Update example with possible yarn context extension

README.md

@@ -452,7 +452,7 @@ numactl -N "$SOCKET" -m "$SOCKET" \
 ./build/bin/llama-server \
     --model "$model"\
     --alias ubergarm/Ling-1T-GGUF \
-    --ctx-size 65536 \
+    --ctx-size 32768 \
     -fa -fmoe -ger \
     -ctk q8_0 -ctv q8_0 \
     -ub 4096 -b 4096 \
@@ -471,6 +471,15 @@ numactl -N "$SOCKET" -m "$SOCKET" \
 
 # optional use this once after downloading to confirm good files
     --validate-quants
+
+# NOTE: if you *really* want over the official 32k supported context consider these options:
+#  64k
+--ctx-size  65536 --rope-scaling yarn --rope-scale 2 --yarn-orig-ctx 32768 --override-kv bailingmoe2.context_length=int:65536
+# 128k (longer extension likely reduces quality of output, always use minimum context required)
+--ctx-size 131072 --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 32768 --override-kv bailingmoe2.context_length=int:131072
+# Details:
+# * https://github.com/ikawrakow/ik_llama.cpp/discussions/839#discussioncomment-14745117
+# * https://github.com/ikawrakow/ik_llama.cpp/issues/873
 ```
 
 ## References