# my-deepseek server settings. Restart (./stop.sh && ./start.sh) after editing.

# Listen address. 127.0.0.1 = this RHEL host only (use an SSH tunnel from Windows).
# 0.0.0.0 = reachable from the LAN (open the firewall port too; Ollama has NO auth).
OLLAMA_HOST=127.0.0.1:11550

# Model tag served (DeepSeek-R1-Distill-Qwen-14B, Q4_K_M, ~9 GB)
MODEL=deepseek-r1:14b

# Context window in tokens. R1 "thinks" before answering, so keep >= 8192.
OLLAMA_CONTEXT_LENGTH=8192
OLLAMA_NUM_PARALLEL=1
OLLAMA_MAX_LOADED_MODELS=1
OLLAMA_KEEP_ALIVE=30m
OLLAMA_NO_CLOUD=1