static.a
+7a12fc880d2a13ee5c7cf2b1e169cb19 libtensorrt_llm_internal_cutlass_kernels_static.pre_cxx11.a
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/tensorrt_llm_internal_cutlass_kernels_static.lib b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/tensorrt_llm_internal_cutlass_kernels_static.lib
index 29372825c2..d5007d0d2c 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/tensorrt_llm_internal_cutlass_kernels_static.lib
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/tensorrt_llm_internal_cutlass_kernels_static.lib
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04f935b72c3af6c2081f19b94216f0bbcecae37511d9958ff6c62304aa022cab
+oid sha256:5ba6f8610db3b967f3de4beeff6394cdd3e56d15916f39110ed932c3c3a65417
 size 88141376
diff --git a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/version.txt b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/version.txt
index 31d57a78d3..15334333c4 100644
--- a/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/version.txt
+++ b/cpp/tensorrt_llm/kernels/internal_cutlass_kernels/x86_64-windows-msvc/version.txt
@@ -1,2 +1,2 @@
-1ce107485d9c6133e135f4dbb0a5c4dc tensorrt_llm_internal_cutlass_kernels_static.lib
-0144f02d9054532ac5b39f5accd89e607f5966a2 commit
\ No newline at end of file
+5c26d1347bb8b47288d598b6d7444900 tensorrt_llm_internal_cutlass_kernels_static.lib
+7adf157833793b3215570b0a95b9c4b2998a620c commit
\ No newline at end of file
diff --git a/examples/baichuan/requirements.txt b/examples/baichuan/requirements.txt
index ac40e979d6..5c1d7709ed 100644
--- a/examples/baichuan/requirements.txt
+++ b/examples/baichuan/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.15.0
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/bloom/requirements.txt b/examples/bloom/requirements.txt
index 0703e0cf3f..4abbbacac2 100644
--- a/examples/bloom/requirements.txt
+++ b/examples/bloom/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/chatglm/requirements.txt b/examples/chatglm/requirements.txt
index 18cc266bdc..e16e87b61a 100644
--- a/examples/chatglm/requirements.txt
+++ b/examples/chatglm/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 protobuf
diff --git a/examples/dbrx/requirements.txt b/examples/dbrx/requirements.txt
index 26b87d79cc..2858ba8096 100644
--- a/examples/dbrx/requirements.txt
+++ b/examples/dbrx/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/falcon/requirements.txt b/examples/falcon/requirements.txt
index 00f735f792..21d904ffb7 100644
--- a/examples/falcon/requirements.txt
+++ b/examples/falcon/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 transformers>=4.31.0
 datasets~=2.14.5
 evaluate~=0.4.1
diff --git a/examples/gemma/requirements.txt b/examples/gemma/requirements.txt
index b4fcb28b0d..85042ed1ca 100644
--- a/examples/gemma/requirements.txt
+++ b/examples/gemma/requirements.txt
@@ -3,7 +3,7 @@
 # WAR the new posting of "nvidia-cudnn-cu12~=9.0".
 # "jax[cuda12_pip]~=0.4.19" specifies "nvidia-cudnn-cu12>=8.9" but actually requires "nvidia-cudnn-cu12~=8.9".
 nvidia-cudnn-cu12~=8.9; platform_machine == "x86_64"
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 flax~=0.8.0
 # jax[cuda12_pip]~=0.4.19; platform_system != "Windows"
 jax~=0.4.19; platform_system == "Windows"
diff --git a/examples/gpt/requirements.txt b/examples/gpt/requirements.txt
index 823f8cc3fb..1cdca2c02c 100644
--- a/examples/gpt/requirements.txt
+++ b/examples/gpt/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/gptj/requirements.txt b/examples/gptj/requirements.txt
index 00c5b7d21f..0060a16312 100644
--- a/examples/gptj/requirements.txt
+++ b/examples/gptj/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/gptneox/requirements.txt b/examples/gptneox/requirements.txt
index 9282f54e7c..c29598edaa 100644
--- a/examples/gptneox/requirements.txt
+++ b/examples/gptneox/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 rouge_score~=0.1.2
 evaluate~=0.4.1
diff --git a/examples/grok/requirements.txt b/examples/grok/requirements.txt
index adf0c917d5..b13b48bc79 100644
--- a/examples/grok/requirements.txt
+++ b/examples/grok/requirements.txt
@@ -1,6 +1,6 @@
 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets==2.14.6
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/internlm/requirements.txt b/examples/internlm/requirements.txt
index 9ccbe7a2ec..d2fa98b74f 100644
--- a/examples/internlm/requirements.txt
+++ b/examples/internlm/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets==2.14.5
 rouge_score~=0.1.2
 sentencepiece~=0.1.99
diff --git a/examples/jais/requirements.txt b/examples/jais/requirements.txt
index 823f8cc3fb..1cdca2c02c 100644
--- a/examples/jais/requirements.txt
+++ b/examples/jais/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/llama/requirements.txt b/examples/llama/requirements.txt
index bb6fada7a8..5a5c7ac02a 100644
--- a/examples/llama/requirements.txt
+++ b/examples/llama/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets==2.14.6
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/llm-api/requirements.txt b/examples/llm-api/requirements.txt
index 45608c14b2..a9f717dc2a 100644
--- a/examples/llm-api/requirements.txt
+++ b/examples/llm-api/requirements.txt
@@ -1,2 +1,2 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
diff --git a/examples/mamba/requirements.txt b/examples/mamba/requirements.txt
index 981ddbfd3a..159c212877 100644
--- a/examples/mamba/requirements.txt
+++ b/examples/mamba/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 transformers>=4.39.0
 datasets~=2.14.5
 evaluate
diff --git a/examples/medusa/requirements.txt b/examples/medusa/requirements.txt
index 0ac1cba1d2..fd75e3f5be 100644
--- a/examples/medusa/requirements.txt
+++ b/examples/medusa/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 rouge_score~=0.1.2
 sentencepiece~=0.1.99
diff --git a/examples/mixtral/requirements.txt b/examples/mixtral/requirements.txt
index 35c402d0c7..d2b3b6edba 100644
--- a/examples/mixtral/requirements.txt
+++ b/examples/mixtral/requirements.txt
@@ -1,4 +1,4 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 transformers==4.38.2
 accelerate==0.25.0
diff --git a/examples/mpt/requirements.txt b/examples/mpt/requirements.txt
index 00c5b7d21f..0060a16312 100644
--- a/examples/mpt/requirements.txt
+++ b/examples/mpt/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/nemotron/requirements.txt b/examples/nemotron/requirements.txt
index e315f2edbf..bf3e39b44d 100644
--- a/examples/nemotron/requirements.txt
+++ b/examples/nemotron/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 nemo-toolkit[all]==2.0.0rc1
 megatron-core==0.8.0
 datasets~=2.14.5
diff --git a/examples/opt/requirements.txt b/examples/opt/requirements.txt
index 00c5b7d21f..0060a16312 100644
--- a/examples/opt/requirements.txt
+++ b/examples/opt/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/phi/requirements.txt b/examples/phi/requirements.txt
index b924a38533..d041d57d6a 100644
--- a/examples/phi/requirements.txt
+++ b/examples/phi/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/quantization/requirements.txt b/examples/quantization/requirements.txt
index dd721e1b93..9e5059e68c 100644
--- a/examples/quantization/requirements.txt
+++ b/examples/quantization/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets>=2.14.4
 nemo-toolkit[all]<=1.20.0,>=1.18.0
 rouge_score~=0.1.2
diff --git a/examples/qwen/requirements.txt b/examples/qwen/requirements.txt
index f291ee341a..a100bbffe8 100644
--- a/examples/qwen/requirements.txt
+++ b/examples/qwen/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.16.0
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/qwenvl/requirements.txt b/examples/qwenvl/requirements.txt
index 46489ca057..a6b70325a5 100644
--- a/examples/qwenvl/requirements.txt
+++ b/examples/qwenvl/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.16.0
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/recurrentgemma/requirements.txt b/examples/recurrentgemma/requirements.txt
index 1e9f04dc58..05d787ade9 100644
--- a/examples/recurrentgemma/requirements.txt
+++ b/examples/recurrentgemma/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 git+https://github.com/google-deepmind/recurrentgemma.git
 flax>=0.8.2
 jax~=0.4.23
diff --git a/examples/redrafter/requirements.txt b/examples/redrafter/requirements.txt
index 0ac1cba1d2..fd75e3f5be 100644
--- a/examples/redrafter/requirements.txt
+++ b/examples/redrafter/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.14.5
 rouge_score~=0.1.2
 sentencepiece~=0.1.99
diff --git a/examples/skywork/requirements.txt b/examples/skywork/requirements.txt
index 17e29e6f2f..563246878f 100644
--- a/examples/skywork/requirements.txt
+++ b/examples/skywork/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets~=2.16.1
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/smaug/requirements.txt b/examples/smaug/requirements.txt
index bb6fada7a8..5a5c7ac02a 100644
--- a/examples/smaug/requirements.txt
+++ b/examples/smaug/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 datasets==2.14.6
 evaluate~=0.4.1
 rouge_score~=0.1.2
diff --git a/examples/whisper/requirements.txt b/examples/whisper/requirements.txt
index 485ef3621b..9c9fed2b4d 100644
--- a/examples/whisper/requirements.txt
+++ b/examples/whisper/requirements.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pypi.nvidia.com
-tensorrt_llm==0.14.0.dev2024092400
+tensorrt_llm==0.14.0.dev2024092401
 tiktoken
 datasets
 kaldialign
diff --git a/tensorrt_llm/version.py b/tensorrt_llm/version.py
index adafbd5942..b1ebedec53 100644
--- a/tensorrt_llm/version.py
+++ b/tensorrt_llm/version.py
@@ -12,4 +12,4 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-__version__ = "0.14.0.dev2024092400"
+__version__ = "0.14.0.dev2024092401"