AWS Neuron SDK 2.16.1: update neuronxcc (#449)

* fix(decoder): no need to set model type when exporting

* chore: bump neuronx-cc version

* chore: store AWS Neuron SDK version

optimum/neuron/modeling_decoder.py

@@ -123,14 +123,11 @@ def __init__(
             neuronx_model.load(compiled_dir)
 
         # Compile the Neuron model (if present compiled artifacts will be reloaded instead of compiled)
-        neuron_cc_flags = os.environ.get("NEURON_CC_FLAGS", "")
-        os.environ["NEURON_CC_FLAGS"] = neuron_cc_flags + " --model-type=transformer"
         checkpoint_id = neuron_config.get("checkpoint_id", None)
         # Only create a cache entry if the model comes from the hub
         cache_entry = None if checkpoint_id is None else ModelCacheEntry(checkpoint_id, config)
         with hub_neuronx_cache(entry=cache_entry):
             neuronx_model.to_neuron()
-        os.environ["NEURON_CC_FLAGS"] = neuron_cc_flags
 
         super().__init__(neuronx_model, config)
 

optimum/neuron/version.py

@@ -13,3 +13,5 @@
 #  limitations under the License.
 
 __version__ = "0.0.18.dev0"
+
+__sdk_version__ = "2.16.1"

setup.py

@@ -55,7 +55,7 @@
     ],
     "neuronx": [
         "wheel",
-        "neuronx-cc==2.12.54.0",
+        "neuronx-cc==2.12.68.0",
         "torch-neuronx==1.13.1.1.13.0",
         "transformers-neuronx==0.9.474",
         "torch==1.13.1.*",

text-generation-inference/Dockerfile

@@ -99,7 +99,7 @@ RUN apt-get update -y \
 ENV PATH="/opt/bin/:/opt/aws/neuron/bin:${PATH}"
 
 RUN pip3 install \
-    neuronx-cc==2.12.54.0 \
+    neuronx-cc==2.12.68.0 \
     torch-neuronx==1.13.1.1.13.0 \
     transformers-neuronx==0.9.474 \
     --extra-index-url=https://pip.repos.neuron.amazonaws.com