diff --git a/comfy/model_patcher.py b/comfy/model_patcher.py
index cdf2893955e6..d888dbcfb08a 100644
--- a/comfy/model_patcher.py
+++ b/comfy/model_patcher.py
@@ -1400,7 +1400,7 @@ def state_dict_for_saving(self, clip_state_dict=None, vae_state_dict=None, clip_
                 continue
             key = "diffusion_model." + k
             unet_state_dict[k] = LazyCastingParam(self, key, comfy.utils.get_attr(self.model, key))
-        return self.model.state_dict_for_saving(unet_state_dict)
+        return self.model.state_dict_for_saving(unet_state_dict, clip_state_dict=clip_state_dict, vae_state_dict=vae_state_dict, clip_vision_state_dict=clip_vision_state_dict)
 
     def __del__(self):
         self.unpin_all_weights()
diff --git a/comfy/text_encoders/ace15.py b/comfy/text_encoders/ace15.py
index 73d71067174d..fce2b67cec97 100644
--- a/comfy/text_encoders/ace15.py
+++ b/comfy/text_encoders/ace15.py
@@ -19,6 +19,7 @@ def sample_manual_loop_no_classes(
     min_tokens: int = 1,
     max_new_tokens: int = 2048,
     audio_start_id: int = 151669,  # The cutoff ID for audio codes
+    audio_end_id: int = 215669,
     eos_token_id: int = 151645,
 ):
     device = model.execution_device
@@ -60,6 +61,7 @@ def sample_manual_loop_no_classes(
         remove_logit_value = torch.finfo(cfg_logits.dtype).min
         # Only generate audio tokens
         cfg_logits[:, :audio_start_id] = remove_logit_value
+        cfg_logits[:, audio_end_id:] = remove_logit_value
 
         if eos_token_id is not None and eos_token_id < audio_start_id and min_tokens < step:
             cfg_logits[:, eos_token_id] = eos_score
diff --git a/comfy/utils.py b/comfy/utils.py
index c1b5368334ef..1337e2205a04 100644
--- a/comfy/utils.py
+++ b/comfy/utils.py
@@ -82,14 +82,12 @@ def scalar(*args, **kwargs):
 def load_safetensors(ckpt):
     f = open(ckpt, "rb")
     mapping = mmap.mmap(f.fileno(), 0, access=mmap.ACCESS_READ)
+    mv = memoryview(mapping)
 
     header_size = struct.unpack("<Q", mapping[:8])[0]
     header = json.loads(mapping[8:8+header_size].decode("utf-8"))
 
-    with warnings.catch_warnings():
-        #We are working with read-only RAM by design
-        warnings.filterwarnings("ignore", message="The given buffer is not writable")
-        data_area = torch.frombuffer(mapping, dtype=torch.uint8)[8 + header_size:]
+    mv = mv[8 + header_size:]
 
     sd = {}
     for name, info in header.items():
@@ -97,7 +95,13 @@ def load_safetensors(ckpt):
             continue
 
         start, end = info["data_offsets"]
-        sd[name] = data_area[start:end].view(_TYPES[info["dtype"]]).view(info["shape"])
+        if start == end:
+            sd[name] = torch.empty(info["shape"], dtype =_TYPES[info["dtype"]])
+        else:
+            with warnings.catch_warnings():
+                #We are working with read-only RAM by design
+                warnings.filterwarnings("ignore", message="The given buffer is not writable")
+                sd[name] = torch.frombuffer(mv[start:end], dtype=_TYPES[info["dtype"]]).view(info["shape"])
 
     return sd, header.get("__metadata__", {}),
 
diff --git a/comfyui_version.py b/comfyui_version.py
index 2e2c12ced561..5d296cd1b680 100644
--- a/comfyui_version.py
+++ b/comfyui_version.py
@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.12.1"
+__version__ = "0.12.2"
diff --git a/pyproject.toml b/pyproject.toml
index c21ee03f1187..1ddcc3596b2c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.12.1"
+version = "0.12.2"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.10"