dankclaw: retire DSpark, use llama.cpp --spec-default (ngram-mod) preset; PARALLEL=2 trial

scripts/serve-llama-frank-large.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
-# Serve unsloth Qwen3.8-27B UD-Q8_K_XL on dankclaw with a DSpark
-# sidecar draft model (draft-dspark). ROCm backend.
+# 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
@@ -25,7 +25,7 @@ fi
 : "${PORT:=7770}"
 : "${CTX_SIZE:=262144}"   # Qwen3.8 native context (env may raise for parallel≥2)
 : "${PARALLEL:=2}"
-: "${SPEC_TYPE:=draft-mtp}"
+: "${SPEC_DEFAULT:=0}"   # 1: use --spec-default (ngram-mod preset) instead of --spec-type
 : "${SPEC_DRAFT_N_MAX:=10}"
 : "${SPEC_NGRAM_MOD_N_MIN:=48}"
 : "${SPEC_NGRAM_MOD_N_MAX:=64}"
@@ -52,29 +52,37 @@ if [[ ! -f "$MODEL" ]]; then
     exit 1
 fi
 
-# Draft-model flag applies only when a separate draft file is provided.
-# Qwen3.8-UD bundles MTP in-model; leave DRAFT_FILE empty for bundled MTP,
-# or set it to a DSpark/DFlash sidecar GGUF for a dedicated draft.
+# 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 [[ -n "$DRAFT_FILE" ]]; then
-    if [[ ! -f "$DRAFT" ]]; then
-        echo "missing required file: $DRAFT" >&2
+if [[ "${SPEC_DEFAULT}" == "1" ]]; then
+    SPEC_ARGS+=(--spec-default)
+else
+    if [[ -z "${SPEC_TYPE:-}" ]]; then
+        echo "SPEC_TYPE must be set when SPEC_DEFAULT=0" >&2
         exit 1
     fi
-    SPEC_ARGS+=(--spec-draft-model "$DRAFT")
-fi
-if [[ -n "${SPEC_DRAFT_P_MIN:-}" ]]; then
-    SPEC_ARGS+=(--spec-draft-p-min "$SPEC_DRAFT_P_MIN")
+    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" \
     "${SPEC_ARGS[@]}" \
-    --spec-type "$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" \
     --fit off \