This is designed to be a test project to test the performance of out-of-the-box transformers. The model is stored here in the project to decrease startup time when running in the cloud but i can't store those large of files in my github...
There are 6 tags available at sperry/1996 docker hub. All the tags are run with a pytorch backend unless otherwise noted. A couple have a tensorflow backend. The docker images were built with the model on the image to avoid downloading on startup.
Env vars:
- ITERATIONS = how many iterations to run (default 25)
- NUMBER_SENTENCES = how many sentences per iteration (default 100)
- MODEL_PATH = which model to load, by default it will load the cardiffnlp/twitter-roberta-base-sentiment stored locally in the image.
The 0.0.1-CPU has python 3.10.9, torch 1.13.1, and transformers 4.25.1 (see Dockerfile for more details)
The 0.0.1 has python 3.10.8, torch 1.12.1, and transformers 4.24.0. (see Dockerfile.multi for more details)
the 0.0.1-INTEL has a pytorch intel optimized runtime with python 3.8.16, torch 1.12.1+cpu, and transformers 4.25.1 (see Dockerfile.intel for more details)
The 0.0.1-PYTORCH has python 3.9.12, torch 1.13.0, and transformers 4.25.1 (see Dockerfile.pytorch for more details)
0.0.1-HF is using the huggingface docker image, transformers version 4.23.1, pytorch (see Dockerfile.hf for more details)
0.0.1-TF has tensorflow 2.11.0 installed (see Dockerfile.tf for more details)
0.0.1-HFTF is using the huggingface docker image with tensorflow backend (see Dockerfile.hftf for more details)