[Docker] bump up neuron sdk v2.21 (#11593)

Signed-off-by: Liangfu Chen <liangfc@amazon.com>

Author: liangfu

Dockerfile.neuron

@@ -1,6 +1,6 @@
 # default base image
 # https://gallery.ecr.aws/neuron/pytorch-inference-neuronx
-ARG BASE_IMAGE="public.ecr.aws/neuron/pytorch-inference-neuronx:2.1.2-neuronx-py310-sdk2.20.2-ubuntu20.04"
+ARG BASE_IMAGE="public.ecr.aws/neuron/pytorch-inference-neuronx:2.5.1-neuronx-py310-sdk2.21.0-ubuntu22.04"
 
 FROM $BASE_IMAGE
 
@@ -22,9 +22,9 @@ WORKDIR ${APP_MOUNT}/vllm
 
 RUN python3 -m pip install --upgrade pip
 RUN python3 -m pip install --no-cache-dir fastapi ninja tokenizers pandas
-RUN python3 -m pip install sentencepiece transformers==4.36.2 -U
+RUN python3 -m pip install sentencepiece transformers==4.45.2 -U
 RUN python3 -m pip install transformers-neuronx --extra-index-url=https://pip.repos.neuron.amazonaws.com -U
-RUN python3 -m pip install --pre neuronx-cc==2.15.* --extra-index-url=https://pip.repos.neuron.amazonaws.com -U
+RUN python3 -m pip install neuronx-cc==2.16.345.0 --extra-index-url=https://pip.repos.neuron.amazonaws.com -U
 
 COPY . .
 ARG GIT_REPO_CHECK=0

requirements-neuron.txt

@@ -2,6 +2,6 @@
 -r requirements-common.txt
 
 # Dependencies for Neuron devices
-transformers-neuronx >= 0.12.0
-torch-neuronx >= 2.1.2
+transformers-neuronx >= 0.13.0
+torch-neuronx >= 2.5.0
 neuronx-cc

vllm/_custom_ops.py

@@ -23,8 +23,7 @@
     import vllm._moe_C  # noqa: F401
     supports_moe_ops = True
 
-# neuron has torch version that doesn't even have impl_abstract
-if TYPE_CHECKING or current_platform.is_neuron():
+if TYPE_CHECKING:
 
     def register_fake(fn):
         return lambda name: fn

vllm/triton_utils/importing.py

@@ -8,7 +8,6 @@
 HAS_TRITON = (
     find_spec("triton") is not None
     and not current_platform.is_xpu()  # Not compatible
-    and not current_platform.is_neuron()  # neuron has too old torch
 )
 
 if not HAS_TRITON: