Upload model

config.json

@@ -1,6 +1,6 @@
 {
   "_commit_hash": null,
-  "_name_or_path": "/home/arthur_huggingface_co/transformers/jukebox-5b-lyrics-converted",
+  "_name_or_path": "jukebox-5b-lyrics",
   "architectures": [
     "JukeboxModel"
   ],
@@ -26,32 +26,69 @@
     "attn_res_scale": false,
     "blocks": 64,
     "conv_res_scale": null,
-    "depth": 79,
     "emb_dropout": 0,
-    "embed_dim": 2048,
-    "encoder_attention_multiplier": 0.25,
-    "encoder_attention_pattern": "RawColumnPreviousRowAttention",
-    "encoder_attn_dropout": 0.0,
-    "encoder_attn_res_scale": false,
-    "encoder_blocks": 32,
-    "encoder_depth": 18,
-    "encoder_emb_dropout": 0.0,
-    "encoder_heads": 4,
-    "encoder_init_scale": 0.1,
-    "encoder_loss_fraction": [
-      0.4,
-      0.0,
-      0.0
-    ],
-    "encoder_mlp_multiplier": 1.0,
-    "encoder_n_vocab": 80,
-    "encoder_resid_dropout": 0.0,
-    "encoder_spread": null,
-    "encoder_width": 1280,
-    "encoder_zero_out": false,
+    "encoder_config": {
+      "act_fn": "quick_gelu",
+      "alignment_head": 2,
+      "alignment_layer": 68,
+      "attention_multiplier": 0.25,
+      "attention_pattern": "RawColumnPreviousRowAttention",
+      "attn_dropout": 0.0,
+      "attn_res_scale": false,
+      "blocks": 32,
+      "conv_res_scale": null,
+      "emb_dropout": 0.0,
+      "encoder_config": null,
+      "encoder_loss_fraction": 0.4,
+      "hidden_size": 1280,
+      "init_scale": 0.1,
+      "is_encoder_decoder": false,
+      "level": 0,
+      "lyric_vocab_size": 80,
+      "mask": true,
+      "max_duration": 600,
+      "max_nb_genres": 1,
+      "merged_decoder": false,
+      "metadata_conditioning": true,
+      "metadata_dims": [
+        604,
+        7898
+      ],
+      "min_duration": 0,
+      "mlp_multiplier": 1.0,
+      "model_type": "jukebox_prior",
+      "music_vocab_size": 2048,
+      "n_ctx": 6144,
+      "n_heads": 4,
+      "nb_relevant_lyric_tokens": 384,
+      "num_layers": 18,
+      "res_conv_depth": 3,
+      "res_conv_width": 128,
+      "res_convolution_multiplier": 1,
+      "res_dilation_cycle": null,
+      "res_dilation_growth_rate": 1,
+      "res_downs_t": [
+        3,
+        2,
+        2
+      ],
+      "res_strides_t": [
+        2,
+        2,
+        2
+      ],
+      "resid_dropout": 0.0,
+      "sampling_rate": 44100,
+      "spread": null,
+      "timing_dims": 64,
+      "zero_out": false
+    },
+    "encoder_loss_fraction": 0.4,
+    "hidden_size": 4800,
     "init_scale": 0.2,
     "is_encoder_decoder": false,
-    "lyric_conditioning": true,
+    "level": 0,
+    "lyric_vocab_size": 80,
     "mask": true,
     "max_duration": 600.0,
     "max_nb_genres": 5,
@@ -63,10 +100,12 @@
     ],
     "min_duration": 23.8,
     "mlp_multiplier": 1.0,
-    "model_type": "jukebox",
+    "model_type": "jukebox_prior",
+    "music_vocab_size": 2048,
     "n_ctx": 8192,
     "n_heads": 8,
     "nb_relevant_lyric_tokens": 512,
+    "num_layers": 79,
     "res_conv_depth": null,
     "res_conv_width": null,
     "res_convolution_multiplier": null,
@@ -86,8 +125,6 @@
     "sampling_rate": 44100,
     "spread": null,
     "timing_dims": 128,
-    "transformers_version": "4.25.0.dev0",
-    "width": 4800,
     "zero_out": false
   },
   "prior_1": {
@@ -100,32 +137,15 @@
     "attn_res_scale": false,
     "blocks": 64,
     "conv_res_scale": true,
-    "depth": 72,
     "emb_dropout": 0,
     "embed_dim": 2048,
-    "encoder_attention_multiplier": null,
-    "encoder_attention_pattern": null,
-    "encoder_attn_dropout": null,
-    "encoder_attn_res_scale": false,
-    "encoder_blocks": null,
-    "encoder_depth": null,
-    "encoder_emb_dropout": null,
-    "encoder_heads": null,
-    "encoder_init_scale": null,
-    "encoder_loss_fraction": [
-      0.4,
-      0.0,
-      0.0
-    ],
-    "encoder_mlp_multiplier": null,
-    "encoder_n_vocab": 79,
-    "encoder_resid_dropout": null,
-    "encoder_spread": null,
-    "encoder_width": null,
-    "encoder_zero_out": null,
+    "encoder_config": null,
+    "encoder_loss_fraction": 0.4,
+    "hidden_size": 1920,
     "init_scale": 1,
     "is_encoder_decoder": false,
-    "lyric_conditioning": true,
+    "level": 1,
+    "lyric_vocab_size": 80,
     "mask": true,
     "max_duration": 600.0,
     "max_nb_genres": 5,
@@ -137,10 +157,12 @@
     ],
     "min_duration": 23.8,
     "mlp_multiplier": 1.0,
-    "model_type": "jukebox",
+    "model_type": "jukebox_prior",
+    "music_vocab_size": 2048,
     "n_ctx": 8192,
     "n_heads": 1,
     "nb_relevant_lyric_tokens": 0,
+    "num_layers": 72,
     "res_conv_depth": 16,
     "res_conv_width": 1024,
     "res_convolution_multiplier": 1,
@@ -160,8 +182,6 @@
     "sampling_rate": 44100,
     "spread": null,
     "timing_dims": 128,
-    "transformers_version": "4.25.0.dev0",
-    "width": 1920,
     "zero_out": false
   },
   "prior_2": {
@@ -174,32 +194,15 @@
     "attn_res_scale": false,
     "blocks": 64,
     "conv_res_scale": false,
-    "depth": 72,
     "emb_dropout": 0,
     "embed_dim": 2048,
-    "encoder_attention_multiplier": null,
-    "encoder_attention_pattern": null,
-    "encoder_attn_dropout": null,
-    "encoder_attn_res_scale": false,
-    "encoder_blocks": null,
-    "encoder_depth": null,
-    "encoder_emb_dropout": null,
-    "encoder_heads": null,
-    "encoder_init_scale": null,
-    "encoder_loss_fraction": [
-      0.4,
-      0.0,
-      0.0
-    ],
-    "encoder_mlp_multiplier": null,
-    "encoder_n_vocab": 79,
-    "encoder_resid_dropout": null,
-    "encoder_spread": null,
-    "encoder_width": null,
-    "encoder_zero_out": null,
+    "encoder_config": null,
+    "encoder_loss_fraction": 0.4,
+    "hidden_size": 1920,
     "init_scale": 1,
     "is_encoder_decoder": false,
-    "lyric_conditioning": false,
+    "level": 2,
+    "lyric_vocab_size": 80,
     "mask": true,
     "max_duration": 600.0,
     "max_nb_genres": 5,
@@ -211,10 +214,12 @@
     ],
     "min_duration": 23.8,
     "mlp_multiplier": 1.0,
-    "model_type": "jukebox",
+    "model_type": "jukebox_prior",
+    "music_vocab_size": 2048,
     "n_ctx": 8192,
     "n_heads": 1,
     "nb_relevant_lyric_tokens": 0,
+    "num_layers": 72,
     "res_conv_depth": 16,
     "res_conv_width": 1024,
     "res_convolution_multiplier": 1,
@@ -234,8 +239,6 @@
     "sampling_rate": 44100,
     "spread": null,
     "timing_dims": 128,
-    "transformers_version": "4.25.0.dev0",
-    "width": 1920,
     "zero_out": false
   },
   "sampling_rate": 44100,
@@ -244,45 +247,6 @@
   "transformers_version": null,
   "vqvae_config": {
     "act_fn": "relu",
-    "codebook_dimension": 2048,
-    "commit": 0.02,
-    "conv_input_shape": 1,
-    "conv_res_scale": false,
-    "embed_dim": 64,
-    "hop_fraction": [
-      0.125,
-      0.5,
-      0.5
-    ],
-    "levels": 3,
-    "lmu": 0.99,
-    "model_type": "",
-    "multipliers": [
-      2,
-      1,
-      1
-    ],
-    "res_conv_depth": 4,
-    "res_conv_width": 32,
-    "res_convolution_multiplier": 1,
-    "res_dilation_cycle": null,
-    "res_dilation_growth_rate": 3,
-    "res_downs_t": [
-      3,
-      2,
-      2
-    ],
-    "res_strides_t": [
-      2,
-      2,
-      2
-    ],
-    "sample_length": 1058304,
-    "transformers_version": "4.25.0.dev0"
-  },
-  "vqvae_config_dict": {
-    "act_fn": "relu",
-    "codebook_dimension": 2048,
     "commit": 0.02,
     "conv_input_shape": 1,
     "conv_res_scale": false,
@@ -292,14 +256,16 @@
       0.5,
       0.5
     ],
+    "init_scale": 0.2,
     "levels": 3,
     "lmu": 0.99,
-    "model_type": "",
+    "model_type": "jukebox_vqvae",
     "multipliers": [
       2,
       1,
       1
     ],
+    "nb_discrete_codes": 2048,
     "res_conv_depth": 4,
     "res_conv_width": 32,
     "res_convolution_multiplier": 1,
@@ -316,6 +282,7 @@
       2
     ],
     "sample_length": 1058304,
-    "transformers_version": "4.24.0.dev0"
+    "transformers_version": "4.25.0.dev0",
+    "zero_out": false
   }
 }