Config update at step 1250

config.json

@@ -0,0 +1,77 @@
+{
+  "modality_dims": {
+    "clip_l": 768,
+    "clip_g": 1280,
+    "t5_xl_l": 2048,
+    "t5_xl_g": 2048
+  },
+  "modality_seq_lens": {
+    "clip_l": 77,
+    "clip_g": 77,
+    "t5_xl_l": 512,
+    "t5_xl_g": 512
+  },
+  "binding_config": {
+    "clip_l": {
+      "t5_xl_l": 0.3
+    },
+    "clip_g": {
+      "t5_xl_g": 0.3
+    },
+    "t5_xl_l": {},
+    "t5_xl_g": {}
+  },
+  "latent_dim": 2048,
+  "seq_len": 77,
+  "encoder_layers": 3,
+  "decoder_layers": 3,
+  "hidden_dim": 1024,
+  "dropout": 0.1,
+  "fusion_strategy": "adaptive_cantor",
+  "fusion_heads": 8,
+  "fusion_dropout": 0.1,
+  "cantor_depth": 8,
+  "cantor_local_window": 3,
+  "alpha_init": 1.0,
+  "beta_init": 0.3,
+  "alpha_lr_scale": 0.1,
+  "beta_lr_scale": 1.0,
+  "beta_kl": 0.1,
+  "beta_reconstruction": 1.0,
+  "beta_cross_modal": 0.05,
+  "beta_alpha_regularization": 0.01,
+  "recon_type": "mse",
+  "modality_recon_weights": {
+    "clip_l": 1.0,
+    "clip_g": 1.0,
+    "t5_xl_l": 0.3,
+    "t5_xl_g": 0.3
+  },
+  "use_kl_annealing": true,
+  "kl_anneal_epochs": 10,
+  "kl_start_beta": 0.0,
+  "batch_size": 8,
+  "num_epochs": 100,
+  "learning_rate": 0.0001,
+  "weight_decay": 1e-05,
+  "gradient_clip": 1.0,
+  "use_scheduler": true,
+  "scheduler_type": "cosine",
+  "num_samples": 10000,
+  "synthetic_ratio": 0.15,
+  "checkpoint_dir": "./checkpoints_lyra_adaptive_cantor",
+  "save_every": 1000,
+  "keep_last_n": 3,
+  "hf_repo": "AbstractPhil/vae-lyra-xl-adaptive-cantor",
+  "push_to_hub": true,
+  "push_every": 2000,
+  "auto_load_from_hub": true,
+  "use_wandb": false,
+  "wandb_project": "vae-lyra-adaptive-cantor",
+  "wandb_entity": null,
+  "log_every": 50,
+  "device": "cuda",
+  "mixed_precision": true,
+  "seed": 42,
+  "num_workers": 0
+}