Where to Find (Near) Real-Time Deep Learning Framework Market Share Data

Why this question is tricky

If you ask for real-time market share of deep learning frameworks, there is no single official dashboard that gives a perfect answer.

What we can get in practice is near real-time adoption signals from multiple public sources.

The best web sources to track framework share

1) Papers With Code

• Website: paperswithcode.com
• What it tells you: research popularity (which framework appears in SOTA repos and paper implementations)
• Strength: very good for research trend momentum
• Limitation: not equal to enterprise production usage

2) Kaggle Survey

• Website: kaggle.com
• What it tells you: practitioner self-reported usage (annual snapshot)
• Strength: broad data-science community signal
• Limitation: annual, not real-time

3) Stack Overflow Survey + Trends

• Websites:
  • survey.stackoverflow.co
  • stackoverflow.com/trends
• What it tells you: community discussion and interest trends
• Strength: useful long-term direction signal
• Limitation: discussion volume is not direct market share

4) PyPI Stats

• Website: pypistats.org
• What it tells you: package download activity (e.g., torch, tensorflow, jax)
• Strength: frequent updates, easy to track over time
• Limitation: downloads can be inflated by CI/CD and mirrors

5) GitHub + Octoverse

• Websites:
  • github.com
  • octoverse.github.com
• What it tells you: repo activity, stars, contributors, ecosystem growth
• Strength: captures open-source developer momentum
• Limitation: stars are noisy and can lag actual production adoption

6) Hugging Face ecosystem signals

• Website: huggingface.co
• What it tells you: practical model and library adoption in modern ML workflows
• Strength: strong signal for current GenAI/NLP/CV usage patterns
• Limitation: ecosystem is broad, not a strict framework vote

How to build a “best-effort live” market-share dashboard

Track three dimensions weekly:

1. Research share
   • Metric examples: Papers With Code framework mentions, SOTA implementation distribution
2. Developer share
   • Metric examples: GitHub stars growth, active contributors, issue/PR velocity
3. Usage proxy share
   • Metric examples: PyPI download trend for torch, tensorflow, jax

Then normalize each metric and combine with weighted scoring:

\[ext{Composite Share} = 0.4 \cdot \text{Research} + 0.3 \cdot \text{Developer} + 0.3 \cdot \text{Usage Proxy}\]

The weights are configurable. If your focus is production usage, increase the usage proxy weight.

Suggested tracking cadence

• Weekly: PyPI + GitHub metrics
• Monthly: Papers With Code trend snapshot
• Quarterly: interpretation review (remove anomalies, tune weights)
• Yearly: compare against Kaggle/Stack Overflow survey baselines

Practical conclusions (2026)

• PyTorch remains strongest in research mindshare and modern open-source momentum.
• TensorFlow/Keras still shows meaningful production and legacy enterprise footprint.
• JAX continues to grow in high-performance research niches.
• Smaller frameworks can win specific verticals, but ecosystem depth decides long-term share.

Final takeaway

There is no perfect real-time “market share” API for deep learning frameworks. The reliable approach is to combine multiple public signals and maintain a transparent, repeatable scoring method.

If you publish the metric definition and weights, your trend report becomes far more credible than any single-number claim.