Merge pull request #3818 from jwatzman/master

Reduce peak memory usage when changing models

1 files changed, 7 insertions, 4 deletions

diff --git a/modules/sd_models.py b/modules/sd_models.py
index 64d5ee0d..f86dc3ed 100644
--- a/modules/sd_models.py
+++ b/modules/sd_models.py
@@ -173,7 +173,9 @@ def load_model_weights(model, checkpoint_info):
             print(f"Global Step: {pl_sd['global_step']}")

 

         sd = get_state_dict_from_checkpoint(pl_sd)

-        missing, extra = model.load_state_dict(sd, strict=False)

+        del pl_sd

+        model.load_state_dict(sd, strict=False)

+        del sd

 

         if shared.cmd_opts.opt_channelslast:

             model.to(memory_format=torch.channels_last)

@@ -197,9 +199,10 @@ def load_model_weights(model, checkpoint_info):
 

         model.first_stage_model.to(devices.dtype_vae)

 

-        checkpoints_loaded[checkpoint_info] = model.state_dict().copy()

-        while len(checkpoints_loaded) > shared.opts.sd_checkpoint_cache:

-            checkpoints_loaded.popitem(last=False)  # LRU

+        if shared.opts.sd_checkpoint_cache > 0:

+            checkpoints_loaded[checkpoint_info] = model.state_dict().copy()

+            while len(checkpoints_loaded) > shared.opts.sd_checkpoint_cache:

+                checkpoints_loaded.popitem(last=False)  # LRU

     else:

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

         checkpoints_loaded.move_to_end(checkpoint_info)