SkillTrendLab
Skill · Frameworks · Framework, LLM and domain signals

PyTorch

The most used framework for deep-learning research and production.

17.6%
of core AI postings mention it
619
postings
384
companies

History starts on September 11, 2026. Movement needs a comparable later edition at least 30 days on.

Where it is mentioned most

Each row divides by that role's postings in this edition, highest share first.

Reading this edition

Postings in 10 role families mention PyTorch in this edition.

In 4 of the 8 role families with a page of their own, at least one posting in five mentions PyTorch: Computer Vision Engineer 38.0%, Data Scientist 31.8%, LLM / NLP Engineer 26.9%, Algorithm Engineer 24.0%.

PyTorch is more concentrated in Computer Vision Engineer postings (38.0%) than across all core AI postings (17.6%): at least twice the overall share.

PyTorch is in Frameworks: a framework, LLM or domain signal. Read it together with the foundation and production skills that appear with it below.

How the readings are made: "more concentrated than overall" means that in a role family with a page of its own, at least 10 postings mention the skill and its share is at least 2 times its share of all core AI postings; "common" means at least 20% of a role's postings mention it. They describe the postings open in this edition, not a trend.

  1. 43 of 113 postings · at least twice overall
  2. 48 of 151 postings
  3. AI Researcher30.0%
    18 of 60 postings · Sample below the page threshold; not read
  4. MLOps / Platform Engineer28.9%
    11 of 38 postings · Sample below the page threshold; not read
  5. 47 of 175 postings
  6. 77 of 321 postings
  7. 337 of 1,741 postings
  8. 10 of 123 postings
  9. 6 of 225 postings
  10. 8 of 305 postings

Skills that appear with it

Each share divides by the postings that mention PyTorch: how many of them also mention each skill, grouped into three lenses.

Foundations and problem-solving

Categories: Languages, Machine learning, Data engineering, Databases, Methods and process

  1. Python90.3%
    Languages · 559 postings
  2. Machine learning · 403 postings
  3. Machine learning · 369 postings
  4. C++40.1%
    Languages · 248 postings
  5. Machine learning · 225 postings
  6. Methods and process · 218 postings
  7. MLOps28.0%
    Machine learning · 173 postings

Production and delivery

Categories: DevOps, Cloud, Tools

  1. Git33.1%
    Tools · 205 postings
  2. Linux30.4%
    DevOps · 188 postings
  3. Docker26.8%
    DevOps · 166 postings

Framework, LLM and domain signals

Categories: Frameworks, LLM applications, Domain knowledge, Office software

  1. Frameworks · 479 postings
  2. LLM45.9%
    LLM applications · 284 postings

Industries and places

Share of this skill's postings.

Industry
  1. Software & Internet
    203
    32.8%
  2. Electronics & Hardware
    124
    20.0%
  3. Semiconductor
    89
    14.4%
  4. Professional Services
    49
    7.9%
  5. Healthcare & Biotech
    40
    6.5%
  6. Finance & Insurance
    36
    5.8%
  7. Manufacturing
    32
    5.2%
  8. Wholesale, Retail & E-commerce
    16
    2.6%
  9. Telecom, Media & Digital Content
    11
    1.8%
  10. Construction & Real Estate
    6
    1.0%
  11. Consulting & Marketing Services
    6
    1.0%
  12. Government & Nonprofit
    3
    0.5%
Place
  1. Taipei
    279
    45.1%
  2. New Taipei
    87
    14.1%
  3. Hsinchu City
    63
    10.2%
  4. Hsinchu County
    62
    10.0%
  5. Taichung
    37
    6.0%
  6. Taoyuan
    28
    4.5%
  7. Kaohsiung
    25
    4.0%
  8. Miaoli
    15
    2.4%
  9. Tainan
    13
    2.1%
  10. Yunlin
    2
    0.3%
  11. Nantou
    2
    0.3%
  12. Hualien
    1
    0.2%

Where this number is read from

Of the 619 postings, how many were found through the employer's own skill tags, the title, and the job description. A posting can carry more than one kind of evidence, so the three add up to more than the posting count.

  1. Employer's skill tags
    253
  2. Job title
    1
  3. Job description
    526
The data on this page
Data as of
September 11, 2026
Core AI postings this edition
3,523
Method version
2026.09.1

This page is not indexed yet: change needs a second edition.