Merge branch 'main' into GraniteCodeSupport

Author: Jack-Khuu

install/.pins/torchao-pin.txt

@@ -1 +1 @@
-7d7c14e898eca3fe66138d2a9445755a9270b800
+2e032c6b0de960dee554dcb08126ace718b14c6d

install/install_requirements.sh

@@ -44,31 +44,20 @@ fi
 
 echo "Using pip executable: $PIP_EXECUTABLE"
 
-#
-# First install requirements in install/requirements.txt. Older torch may be
-# installed from the dependency of other models. It will be overridden by
-# newer version of torch nightly installed later in this script.
-#
-
-(
-  set -x
-  $PIP_EXECUTABLE install -r install/requirements.txt --extra-index-url https://download.pytorch.org/whl/nightly/cu121
-)
-
 # Since torchchat often uses main-branch features of pytorch, only the nightly
 # pip versions will have the required features. The PYTORCH_NIGHTLY_VERSION value should
 # agree with the third-party/pytorch pinned submodule commit.
 #
 # NOTE: If a newly-fetched version of the executorch repo changes the value of
 # PYTORCH_NIGHTLY_VERSION, you should re-run this script to install the necessary
 # package versions.
-PYTORCH_NIGHTLY_VERSION=dev20241213
+PYTORCH_NIGHTLY_VERSION=dev20241218
 
 # Nightly version for torchvision
-VISION_NIGHTLY_VERSION=dev20241213
+VISION_NIGHTLY_VERSION=dev20241218
 
 # Nightly version for torchtune
-TUNE_NIGHTLY_VERSION=dev20241126
+TUNE_NIGHTLY_VERSION=dev20241218
 
 # Uninstall triton, as nightly will depend on pytorch-triton, which is one and the same
 (
@@ -96,6 +85,16 @@ REQUIREMENTS_TO_INSTALL=(
   torchtune=="0.5.0.${TUNE_NIGHTLY_VERSION}"
 )
 
+#
+# First install requirements in install/requirements.txt. Older torch may be
+# installed from the dependency of other models. It will be overridden by
+# newer version of torch nightly installed later in this script.
+#
+(
+  set -x
+  $PIP_EXECUTABLE install -r install/requirements.txt --extra-index-url "${TORCH_NIGHTLY_URL}"
+)
+
 # Install the requirements. --extra-index-url tells pip to look for package
 # versions on the provided URL if they aren't available on the default URL.
 (

torchchat/utils/quantize.py

@@ -932,13 +932,15 @@ def quantized_model(self) -> nn.Module:
         libs = glob.glob(f"{torchao_build_path}/cmake-out/lib/libtorchao_ops_aten.*")
         libs = list(filter(lambda l: (l.endswith("so") or l.endswith("dylib")), libs))
         torch.ops.load_library(libs[0])
+        print("Loaded torchao cpu ops.")
     except Exception as e:
         print("Unabled to load torchao cpu ops library. Slow fallback kernels will be used.")
 
     try:
         libname = "libtorchao_ops_mps_aten.dylib"
         libpath = f"{torchao_build_path}/cmake-out/lib/{libname}"
         torch.ops.load_library(libpath)
+        print("Loaded torchao mps ops.")
     except Exception as e:
         print("Unabled to load torchao mps ops library.")