From 0ca7e7a21981c7d42fa6a076072ea8883b4854df Mon Sep 17 00:00:00 2001
From: Adrian Gunnar Lauterer <adrian@lauterer.it>
Date: Thu, 2 Apr 2026 21:54:46 +0200
Subject: [PATCH] added gemma 4

---
 modules/llama-swap.nix | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/modules/llama-swap.nix b/modules/llama-swap.nix
index 1f07e2b..d0131b7 100644
--- a/modules/llama-swap.nix
+++ b/modules/llama-swap.nix
@@ -83,10 +83,29 @@
             aliases = [ "coder" ];
             macros = {
               hf_repo = "unsloth/GLM-4.7-Flash-REAP-23B-A3B-GGUF"; #Reap should allow more context in gpu mem
-              ctx = 64000; #supports 200k.
+              ctx = 48000; 
             };
           };
 
+          "gemma4" = {
+            cmd = "${llama-server} --port $\{PORT\} --host 0.0.0.0 --ctx-size $\{ctx\}  $\{kv_cache\} -ngl $\{ngl\} --hf-repo $\{hf_repo\}";
+            ttl = 900;
+            macros = {
+              hf_repo = "unsloth/gemma-4-26B-A4B-it-GGUF:UD-IQ3_XSS";
+              ctx = 64000;
+            };
+          };
+
+          "gemma4E4" = {
+            cmd = "${llama-server} --port $\{PORT\} --host 0.0.0.0 --ctx-size $\{ctx\}  $\{kv_cache\} -ngl $\{ngl\} --hf-repo $\{hf_repo\}";
+            ttl = 900;
+            macros = {
+              hf_repo = "unsloth/gemma-4-E4B-it-GGUF";
+              ctx = 128000;
+            };
+          };
+
+
           "ministal-3-8b-reasonning" = {
             cmd = "${llama-server} --port $\{PORT\} --host 0.0.0.0 --ctx-size $\{ctx\} -ngl $\{ngl\} --hf-repo $\{hf_repo\}";
             aliases = [ "ministral3" ];