# Ollama with the MLX runtime (macOS): from this folder, run
#   ollama create simplixity-writer:4b-v1-q4_k_m -f Modelfile --quantize q4_K_M
# Ollama on Linux and Windows cannot import safetensors: use a GGUF file from
# stneric/simplixity-writer-4b-GGUF and replace "FROM ." with its full path.
# RENDERER and PARSER are Ollama's Qwen3.5 chat format. Without them the system
# prompt never reaches the model. num_ctx is the window the model was trained
# for; Simplixity reads it and fits the sources to it. draft_num_predict 2
# turns on speculative decoding with the model's MTP layer (Ollama 0.40+).
FROM .
RENDERER qwen3.5
PARSER qwen3.5
PARAMETER num_ctx 8192
PARAMETER draft_num_predict 2