From 7ce98859f178a97598046e00a9b6cb4677fe06c6 Mon Sep 17 00:00:00 2001
From: Developer BadStorm <developer@badstorm.xyz>
Date: Sat, 8 Aug 2026 16:58:46 +0000
Subject: [PATCH] Update containers/llamacpp/llamacpp.container

---
 containers/llamacpp/llamacpp.container | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/containers/llamacpp/llamacpp.container b/containers/llamacpp/llamacpp.container
index ca486e7..b15d94e 100644
--- a/containers/llamacpp/llamacpp.container
+++ b/containers/llamacpp/llamacpp.container
@@ -23,6 +23,14 @@ Environment=LLAMA_ARG_LOAD_MODE=none
 Environment=LLAMA_ARG_CTX_SIZE=131072
 Environment=LLAMA_ARG_HF_REPO=unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q2_K
 
+# Improvements
+Environment=LLAMA_ARG_CACHE_REUSE=256
+Environment=LLAMA_ARG_BATCH=4096
+Environment=LLAMA_ARG_UBATCH=1024
+Environment=LLAMA_ARG_CACHE_TYPE_K=q4_1
+Environment=LLAMA_ARG_CACHE_TYPE_V=q4_1
+#Environment=LLAMA_CHAT_TEMPLATE_KWARGS={"thinking":true,"reasoning_effort":"medium"}
+
 # MTP
 Environment=LLAMA_ARG_SPEC_TYPE=draft-mtp
 Environment=LLAMA_ARG_SPEC_DRAFT_N_MAX=6