Checkpoint: danbooru-50k-v1/epoch2_step4585_20251117_051214

danbooru-50k-v1/epoch2_step4585_20251117_051214/config.json

@@ -0,0 +1,115 @@
+{
+  "sub_name": "danbooru-50k-v1",
+  "num_opinion_anchors": 225,
+  "pentachoron_dim": 256,
+  "scales": [
+    128,
+    256,
+    512
+  ],
+  "scale_hidden_dims": {
+    "128": 256,
+    "256": 512,
+    "512": 1024
+  },
+  "alpha_init": 0.1,
+  "alpha_learnable": true,
+  "alpha_per_scale": true,
+  "beta_init": 0.5,
+  "beta_learnable": true,
+  "beta_per_scale": true,
+  "gamma_learnable": true,
+  "learn_layer_weights": true,
+  "siglip_model": "google/siglip-so400m-patch14-384",
+  "clip_tokenizer": "openai/clip-vit-large-patch14",
+  "illustrious_clip_path": "./models/NAI-11-epsilon_clip_l.safetensors",
+  "clip_skip": 0,
+  "siglip_layer_indices": [
+    12,
+    16,
+    20,
+    23,
+    25,
+    26
+  ],
+  "clip_layer_indices": null,
+  "use_gradient_checkpointing": false,
+  "share_scale_embeddings": true,
+  "dataset_name": "animetimm/danbooru-wdtagger-v4-w640-ws-50k",
+  "image_size": 384,
+  "max_tag_length": 77,
+  "batch_size": 32,
+  "num_epochs": 3,
+  "learning_rate": 0.0001,
+  "weight_decay": 0.01,
+  "warmup_steps": 1000,
+  "gradient_clip": 1.0,
+  "gradient_accumulation_steps": 1,
+  "token_loss_weight": 1.0,
+  "geometric_weight": 0.1,
+  "fusion_strategy": "learned_weighted",
+  "text_dropout_prob": 0.3,
+  "text_noise_std": 0.1,
+  "text_noise_prob": 0.5,
+  "vision_only_text": "general: blank_image",
+  "text_dropout_schedule": "linear",
+  "text_dropout_start": 0.1,
+  "text_dropout_end": 0.5,
+  "checkpoint_dir": "./checkpoints/liminal_staircase_danbooru",
+  "save_every": 500,
+  "hf_repo_id": "AbstractPhil/liminal-staircase-v2",
+  "hf_upload_every": 1000,
+  "hf_private": false,
+  "resume": true,
+  "log_dir": "./logs/liminal_staircase_danbooru",
+  "log_every": 5,
+  "device": "cuda",
+  "timestamp": "2025-11-17T05:12:15.951499",
+  "step": 4585,
+  "epoch": 2,
+  "val_loss": 3.1854735561039136,
+  "fusion_diagnostics": {
+    "layer_weights": [
+      0.05383476987481117,
+      0.05307381600141525,
+      0.05373457819223404,
+      0.05356656759977341,
+      0.05366039276123047,
+      0.05355314910411835,
+      0.05681484192609787,
+      0.05716554448008537,
+      0.05682314559817314,
+      0.056655507534742355,
+      0.05661669373512268,
+      0.05677459016442299,
+      0.056852370500564575,
+      0.05770973861217499,
+      0.05561613664031029,
+      0.05701824277639389,
+      0.05642232671380043,
+      0.05410761386156082
+    ],
+    "scale_weights": [
+      0.3062363862991333,
+      0.3039093315601349,
+      0.3898542523384094
+    ],
+    "alpha_per_scale": [
+      0.27752038836479187,
+      0.27752038836479187,
+      0.27752038836479187
+    ],
+    "beta_per_scale": [
+      0.5901311635971069,
+      0.5916643142700195,
+      0.5984686613082886
+    ],
+    "scale_statistics": {},
+    "text_modality_stats": {
+      "clean": "30.0%",
+      "noisy": "29.7%",
+      "sentinel": "40.3%"
+    }
+  },
+  "is_best": true
+}

danbooru-50k-v1/epoch2_step4585_20251117_051214/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f169684ac83dddb6425e8b09d4f1276cd9d828689667e4dda3dccb182aad2992
+size 328002100

danbooru-50k-v1/epoch2_step4585_20251117_051214/training_state.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4dccc1dbc0d1a59639b4d067bdbf9fb66117a9aa942d18f712011bba2685042f
+size 606530821