load models to VRAM when using `--lowram` param

load models to VRM instead of RAM (for machines which have bigger VRM than RAM such as free Google Colab server)

1 files changed, 13 insertions, 2 deletions

diff --git a/modules/sd_models.py b/modules/sd_models.py
index 0a55b4c3..78a198b9 100644
--- a/modules/sd_models.py
+++ b/modules/sd_models.py
@@ -134,7 +134,12 @@ def load_model_weights(model, checkpoint_info):
 

     print(f"Loading weights [{sd_model_hash}] from {checkpoint_file}")

 

-    pl_sd = torch.load(checkpoint_file, map_location="cpu")

+    if shared.cmd_opts.lowram:

+        print("Load to VRAM if GPU is available (low RAM)")

+        pl_sd = torch.load(checkpoint_file)

+    else:

+        pl_sd = torch.load(checkpoint_file, map_location="cpu")

+

     if "global_step" in pl_sd:

         print(f"Global Step: {pl_sd['global_step']}")

 

@@ -158,7 +163,13 @@ def load_model_weights(model, checkpoint_info):
 

     if os.path.exists(vae_file):

         print(f"Loading VAE weights from: {vae_file}")

-        vae_ckpt = torch.load(vae_file, map_location="cpu")

+

+        if shared.cmd_opts.lowram:

+            print("Load to VRAM if GPU is available (low RAM)")

+            vae_ckpt = torch.load(vae_file)

+        else:

+            vae_ckpt = torch.load(vae_file, map_location="cpu")

+

         vae_dict = {k: v for k, v in vae_ckpt["state_dict"].items() if k[0:4] != "loss"}

 

         model.first_stage_model.load_state_dict(vae_dict)