fix: disable --fit in frank-default-large serve (full offload fits; fit aborted on explicit n-gpu-layers)

scripts/serve-llama-frank-large.sh

@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
 # Serve HauhauCS Qwen3.6-27B-Uncensored-HauhauCS-Aggressive:Q8_K_P
-# with DFlash draft + ngram-mod speculative decoding on dankclaw
-# (AMD Strix Halo gfx1151, ROCm or Vulkan backend).
+# with DFlash draft speculative decoding (draft-dflash) on dankclaw
+# (AMD Strix Halo gfx1151, ROCm backend).
 # Runtime: llama.cpp llama-server.
 #
 # Tunables come from frank-default-large.env in this directory unless the
@@ -61,6 +61,7 @@ exec "$LLAMA_SERVER" \
     --spec-ngram-mod-n-match "$SPEC_NGRAM_MOD_N_MATCH" \
     --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" \