# /etc/conf.d/fastflowlm
# Configuration for the FastFlowLM (FLM) OpenRC service.

# REQUIRED. Local user the daemon runs as. Must have NPU access
# (the amdxdna kernel stack is typically world-accessible via
# /dev/accel/accel0; if you restricted permissions, add this user
# to the relevant group). The service refuses to start if unset.
FLM_USER=""

# Main chat-side model loaded at startup. List available models
# with `flm list`. Default is the smallest qwen3 (NPU-fastest).
#FLM_MODEL="qwen3:0.6b"

# Listener address. FLM defaults to 127.0.0.1; leave FLM_PORT
# empty to use FLM's deterministic default port (verify with
# `flm port`).
#FLM_HOST="127.0.0.1"
#FLM_PORT=""

# Co-loaded sidecars on the NPU. 1 enables, 0 disables.
#   FLM_ASR=1   — Whisper-V3 Turbo for STT
#   FLM_EMBED=1 — embed-gemma:300m for embeddings
# Both can run alongside the chat model without unloading.
#FLM_ASR="1"
#FLM_EMBED="1"

# Power mode: powersaver, balanced, performance, turbo.
# performance is the sane default; turbo runs the NPU harder.
#FLM_PMODE="performance"

# Log file (opened by supervise-daemon as root before drop-priv).
#FLM_LOG="/var/log/flm.log"

# Extra CLI args appended before the model tag. Examples:
#   -c 8192                  # context length
#   --quiet
#FLM_EXTRA_OPTS=""