#!/usr/bin/env bash
# Serve inclusionAI Ling-3.0-flash (Q6_K) on dankclaw as a TRIAL, replacing
# frank-default-large temporarily.
#
# Runtime: llama.cpp FORK (aetherbird bailingmoe3-support) - stock llama.cpp
# cannot load the bailingmoe3 architecture. Backend: Vulkan (RADV) on the
# Strix Halo APU; memory is the shared 128 GB pool (96 GiB UMA + GTT that
# Vulkan exposes as an ~111 GiB heap). MTP drafter is bundled in the same
# GGUF (block 42) - no separate --spec-draft-model needed.
#
# Tie-back: to restore the Qwen3.6-27B service, stop frank-dragoon-ling and
# start frank-default-large (its env/unit are committed and unchanged).
#
# Tunables come from frank-default-large.ling.env in this directory unless
# FRANK_LING_ENV_FILE overrides that path.
set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
ENV_FILE="${FRANK_LING_ENV_FILE:-$SCRIPT_DIR/frank-default-large.ling.env}"
if [[ -f "$ENV_FILE" ]]; then
    set -a
    # shellcheck disable=SC1090
    source "$ENV_FILE"
    set +a
fi

# --- defaults -------------------------------------------------------------
: "${MODEL_DIR:=/home/dank/models}"
: "${MODEL_FILE:=Ling-3.0-flash-Q6_K.gguf}"
: "${ALIAS:=frank-default-large}"   # keep the same API alias for clients
: "${HOST:=0.0.0.0}"
: "${PORT:=7770}"
: "${CTX_SIZE:=32768}"
: "${PARALLEL:=2}"
: "${SPEC_TYPE:=draft-mtp}"         # MTP block is bundled in the model GGUF
: "${SPEC_DRAFT_N_MAX:=3}"
: "${SPEC_NGRAM_MOD_N_MIN:=48}"
: "${SPEC_NGRAM_MOD_N_MAX:=64}"
: "${SPEC_NGRAM_MOD_N_MATCH:=24}"
: "${DEVICE:=Vulkan0}"
: "${N_GPU_LAYERS:=all}"
: "${BATCH_SIZE:=512}"
: "${UBATCH_SIZE:=512}"
: "${TEMPERATURE:=1.0}"
: "${TOP_K:=20}"
: "${TOP_P:=0.95}"
: "${MIN_P:=0.0}"
: "${THREADS:=12}"
: "${THREADS_HTTP:=1}"
: "${VERBOSITY:=0}"      # ggml log verbosity 0-5; 4 = verbose, 5 = debug
: "${LOG_FILE:=}"         # optional; if set, --log-file path

# Fork binary; stock llama.cpp cannot load bailingmoe3.
LLAMA_SERVER="${LLAMA_SERVER:-/home/dank/llama.cpp-ling/build/bin/llama-server}"
MODEL="$MODEL_DIR/$MODEL_FILE"

if [[ ! -x "$LLAMA_SERVER" ]]; then
    echo "missing fork binary: $LLAMA_SERVER (build required)" >&2
    exit 1
fi
if [[ ! -f "$MODEL" ]]; then
    echo "missing model: $MODEL" >&2
    exit 1
fi

exec "$LLAMA_SERVER" \
    $(if [[ -n "$LOG_FILE" ]]; then echo "--log-file $LOG_FILE"; fi) \
    --model "$MODEL" \
    --spec-type "$SPEC_TYPE" \
    --spec-draft-n-max "$SPEC_DRAFT_N_MAX" \
    --spec-ngram-mod-n-min "$SPEC_NGRAM_MOD_N_MIN" \
    --spec-ngram-mod-n-max "$SPEC_NGRAM_MOD_N_MAX" \
    --spec-ngram-mod-n-match "$SPEC_NGRAM_MOD_N_MATCH" \
    --device "$DEVICE" \
    --gpu-layers "$N_GPU_LAYERS" \
    --flash-attn auto \
    --ctx-size "$CTX_SIZE" \
    --parallel "$PARALLEL" \
    --batch-size "$BATCH_SIZE" \
    --ubatch-size "$UBATCH_SIZE" \
    --temp "$TEMPERATURE" \
    --top-k "$TOP_K" \
    --top-p "$TOP_P" \
    --min-p "$MIN_P" \
    --threads "$THREADS" \
    --threads-http "$THREADS_HTTP" \
    -lv "$VERBOSITY" \
    --alias "$ALIAS" \
    --host "$HOST" \
    --port "$PORT"