In a post from last summer, I noted how rapidly PyTorch was gaining users in the machine learning research community. At that time PyTorch was growing 194% year-over-year (compared to a 23% growth rate for TensorFlow). That post used research papers, specifically simple full-text searches of papers posted on the popular e-print service arXiv.org.
With that said, startups would be remiss to ignore the larger share of TensorFlow among enterprise users. TensorFlow remains more popular among companies as it still has more deployment and MLOps tools, and a broader collection of companies ready to provide enterprise support. The depth and breadth of the TensorFlow ecosystem was on full display at TensorFlow World last November. It remains to be seen how long these advantages persist.
The chart below displays the count of the number of job postings that mention “TensorFlow” or “PyTorch” respectively:
What about the relative size of the job market for each of these tools? The total number of PyTorch job postings across the nine metro areas in the chart, is about 60% that of the number of TensorFlow job postings:
Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.
- Rajat Monga: “The evolution of TensorFlow and of machine learning infrastructure”.
- Nir Shavit: “The combination of the right software and commodity hardware will prove capable of handling most machine learning tasks”.
- Morten Dahl: “The state of privacy-preserving machine learning”.
- Dean Wampler: “Scalable Machine Learning, Scalable Python, For Everyone”.