From 9fdec02914d23aadf8152c8e4829c7a7b30f12f0 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Wed, 30 Apr 2025 14:49:52 +0800
Subject: [PATCH] unclamp glm4 in debug

---
 gpttype_adapter.cpp | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 6d649af94..d83628e98 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -1908,8 +1908,14 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
         clamped_max_context_length = 16384;
     }
     if (isGguf && file_format_meta.model_architecture == GGUFArch::ARCH_GLM4 && kcpp_data->n_batch > 16) {
-        printf("GLM-4 is broken on larger batch sizes. Clamping batch size to 16.\n");
-        kcpp_data->n_batch = kcpp_data->n_ubatch = 16;
+        if(debugmode==1)
+        {
+            printf("GLM-4 is broken on larger batch sizes. Clamp ignored in debug.\n");
+        } else {
+            printf("GLM-4 is broken on larger batch sizes. Clamping batch size to 16.\n");
+            kcpp_data->n_batch = kcpp_data->n_ubatch = 16;
+        }
+
     }
 
     kcpp_data->n_ctx = clamped_max_context_length;