Morph MORPH
SIGN IN SIGN UP
dhruvhead / Music-Source-Separation-Training UNCLAIMED
0 0 1 Python

COMMITS

main
January 23, 2026
K
Fixed problem when chunk_size > length
Kisel committed
9db76f5
January 20, 2026
U
BS Mamba2 model weights has been added
user-miner1 committed
690b6f8
U
BS Mamba2 model was updated
user-miner1 committed
3559543
January 7, 2026
R
Merge pull request #217 from kiselecheck/main
Roman Solovyev committed
c6a0226
K
Merge remote-tracking branch 'origin/main'
Kisel committed
bae6fa6
K
Merge remote-tracking branch 'origin/main'
Kisel committed
d4e4f3e
K
Add supporting active_stems_list
Kisel committed
ead9986
January 5, 2026
U
Fixed saving files bug
user-miner1 committed
0b501f9
K
return
Kisel committed
4048168
K
add annotations for dataset_type 6 and 7
Kisel committed
17d8427
R
Merge pull request #216 from kiselecheck/main
Roman Solovyev committed
d3b11e6
K
Merge branch 'ZFTurbo:main' into main
Kisel committed
25cf6c4
K
add annotations for dataset_type 6 and 7
Kisel committed
8c72c56
K
fix length track + cache for dataset_type 7
Kisel committed
bbf2dc9
January 2, 2026
R
Merge pull request #215 from kiselecheck/main
Roman Solovyev committed
3df209d
December 28, 2025
K
fix
Kisel committed
2c4e019
K
Merge branch 'main' into main
Kisel committed
ba1f67f
K
add safe_mode
Kisel committed
289917e
K
freeze layers
Kisel committed
da36684
K
add k_sdr metric
Kisel committed
6bdc2df
K
Revert "add k_sdr metric for normalized SDR scaling"
Kisel committed
2127d5e
K
Merge remote-tracking branch 'origin/main'
Kisel committed
9ebcfa4
K
add dataset_type 6 and dataset_type 7
Kisel committed
961f211
December 27, 2025
K
support multiple mixture formats and filter metrics by available instruments
Kisel committed
60afb26
K
refactor code for dataset_type 7 compatibility and add safe mode to skip corrupted batches
Kisel committed
1f89c47
K
new choices for arguments, logging
Kisel committed
15b517d
K
safe import muon
Kisel committed
16f5f27
K
add k_sdr metric for normalized SDR scaling
Kisel committed
7383fac
K
add recursive input processing and preserve folder structure in output
Kisel committed
d03f2e0
K
Now we can train only on stems present in the audio, not on non-existent stems
Kisel committed
3810e34