Text-only Q4_K_M GGUF quantization of mistralai/Ministral-3-3B-Base-2512. No vision projector is included.

The original BF16 weights were first converted to F16 GGUF with the official llama.cpp converter at tag b9402 using convert_hf_to_gguf.py --outtype f16 --mistral-format. That F16 GGUF was then quantized with the official llama-quantize build 10360 using Q4_K_M. No importance matrix was used.

The file was successfully loaded with KeyType's vendored llama.cpp b9402 runtime. Its autocomplete profile self-check passed 6 of 6 checks, and an actual decode smoke test passed.