frank-default-large: ubatch 512, disable speculative decoding

- UBATCH_SIZE: 4096 -> 512 (reduce GPU memory pressure)
- SPEC_TYPE: ngram-mod -> none (disable speculative decoding entirely)

scripts/serve-frank-large.sh

@@ -47,8 +47,8 @@ fi
 : "${LISTEN_PORT:=7770}"
 : "${CTX_SIZE:=401408}"
 : "${PARALLEL:=2}"
-: "${BATCH_SIZE:=4096}"
-: "${UBATCH_SIZE:=4096}"
+:: "${BATCH_SIZE:=4096}"
+:: "${UBATCH_SIZE:=512}"
 : "${NUM_THREADS:=16}"
 : "${THREAD_BATCH:=16}"
 : "${CACHE_TYPE_K:=f16}"
@@ -67,8 +67,8 @@ fi
 : "${PRESENCE_PENALTY:=1.5}"
 : "${REASONING:=on}"
 : "${REASONING_FORMAT:=deepseek}"
-: "${HSA_OVERRIDE_GFX_VERSION:=11.5.1}"
-: "${SPEC_TYPE:=ngram-mod}"
+:: "${HSA_OVERRIDE_GFX_VERSION:=11.5.1}"
+:: "${SPEC_TYPE:=none}"
 : "${DRAFT_MODEL_DIR:=/home/dank/models/mtp-Qwen3.8-Flash-Next-FAST}"
 : "${DRAFT_MODEL_FILE:=Qwen3.8-Flash-Next-MTP-ROCmFP4-FAST.gguf}"
 : "${SPEC_DRAFT_N_MAX:=3}"