Checkpoint: 20251117_015502_danbooru-50k-v1_step6

checkpoints/20251117_015502_danbooru-50k-v1_step6/config.json

@@ -0,0 +1,102 @@
+{
+  "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",
+  "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": 50,
+  "device": "cuda",
+  "timestamp": "2025-11-17T01:55:03.560895",
+  "step": 6,
+  "epoch": 0,
+  "val_loss": Infinity,
+  "fusion_diagnostics": {
+    "layer_weights": [
+      0.05555903539061546,
+      0.05554059520363808,
+      0.05553547292947769,
+      0.05557769536972046,
+      0.05554599314928055,
+      0.05556809529662132,
+      0.05554487556219101,
+      0.05558248609304428,
+      0.05553889647126198,
+      0.055535610765218735,
+      0.05558742955327034,
+      0.05557844415307045,
+      0.05556866526603699,
+      0.05554370582103729,
+      0.05553441122174263,
+      0.05558684095740318,
+      0.05553947389125824,
+      0.055532246828079224
+    ],
+    "scale_weights": [
+      0.33324742317199707,
+      0.33328598737716675,
+      0.33346661925315857
+    ],
+    "alpha_per_scale": [
+      0.2624704837799072,
+      0.2624704837799072,
+      0.2624704837799072
+    ],
+    "beta_per_scale": [
+      0.6224586963653564,
+      0.6223307847976685,
+      0.6223313808441162
+    ],
+    "scale_statistics": {}
+  }
+}

checkpoints/20251117_015502_danbooru-50k-v1_step6/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c49006c8008c0cb256569378155a99939a20ea1fb5f9af6b99bd433b1977e538
+size 328002100

checkpoints/20251117_015502_danbooru-50k-v1_step6/training_state.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3206476e8080fe5beb65ced45a926c5630286b0edc5fd1e150bb9c010608d845
+size 606530757