add usage tips

README.md

@@ -255,11 +255,14 @@ custom=$(
     --threads 8 \
     --host 127.0.0.1 \
     --port 8080 \
-    --no-mmap
-
-# for full offload onto GPU just add -ngl 99 and set threads to 1
+    --no-mmap \
+    --jinja
 ```
-If you have a properly fixed chat template, you can use it like this `--jinja --chat-template-file ./myFixedTemplate.jinja`.
+
+Tips:
+* for full offload onto GPU just add `-ngl 99` and use one thread with `--threads 1`
+* to save space on kv-cache use `-ctk q8_0` which is all you need given this is MLA
+* bring your own jinja chat template with `--jinja --chat-template-file ./myFixedTemplate.jinja`
 
 ## References
 * [ik_llama.cpp PR#995](https://github.com/ikawrakow/ik_llama.cpp/pull/995)