ling: batch 8192 (ubatch 3072); add -lv verbosity + log file for stop-investigation

scripts/frank-default-large.ling.env

@@ -32,8 +32,13 @@ DEVICE=Vulkan0
 N_GPU_LAYERS=all
 FLASH_ATTN=auto
 # Batch kept small-leaning for the tight memory budget; tune with CTX_SIZE.
-BATCH_SIZE=3072
+BATCH_SIZE=8192
 UBATCH_SIZE=3072
+
+# Diagnostics: high verbosity + a dedicated log file for investigating the
+# premature 'stop' reports. -lv 4 = verbose (ggml); 5 = debug.
+VERBOSITY=4
+LOG_FILE=/tmp/frank-ling.log
 # Sampling defaults (model card: reasoning-capable; leave temp/top to default)
 TEMPERATURE=1.0
 TOP_K=20

scripts/serve-ling-flash.sh

@@ -47,6 +47,8 @@ fi
 : "${MIN_P:=0.0}"
 : "${THREADS:=12}"
 : "${THREADS_HTTP:=1}"
+: "${VERBOSITY:=0}"      # ggml log verbosity 0-5; 4 = verbose, 5 = debug
+: "${LOG_FILE:=}"         # optional; if set, --log-file path
 
 # Fork binary; stock llama.cpp cannot load bailingmoe3.
 LLAMA_SERVER="${LLAMA_SERVER:-/home/dank/llama.cpp-ling/build/bin/llama-server}"
@@ -62,6 +64,7 @@ if [[ ! -f "$MODEL" ]]; then
 fi
 
 exec "$LLAMA_SERVER" \
+    $(if [[ -n "$LOG_FILE" ]]; then echo "--log-file $LOG_FILE"; fi) \
     --model "$MODEL" \
     --spec-type "$SPEC_TYPE" \
     --spec-draft-n-max "$SPEC_DRAFT_N_MAX" \
@@ -81,6 +84,7 @@ exec "$LLAMA_SERVER" \
     --min-p "$MIN_P" \
     --threads "$THREADS" \
     --threads-http "$THREADS_HTTP" \
+    -lv "$VERBOSITY" \
     --alias "$ALIAS" \
     --host "$HOST" \
     --port "$PORT"