bitnetd.conf.example

# bitnetd.conf — BitNet Inference Daemon Configuration
# See bitnetd.conf(5) for full documentation.

[daemon]
# user = _bitnetd
# group = _bitnetd
pidfile = /tmp/bitnetd.pid
logfile = /tmp/bitnetd.log
loglevel = info

[model]
# Path to the GGUF model file (required).
# For local native-backend smoke testing with a BitNet-b1.58-2B-4T checkout:
#   path = ../bitnet-c11/models/BitNet-b1.58-2B-4T/ggml-model-i2_s.gguf
path = /path/to/ggml-model-i2_s.gguf
threads = 4
ctx_size = 2048
temperature = 0.8
top_k = 40
top_p = 0.95

[server]
listen = 127.0.0.1:8080
# unix_socket = /tmp/bitnetd.sock
max_connections = 64
request_timeout = 30
max_body_size = 65536
backlog = 128

[backend]
# Path to llama-server binary (subprocess backend only; default: searches $PATH)
# server_path = /usr/local/bin/llama-server
#
# For the native backend, server_path is ignored — inference runs in-process.
# Build with:  make BACKEND=native BITNET_C11_DIR=../bitnet-c11
# Smoke test:  BITNET_C11_DIR=../bitnet-c11 BITNET_MODEL=<path> make test-native
port = 18088

[security]
rate_limit = 60

[metrics]
enabled = true