Push model using huggingface_hub.

config.json

@@ -1,6 +1,6 @@
 {
   "audio_channels": 2,
-  "bottom_channels": 0,
+  "bottom_channels": 512,
   "cac": true,
   "channels": 48,
   "channels_time": null,
@@ -9,7 +9,7 @@
   "dconv_comp": 8,
   "dconv_depth": 2,
   "dconv_init": 0.001,
-  "dconv_mode": 1,
+  "dconv_mode": 3,
   "depth": 4,
   "emb_scale": 10,
   "emb_smooth": true,
@@ -17,7 +17,7 @@
   "freq_emb": 0.2,
   "growth": 2,
   "kernel_size": 8,
-  "multi_freqs": null,
+  "multi_freqs": [],
   "multi_freqs_depth": 3,
   "nfft": 4096,
   "norm_groups": 4,
@@ -25,7 +25,7 @@
   "rescale": 0.1,
   "rewrite": true,
   "samplerate": 44100,
-  "segment": 10,
+  "segment": 7.8,
   "sources": [
     "drums",
     "bass",
@@ -42,7 +42,7 @@
   ],
   "t_cape_mean_normalize": true,
   "t_cross_first": false,
-  "t_dropout": 0.0,
+  "t_dropout": 0.02,
   "t_emb": "sin",
   "t_gelu": true,
   "t_global_window": 100,
@@ -61,7 +61,7 @@
   "t_norm_in_group": false,
   "t_norm_out": true,
   "t_sin_random_shift": 0,
-  "t_sparse_attn_window": 500,
+  "t_sparse_attn_window": 400,
   "t_sparse_cross_attn": false,
   "t_sparse_self_attn": false,
   "t_sparsity": 0.95,