From 9c52a834038dd6d43f4c0dbe8d9b6bce1af2655c Mon Sep 17 00:00:00 2001
From: Wing Lian <wing.lian@gmail.com>
Date: Fri, 8 Sep 2023 02:06:12 -0400
Subject: [PATCH] load model faster w low_cpu_mem_usage

---
 src/axolotl/utils/models.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/axolotl/utils/models.py b/src/axolotl/utils/models.py
index 988ed29ba..3057d5da5 100644
--- a/src/axolotl/utils/models.py
+++ b/src/axolotl/utils/models.py
@@ -340,6 +340,7 @@ def load_model(
                 base_model,
                 config=config,
                 trust_remote_code=cfg.trust_remote_code or False,
+                low_cpu_mem_usage=True,
             ).half()
             model = tp.tensor_parallel(model, sharded=False)
         else: