Prepare for release and some bug fixes (#1182)

microsoft · Jan 13, 2025 · 874d9b6 · 874d9b6
1 parent 49eb184
commit 874d9b6
Show file tree

Hide file tree

Showing 7 changed files with 9 additions and 6 deletions.
diff --git a/.pipelines/stages/jobs/steps/python-validation-step.yml b/.pipelines/stages/jobs/steps/python-validation-step.yml
@@ -35,6 +35,9 @@ steps:
         python -m pip install -r test/python/directml/torch/requirements.txt
         python -m pip install -r test/python/directml/ort/requirements.txt
       }
+      elseif ("$(arch)" -eq "arm64") {
+        python -m pip install onnxruntime-qnn
+      }
       else {
         python -m pip install -r test/python/cpu/torch/requirements.txt
         python -m pip install -r test/python/cpu/ort/requirements.txt

diff --git a/src/models/kv_cache.cpp b/src/models/kv_cache.cpp
@@ -367,7 +367,7 @@ void DefaultKeyValueCache::RewindPastTensorsTo(size_t index) {
 // Copy present state to past state reordered by the beam_indices
 template <typename ScoreType>
 void DefaultKeyValueCache::PickPastState(DeviceSpan<int32_t> beam_indices_device, int index) {
-  std::span<int32_t> beam_indices = beam_indices_device.Span();
+  std::span<int32_t> beam_indices = beam_indices_device.CopyDeviceToCpu();
   auto block_size_per_beam = shape_[1] * shape_[2] * shape_[3];
   auto element_count = shape_[0] * block_size_per_beam;
 

diff --git a/src/python/setup.py.in b/src/python/setup.py.in
@@ -26,7 +26,7 @@ package_name = '@TARGET_NAME@'
 def _onnxruntime_dependency() -> str:
     dependency = None
     # Use stable version as default
-    ort_version = os.environ.get("ONNXRUNTIME_VERSION", "1.20.0")
+    ort_version = os.environ.get("ONNXRUNTIME_VERSION", "1.20.1")
 
     if package_name == "onnxruntime-genai":
         dependency = "onnxruntime"

diff --git a/test/python/cpu/ort/requirements.txt b/test/python/cpu/ort/requirements.txt
@@ -1,2 +1,2 @@
 # -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
-onnxruntime==1.20.0
+onnxruntime==1.20.1
diff --git a/test/python/cuda/ort/requirements.txt b/test/python/cuda/ort/requirements.txt
@@ -1,2 +1,2 @@
 # -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
-onnxruntime-gpu==1.20.0
+onnxruntime-gpu==1.20.1
diff --git a/test/python/directml/ort/requirements.txt b/test/python/directml/ort/requirements.txt
@@ -1,2 +1,2 @@
 # -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
-onnxruntime-directml==1.20.0
+onnxruntime-directml==1.20.1
diff --git a/test/python/macos/ort/requirements.txt b/test/python/macos/ort/requirements.txt
@@ -1,2 +1,2 @@
 # -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
-onnxruntime==1.20.0
+onnxruntime==1.20.1