#!/usr/bin/env bash
# Serve unsloth Qwen3.8-27B UD-Q8_K_XL on dankclaw with llama.cpp's
# --spec-default preset (ngram-mod). ROCm backend.
# Runtime: llama.cpp llama-server (>= b10411).
#
# Tunables come from frank-default-large.env in this directory unless the
# FRANK_LARGE_ENV_FILE environment variable overrides that path.
set -euo pipefail

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

# --- defaults (used when the env file does not set them) -----------------
: "${MODEL_DIR:=/home/dank/models}"
: "${MODEL_FILE:=Qwen3.8-27B-UD-Q8_K_XL.gguf}"
: "${DRAFT_FILE:=}"        # set: DSpark/DFlash/MTP/... sidecar GGUF in MODEL_DIR
: "${ALIAS:=frank-default-large}"
: "${HOST:=0.0.0.0}"
: "${PORT:=7770}"
: "${CTX_SIZE:=262144}"   # Qwen3.8 native context (env may raise for parallel≥2)
: "${PARALLEL:=2}"
: "${SPEC_DEFAULT:=0}"   # 1: use --spec-default (ngram-mod preset) instead of --spec-type
: "${MTP:=0}"            # 1: prepend --spec-type draft-mtp before --spec-default (MTP drafts, ngram fallback)
: "${SPEC_DRAFT_N_MAX:=10}"
: "${SPEC_NGRAM_MOD_N_MIN:=48}"
: "${SPEC_NGRAM_MOD_N_MAX:=64}"
: "${SPEC_NGRAM_MOD_N_MATCH:=24}"
: "${DEVICE:=ROCm0}"
: "${N_GPU_LAYERS:=all}"
: "${CACHE_TYPE_K:=q8_0}"
: "${CACHE_TYPE_V:=q8_0}"
: "${BATCH_SIZE:=8192}"
: "${UBATCH_SIZE:=2048}"
: "${TEMPERATURE:=1.0}"
: "${TOP_K:=20}"
: "${TOP_P:=0.95}"
: "${MIN_P:=0.0}"
: "${THREADS:=12}"
: "${THREADS_HTTP:=1}"

LLAMA_SERVER="${LLAMA_SERVER:-/home/dank/llama.cpp/build/bin/llama-server}"
MODEL="$MODEL_DIR/$MODEL_FILE"
DRAFT="$MODEL_DIR/$DRAFT_FILE"

if [[ ! -f "$MODEL" ]]; then
    echo "missing required file: $MODEL" >&2
    exit 1
fi

# Spec flags.  --spec-default is a preset that enables ngram-mod
# (n_match=24, n_min=48, n_max=64) and takes no draft model.  The explicit
# --spec-type / draft-model path is used when SPEC_DEFAULT is 0.
SPEC_ARGS=()
if [[ "${SPEC_DEFAULT}" == "1" ]]; then
    if [[ "${MTP}" == "1" ]]; then
        # MTP first (bundled in-model head), then the --spec-default ngram-mod
        # fallback.  Order matters: the chain drafts through impls in types
        # vector order.  --spec-type appends, so it must precede --spec-default.
        SPEC_ARGS+=(--spec-type draft-mtp)
    fi
    SPEC_ARGS+=(--spec-default)
else
    if [[ -z "${SPEC_TYPE:-}" ]]; then
        echo "SPEC_TYPE must be set when SPEC_DEFAULT=0" >&2
        exit 1
    fi
    SPEC_ARGS+=(--spec-type "$SPEC_TYPE")
    if [[ -n "$DRAFT_FILE" ]]; then
        if [[ ! -f "$DRAFT" ]]; then
            echo "missing required file: $DRAFT" >&2
            exit 1
        fi
        SPEC_ARGS+=(--spec-draft-model "$DRAFT")
    fi
    if [[ -n "${SPEC_DRAFT_P_MIN:-}" ]]; then
        SPEC_ARGS+=(--spec-draft-p-min "$SPEC_DRAFT_P_MIN")
    fi
    SPEC_ARGS+=(--spec-draft-n-max "$SPEC_DRAFT_N_MAX")
    SPEC_ARGS+=(--spec-ngram-mod-n-min "$SPEC_NGRAM_MOD_N_MIN")
    SPEC_ARGS+=(--spec-ngram-mod-n-max "$SPEC_NGRAM_MOD_N_MAX")
    SPEC_ARGS+=(--spec-ngram-mod-n-match "$SPEC_NGRAM_MOD_N_MATCH")
fi

exec "$LLAMA_SERVER" \
    --model "$MODEL" \
    "${SPEC_ARGS[@]}" \
    --device "$DEVICE" \
    --gpu-layers "$N_GPU_LAYERS" \
    --fit off \
    --cache-type-k "$CACHE_TYPE_K" \
    --cache-type-v "$CACHE_TYPE_V" \
    --ctx-size "$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 "$THREADS_HTTP" \
    --alias "$ALIAS" \
    --host "$HOST" \
    --port "$PORT"