GGUF Downloads
GGUF builds run under llama.cpp and everything built on it — Ollama, LM Studio, KoboldCpp, llamafile — on CPU, with no Python and no GPU required. At these model sizes that means a chat model that fits comfortably in a few hundred megabytes and runs on nearly anything.
| Repository | Params | Notes |
|---|---|---|
| Boris-1.3-125M-Instruct-GGUF | 125M | Current recommended build |
| Boris-1.3-75M-Instruct-GGUF | 77.4M | Smallest chat model |
| Boris-250M-Instruct-GGUF | 250M | Largest; imatrix quants |
| Boris-125M-Instruct-GGUF | 125M | Boris 1.0; imatrix quants |
| Boris-75M-Instruct-GGUF | 77.4M | Boris 1.0; imatrix quants |
Each repository lists its available quantizations on the Hugging Face files tab. Q8_0 is close to lossless and still tiny at this scale, so unless you have a specific reason to go lower, start there.
Pulling one
# llama.cpp resolves the repo and picks a quant for you llama-cli -hf opencerebral/Boris-1.3-125M-Instruct-GGUF -cnv # or fetch a specific file huggingface-cli download opencerebral/Boris-1.3-125M-Instruct-GGUF \ --include "*Q8_0.gguf" --local-dir .
Running instructions are on the Usage page.