AI People Wiki

Builders & open source

Pierrick Hymbert

Developer (BNP Paribas is the company listed on Hymbert's GitHub profile), BNP Paribas

  • Developer (BNP Paribas is the company listed on Hymbert's GitHub profile), BNP Paribas1

Pierrick Hymbert is a developer whose GitHub profile lists BNP Paribas as company. Hymbert's commits to llama.cpp, the C/C++ runtime for running language models on local hardware, cover the project's HTTP server, the CI workflows that test and benchmark it, and support for model architectures including DBRX and PhiMoE; the most recent dates from January 2025. Hymbert also added an eval-callback example for debugging, imatrix and dataset metadata in quantized GGUF files, and documentation on how to add a model.12

At a glance

Current role
Developer (BNP Paribas is the company listed on Hymbert's GitHub profile)
Sector
Company
Work listed
1
Sources cited
3
Profile reviewed

Open-source work

Hymbert's llama.cpp contributions in 2024 included tooling for large model files. Commits added gguf-split, which splits and merges GGUF files by batches of tensors, taught the model loader to read models sharded across several GGUF files, and let llama.cpp download a model directly from a URL with the --model-url option.2

Much of Hymbert's work concerned llama.cpp's HTTP server and its continuous integration. Commits set up functional tests for the server and a benchmark workflow that monitors server performance and comments on pull requests, along with server fixes such as reporting token usage in streamed OpenAI-compatible responses and stopping generation at the training context length.2

Hymbert also added support for the DBRX (April 2024) and PhiMoE (January 2025) model architectures, an eval-callback example for debugging computation graphs, imatrix and dataset metadata in quantized GGUF files, and documentation on how to add a new model to llama.cpp.2

Work and contributions

  • llama.cpp Open-source code

    Commits authored from February 2024 to January 2025 include server test and benchmark CI workflows, server fixes such as token usage in streamed OpenAI-compatible responses, support for the DbrxForCausalLM and PhiMoE architectures, the eval-callback debugging example, imatrix and dataset metadata in quantized GGUF files, and a guide to adding a model. The repository is MIT-licensed.23

Timeline

  1. Added the --model-url option for loading a model from a URL in llama.cpp.2
  2. Added gguf-split for splitting and merging GGUF model files.2
  3. Added support for loading models sharded across multiple GGUF files.2
  4. Added continuous performance monitoring with pull-request comments for the llama.cpp server.2
  5. Wrote llama.cpp documentation on how to add a model.2
  6. Added support for the DbrxForCausalLM architecture.2
  7. Added imatrix and dataset metadata to quantized GGUF files.2
  8. Added support for the PhiMoE architecture.2

Where they work

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

3 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. phymbert (Pierrick Hymbert) on GitHubGitHub · Official page · accessed Oct 21, 2026
  2. ggml-org/llama.cpp commits authored by phymbertggml-org (GitHub repository) · Repository · accessed Oct 21, 2026
  3. ggml-org/llama.cpp LICENSE (MIT)ggml-org (GitHub repository) · License · accessed Oct 21, 2026