Bartowski

Open-source work

Bartowski's earliest model uploads on Hugging Face, from late October 2023, were ExLlamaV2 (exl2) quantizations of fine-tuned Mistral 7B models. The account now centres on GGUF quantizations for llama.cpp and compatible runtimes, and its model cards record the llama.cpp release used and the use of imatrix calibration data.⁵¹²

Bartowski's commits to llama.cpp began in May 2024, including support for the Smaug 70B model, and have continued with model-architecture support (an Arcee AFM model, GLM 4.6 and AfmoeForCausalLM), conversion-script fixes and chat-template additions. Many later commits concern quantization quality and speed, such as fixes for IQ quantization types, faster AVX2 prompt processing for IQ models and a shared imatrix loader, along with a per-slot context option for llama-server.⁴

The llama.cpp CODEOWNERS file lists bartowski1182 as a co-owner of the tiled CPU code under ggml/src/ggml-cpu/tiled. On the Hugging Face profile, Bartowski gives a current role as Senior Machine Learning Engineer at Red Hat.³¹

Work and contributions

• llama.cpp Open-source code

  CODEOWNERS lists bartowski1182 with one other contributor for ggml/src/ggml-cpu/tiled. Commits authored from September 2025 to September 2026 include IQ-quant correctness fixes and AVX2 prompt-processing speedups, a shared imatrix loader, per-slot context for llama-server, and support for architectures such as GLM 4.6 and AfmoeForCausalLM. The repository is MIT-licensed.³⁴⁶

• GGUF quantizations on Hugging Face Open weights

  Publishes quantized GGUF versions of open models under the bartowski account. The model card for Qwen3.8-27B-GGUF, for example, says the quants were made with llama.cpp release b10896 using imatrix calibration, and gives the repository's license as Apache-2.0.¹²

Timeline

1. Oct 28, 2023Published a first ExLlamaV2 quantization on the bartowski Hugging Face account.⁵
2. May 26, 2024Added Smaug 70B support to llama.cpp.⁴
3. May 9, 2025Added a --parse-special option to llama.cpp's imatrix tool for parsing special tokens.⁴
4. Jun 15, 2025Added llama.cpp support for Arcee AI's AFM model.⁴
5. Sep 30, 2025Added llama.cpp support for GLM 4.6.⁴
6. Nov 14, 2025Added AfmoeForCausalLM model support to llama.cpp.⁴
7. Jun 4, 2026Moved duplicated imatrix code in llama.cpp into a single shared imatrix loader.⁴
8. Aug 31, 2026Sped up large-batch prompt processing of IQ-quantized models on AVX2 CPUs in llama.cpp.⁴

Where they work

• Red Hat 8 people in the wiki
• llama.cpp project 12 people in the wiki

Related people

• Matt HicksAlso at Red Hat
• Robert ShawAlso at Red Hat
• Chris WrightAlso at Red Hat
• Jeff GeerlingAlso at Red Hat
• Stefanie ChirasAlso at Red Hat
• Oleksiy OstapenkoLanguage models · Inference and serving
• Furu WeiLanguage models · Inference and serving
• Waseem AlShikhLanguage models · Inference and serving
• Younes BelkadaLanguage models · Inference and serving
• Azalia MirhoseiniLanguage models · Inference and serving