PeopleofSuperintelligence

Builder

Pete Walsh

Also known as Evan Pete Walsh

Pete Walsh is a research engineer who works on open-source language models. By 2020, he described NLP work on intent classification and entity extraction; from 2022 through 2026, he coauthored research on transformer training, evaluation tools, open datasets, and language and vision-language models. His projects include Catwalk, Pentathlon, Dolma, OLMo, Molmo and PixMo, OLMo 2, Olmo 3, and Olmo Hybrid. As of October 2026, he works on open-source LLMs and is part of Microsoft's Superintelligence team.

Now at
Microsoft
Works
3
Links
GitHub

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. By 2020

    NLP engineering work

    By 2020, he describes work on NLP tasks such as intent classification and entity extraction. He also tests machine-learning models and writes Python applications for AI products.

  2. 2022

    Staged transformer training

    In 2022, he coauthors Staged Training for Transformer Language Models, which grows a small model by adding depth and width. The method yields up to 22% compute savings over a strong baseline trained from scratch.

  3. 2023

    Catwalk and Pentathlon

    In 2023, he coauthors Catwalk, a unified interface to language-model datasets and models. He also coauthors Pentathlon, an inference-efficiency benchmark that measures latency, throughput, memory overhead, and energy consumption.

  4. 2024

    Dolma and OLMo

    In 2024, he coauthors Dolma, an open three-trillion-token English corpus and data-curation toolkit, and OLMo, a language model released with open training data and training and evaluation code.

  5. September 2024

    Molmo and PixMo

    In September 2024, he coauthors Molmo, a family of vision-language models, and PixMo, a collection of new training datasets. The release makes model weights, datasets, and source code available.

  6. 2025

    OLMo 2 and Olmo 3

    In October 2025, he coauthors a revised version of 2 OLMo 2 Furious, which describes OLMo 2 models at 7B, 13B, and 32B scales with released weights, full training data, code, recipes, logs, and thousands of intermediate checkpoints. In December 2025, he coauthors Olmo 3, a 7B and 32B model family whose release includes every training stage, checkpoint, data point, and dependency.

  7. 2026

    A hybrid 7B model

    In 2026, he coauthors Olmo Hybrid, a 7B model that replaces sliding-window layers with Gated DeltaNet layers and outperforms Olmo 3 across standard pretraining and mid-training evaluations.

  8. As of October 2026

    Microsoft Superintelligence team

    By May 2026, he is among the former Ai2 researchers on Microsoft's Superintelligence team. The team focuses on model post-training and applications in health care and energy. He works as a research engineer on open-source LLMs.

Career

  1. Since 2026MicrosoftResearch Engineer
  2. Until 2026Allen Institute for AI (Ai2)

From the biography.

  1. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Microsoft. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Microsoft. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  4. Google has the highest combined h-index among AI organizationsIncludes Microsoft. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Pete Walsh yet. Send us a free-licensed photo.

  1. What Does A Data Scientist Do On A Daily Basis | Quora Answers analyticsvidhya.com. Checked 21 October 2026.
  2. Staged Training for Transformer Language Models arxiv.org. Checked 21 October 2026.
  3. Catwalk: A Unified Language Model Evaluation Framework for Many Datasets arxiv.org. Checked 21 October 2026.
  4. Efficiency Pentathlon: A Standardized Arena for Efficiency Evaluation arxiv.org. Checked 21 October 2026.
  5. Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research arxiv.org. Checked 21 October 2026.
  6. GitHub - allenai/dolma: Data and tools for generating and inspecting OLMo pre-training data github.com. Checked 21 October 2026.
  7. OLMo: Accelerating the Science of Language Models arxiv.org. Checked 21 October 2026.
  8. Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models arxiv.org. Checked 21 October 2026.
  9. 2 OLMo 2 Furious arxiv.org. Checked 21 October 2026.
  10. Olmo 3 arxiv.org. Checked 21 October 2026.
  11. Olmo Hybrid: From Theory to Practice and Back arxiv.org. Checked 21 October 2026.
  12. Microsoft adds at least 10 former Ai2 researchers to its Superintelligence team completeaitraining.com. Checked 21 October 2026.
  13. epwalsh (Pete Walsh) · GitHub github.com. Checked 21 October 2026.