Builders & open source
Johannes Gäßler
Experimental particle physicist, Karlsruhe Institute of Technology
- Experimental particle physicist, Karlsruhe Institute of Technology3
- Member of the CUDA backend code-owner team (ggml-org/ggml-cuda), llama.cpp project1
Johannes Gäßler contributes to llama.cpp, the C/C++ runtime for running language models on local hardware, and the project's CODEOWNERS file lists Gäßler in the team that owns its CUDA backend. The same file names Gäßler as co-owner of the GGUF file code (gguf.cpp) and owner of the ggml optimizer code (ggml-opt.cpp) and the lookup example. Gäßler's 2026 commits to the repository cover CUDA matrix-multiplication and FlashAttention kernels, HIP tuning for AMD GPUs, and tensor-parallel fixes. Gäßler's GitHub profile gives the Karlsruhe Institute of Technology as company and describes Gäßler as an experimental particle physicist.123
At a glance
- Current roles
- Experimental particle physicist +1 more
- Sector
- University
- Work listed
- 1
- Sources cited
- 12
- Profile reviewed
Open-source work
Gäßler's first contributions to llama.cpp, in April and May 2023, concerned CUDA builds and reproducibility. In July 2023 Gäßler added quantized matrix-matrix multiplication (MMQ) for the CUDA backend, and in June 2024 added quantized KV-cache support to the vector FlashAttention kernels.4567
In September 2024 Gäßler added backend support for numerical optimization to ggml, followed in May 2025 by LLM training support in llama.cpp and ggml, and in December 2024 added tests for the GGUF file format code. In December 2025 Gäßler added code that automatically sets parameters the user has not chosen so as to maximize GPU utilization. The CODEOWNERS file now lists Gäßler in the CUDA backend team, as owner of ggml-opt.cpp, the lookup example and common/fit, and as co-owner of gguf.cpp.8910111
Research interests
Gäßler's GitHub profile describes an experimental particle physicist at the Karlsruhe Institute of Technology with further interests in statistics, generative neural networks and hardware acceleration.3
Work and contributions
llama.cpp Open-source code
CODEOWNERS lists Gäßler in the ggml-org/ggml-cuda team that owns the CUDA backend, as co-owner of gguf.cpp, and as owner of ggml-opt.cpp, ggml-backend-meta.cpp, the lookup example, and common/fit. Commits authored in 2026 cover CUDA MMQ and FlashAttention kernels, HIP tuning for AMD GPUs, and tensor parallelism. The repository is MIT-licensed.1212
Timeline
- Made a first contribution to llama.cpp, adding CUDA include and library paths to the Makefile.4
- Added quantized matrix-matrix multiplication (MMQ) to the llama.cpp CUDA backend.6
- Added quantized KV-cache support to the CUDA vector FlashAttention kernels.7
- Added backend support for numerical optimization to ggml.8
- Added tests for the GGUF file format code.10
- Added LLM training support to llama.cpp and ggml.9
- Added automatic fitting of unset parameters to maximize GPU utilization in llama.cpp.11
Where they work
- Karlsruhe Institute of Technology 1 person in the wiki
- llama.cpp project 12 people in the wiki
Related people
- Jeff JohnsonChips and hardware · Inference and serving
- Ian BuckChips and hardware · ML frameworks and libraries
- Gavin UbertiChips and hardware · Inference and serving
- Bill DallyChips and hardware · Inference and serving
- Hani JamjoomChips and hardware · Inference and serving
- George HotzML frameworks and libraries · Chips and hardware
- Aaron TeoInference and serving · Chips and hardware
- Tim DavisML frameworks and libraries · Inference and serving
- Patrick von PlatenML frameworks and libraries · Inference and serving
- Azalia MirhoseiniChips and hardware · Inference and serving
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
12 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- ggml-org/llama.cpp CODEOWNERS
- ggml-org/llama.cpp commits authored by JohannesGaessler
- JohannesGaessler (Johannes Gäßler) on GitHub
- llama.cpp commit: Add Manjaro CUDA include and lib dirs to Makefile (#1212)
- llama.cpp commit: Documented CUDA reproducibility, added warning (#1346)
- llama.cpp commit: CUDA: Quantized matrix matrix multiplication (#2160)
- llama.cpp commit: CUDA: quantized KV support for FA vec (#7527)
- llama.cpp commit: ggml/examples: add backend support for numerical optimization (ggml/949)
- llama.cpp commit: llama/ggml: add LLM training support (#10544)
- llama.cpp commit: tests: add tests for GGUF (#10830)
- llama.cpp commit: llama: automatically set parameters to maximize GPU utilization
- ggml-org/llama.cpp LICENSE (MIT)