dankclaw: add MTP=1 toggle - prepend draft-mtp before --spec-default ngram-mod fallback

scripts/frank-default-large.env

@@ -19,8 +19,10 @@ CTX_SIZE=524288
 PARALLEL=2
 # Spec: llama.cpp --spec-default preset = ngram-mod (n_match 24, n_min 48,
 # n_max 64).  No draft model.  Switched from DSpark 2026-08-15 after the
-# DSpark multi-slot ROCm crash.  See [[qwen3.8-spec-matrix]].
+# DSpark multi-slot ROCm crash.  MTP=1 prepends the bundled in-model MTP
+# head (draft-mtp) before the ngram-mod fallback.  See [[qwen3.8-spec-matrix]].
 SPEC_DEFAULT=1
+MTP=1
 SPEC_TYPE=
 DRAFT_FILE=
 SPEC_DRAFT_N_MAX=

scripts/frank-default-large.service

@@ -1,5 +1,5 @@
 [Unit]
-Description=llama.cpp Qwen3.8-27B UD-Q8_K_XL spec-default ngram-mod (frank-default-large)
+Description=llama.cpp Qwen3.8-27B UD-Q8_K_XL draft-mtp+ngram-mod (frank-default-large)
 After=network.target
 
 [Service]

scripts/serve-llama-frank-large.sh

@@ -26,6 +26,7 @@ fi
 : "${CTX_SIZE:=262144}"   # Qwen3.8 native context (env may raise for parallel≥2)
 : "${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}"
 : "${SPEC_NGRAM_MOD_N_MAX:=64}"
@@ -57,6 +58,12 @@ fi
 # --spec-type / draft-model path is used when SPEC_DEFAULT is 0.
 SPEC_ARGS=()
 if [[ "${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