Skip to content

Instantly share code, notes, and snippets.

Last active Oct 29, 2021
What would you like to do?
Dockerfile for benchmarking Bert Base Uncased Model in GCP (multiple threads execution + openmp disabled for onnxruntime)
# Install TF entreprise
WORKDIR /workspace
# onnxruntime version >= 1.8 replaced OMP with thread pool:
RUN pip install --no-cache-dir --upgrade pip && \
pip install --upgrade torch==1.8.1+cpu -f && \
pip install coloredlogs sympy onnx tf2onnx onnxruntime==1.8.1 transformers==4.6.0 py-cpuinfo py3nvml
# Service account credentials are required to store benchmark results in a GCS bucket.
# Requirements before building this image.
# 1. Create a gcs bucket named gs://bert-inference-results
# 2. Create a service account and grant it access to the GCS bucket.
RUN gcloud auth activate-service-account --key-file=/workspace/<SERVICE_ACCOUNT_KEY_FILE>.json
COPY ./ /workspace/
COPY ./bert-base-uncased /workspace/bert-base-uncased
# Zip the onnxruntime customized benchmark scripts from
# Add the zipped file to workspace
ADD onnxruntime_benchmark.tar.gz /workspace
CMD [ "./" ]
Copy link

mwakaba2 commented Sep 10, 2021 can be found here

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment