preparing release

README.md

@@ -39,6 +39,8 @@ width="800px"></p>
 
 See the [release notes](./docs/release.md) for more details.
 
+- 17/12/2021: Releasing v3.0.3: bug fixes  (thanks @keunwoochoi), memory drastically
+    reduced on GPU (thanks @famzah) and new multi-core evaluation on CPU (`-j` flag).
 - 12/11/2021: Releasing **Demucs v3** with hybrid domain separation. Strong improvements
 	on all sources. This is the model that won Sony MDX challenge.
 - 11/05/2021: Adding support for MusDB-HQ and arbitrary wav set, for the MDX challenge. For more information
@@ -179,6 +181,8 @@ slower. Don't use it unless you have a GPU.
 The `--overlap` option controls the amount of overlap between prediction windows (for Demucs one window is 10 seconds).
 Default is 0.25 (i.e. 25%) which is probably fine.
 
+The `-j` flag allow to specify a number of parallel jobs (e.g. `demucs -j 2 myfile.mp3`).
+This will multiply by the same amount the RAM used so be careful!
 
 ### Memory requirements for GPU acceleration
 

demucs/__init__.py

@@ -4,4 +4,4 @@
 # This source code is licensed under the license found in the
 # LICENSE file in the root directory of this source tree.
 
-__version__ = "3.0.2a2"
+__version__ = "3.0.2"

demucs/apply.py

@@ -116,7 +116,6 @@ def tensor_chunk(tensor_or_chunk):
         return TensorChunk(tensor_or_chunk)
 
 
-
 def apply_model(model, mix, shifts=1, split=True,
                 overlap=0.25, transition_power=1., progress=False, device=None,
                 num_workers=0, pool=None):
@@ -139,10 +138,11 @@ def apply_model(model, mix, shifts=1, split=True,
     """
     if device is None:
         device = mix.device
-    if pool is None and num_workers > 0:
-        pool = ThreadPoolExecutor(num_workers)
-    elif pool is None:
-        pool = DummyPoolExecutor()
+    if pool is None:
+        if num_workers > 0:
+            pool = ThreadPoolExecutor(num_workers)
+        else:
+            pool = DummyPoolExecutor()
     kwargs = {
         'shifts': shifts,
         'split': split,

demucs/separate.py

@@ -90,12 +90,8 @@ def main():
     parser.add_argument("-j", "--jobs",
                         default=None,
                         type=int,
-                        const=-1,
-                        nargs='?',
                         help="Number of jobs. This can increase memory usage but will "
-                             "be much faster when multiple cores are available. "
-                             "Either pass a number of parallel workers or half the nb of CPU will "
-                             "be selected (near optimal in most cases because of SSE).")
+                             "be much faster when multiple cores are available.")
 
     args = parser.parse_args()
 
@@ -127,7 +123,8 @@ def main():
         ref = wav.mean(0)
         wav = (wav - ref.mean()) / ref.std()
         sources = apply_model(model, wav[None], device=args.device, shifts=args.shifts,
-                              split=args.split, overlap=args.overlap, progress=True, num_workers=2)[0]
+                              split=args.split, overlap=args.overlap, progress=True,
+                              num_workers=args.jobs)[0]
         sources = sources * ref.std() + ref.mean()
 
         track_folder = out / track.name.rsplit(".", 1)[0]

demucs/utils.py

@@ -118,7 +118,7 @@ def __init__(self, func, *args, **kwargs):
         def result(self):
             return self.func(*self.args, **self.kwargs)
 
-    def __init__(self, workers):
+    def __init__(self, workers=0):
         pass
 
     def submit(self, func, *args, **kwargs):

docs/release.md

@@ -6,6 +6,8 @@ Fix bug in weights used for different sources. Thanks @keunwoochoi for the repor
 
 Improving drastically memory usage on GPU for long files. Thanks a lot @famzah for providing this.
 
+Adding multithread evaluation on CPU (`-j` option).
+
 ## V3.0.1, 12th of November 2021
 
 Release of Demucs v3, featuring hybrid domain separation and much more.