fix for hf

.gitignore

@@ -0,0 +1,144 @@
+run_*.sh
+log/
+saves
+saves/
+weights/
+weights
+output/
+output
+pretrained/
+workspace
+workspace/
+ext_weights/
+ext_weights
+.checkpoints/
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/

app.py

@@ -5,6 +5,13 @@ from pathlib import Path
 import gradio as gr
 import torch
 import torchaudio
+import os
+
+try:
+    import mmaudio
+except ImportError:
+    os.system("pip install -e .")
+    import mmaudio
 
 from mmaudio.eval_utils import (ModelConfig, all_model_cfg, generate, load_video, make_video,
                                 setup_eval_logging)

mmaudio/ext/autoencoder/autoencoder.py

@@ -19,7 +19,7 @@ class AutoEncoderModule(nn.Module):
         super().__init__()
         self.vae: VAE = get_my_vae(mode).eval()
         vae_state_dict = torch.load(vae_ckpt_path, weights_only=True, map_location='cpu')
-        self.vae.load_state_dict(vae_state_dict)
+        self.vae.load_state_dict(vae_state_dict, strict=False)
         self.vae.remove_weight_norm()
 
         if mode == '16k':

mmaudio/ext/autoencoder/vae.py

@@ -75,11 +75,15 @@ class VAE(nn.Module):
         super().__init__()
 
         if data_dim == 80:
-            self.data_mean = nn.Buffer(torch.tensor(DATA_MEAN_80D, dtype=torch.float32).cuda())
-            self.data_std = nn.Buffer(torch.tensor(DATA_STD_80D, dtype=torch.float32).cuda())
+            # self.data_mean = torch.tensor(DATA_MEAN_80D, dtype=torch.float32).cuda()
+            # self.data_std = torch.tensor(DATA_STD_80D, dtype=torch.float32).cuda()
+            self.register_buffer('data_mean', torch.tensor(DATA_MEAN_80D, dtype=torch.float32))
+            self.register_buffer('data_std', torch.tensor(DATA_STD_80D, dtype=torch.float32))
         elif data_dim == 128:
-            self.data_mean = nn.Buffer(torch.tensor(DATA_MEAN_128D, dtype=torch.float32).cuda())
-            self.data_std = nn.Buffer(torch.tensor(DATA_STD_128D, dtype=torch.float32).cuda())
+            # torch.tensor(DATA_MEAN_128D, dtype=torch.float32).cuda()
+            # self.data_std = torch.tensor(DATA_STD_128D, dtype=torch.float32).cuda()
+            self.register_buffer('data_mean', torch.tensor(DATA_MEAN_128D, dtype=torch.float32))
+            self.register_buffer('data_std', torch.tensor(DATA_STD_128D, dtype=torch.float32))
 
         self.data_mean = self.data_mean.view(1, -1, 1)
         self.data_std = self.data_std.view(1, -1, 1)

mmaudio/model/embeddings.py

@@ -21,12 +21,11 @@ class TimestepEmbedder(nn.Module):
         assert dim % 2 == 0, 'dim must be even.'
 
         with torch.autocast('cuda', enabled=False):
-            self.freqs = nn.Buffer(
+            self.freqs = (
                 1.0 / (10000**(torch.arange(0, frequency_embedding_size, 2, dtype=torch.float32) /
-                               frequency_embedding_size)),
-                persistent=False)
+                               frequency_embedding_size)))
             freq_scale = 10000 / max_period
-            self.freqs = freq_scale * self.freqs
+            self.freqs = nn.Parameter(freq_scale * self.freqs)
 
     def timestep_embedding(self, t):
         """

mmaudio/model/networks.py

@@ -166,8 +166,10 @@ class MMAudio(nn.Module):
                                           self._clip_seq_len,
                                           device=self.device)
 
-        self.latent_rot = nn.Buffer(latent_rot, persistent=False)
-        self.clip_rot = nn.Buffer(clip_rot, persistent=False)
+        # self.latent_rot = latent_rot.to(self.device)
+        # self.clip_rot = clip_rot.to(self.device)
+        self.register_buffer('latent_rot', latent_rot)
+        self.register_buffer('clip_rot', clip_rot)
 
     def update_seq_lengths(self, latent_seq_len: int, clip_seq_len: int, sync_seq_len: int) -> None:
         self._latent_seq_len = latent_seq_len
@@ -346,7 +348,7 @@ class MMAudio(nn.Module):
         if 'clip_rot' in src_dict:
             del src_dict['clip_rot']
 
-        self.load_state_dict(src_dict, strict=True)
+        self.load_state_dict(src_dict, strict=False)
 
     @property
     def device(self) -> torch.device:

requirements.txt

@@ -1,4 +1,4 @@
-torch >= 2.5.1
+torch == 2.4.0
 torchaudio
 torchvision
 python-dotenv