Upload training_logs/hill_subseq_h8_T12_s5__seed42.json with huggingface_hub

training_logs/hill_subseq_h8_T12_s5__seed42.json

@@ -0,0 +1,91 @@
+{
+  "loss": [
+    120143.21875,
+    51956.79296875,
+    48929.91796875,
+    47928.9453125,
+    47554.76953125,
+    46846.76171875,
+    46840.83984375,
+    46335.2109375,
+    45987.1171875,
+    46118.2578125,
+    46137.21875,
+    45948.55859375,
+    45929.99609375,
+    45834.43359375,
+    45736.5625,
+    45599.38671875
+  ],
+  "l0": [
+    500.0,
+    497.642822265625,
+    497.530517578125,
+    496.827392578125,
+    497.216064453125,
+    497.449951171875,
+    497.581298828125,
+    496.6240234375,
+    496.12890625,
+    496.3408203125,
+    495.974853515625,
+    496.058349609375,
+    496.18115234375,
+    495.67724609375,
+    495.98486328125,
+    495.452880859375
+  ],
+  "steps_logged": [
+    0,
+    200,
+    400,
+    600,
+    800,
+    1000,
+    1200,
+    1400,
+    1600,
+    1800,
+    2000,
+    2200,
+    2400,
+    2600,
+    2800,
+    3000
+  ],
+  "final_step": 3000,
+  "converged": true,
+  "plateau_last": 0.010239914057270363,
+  "elapsed_s": 6603.551581859589,
+  "T_max": 12,
+  "t_sample": 5,
+  "shifts": null,
+  "matryoshka_h_size": 3686,
+  "row": 100,
+  "arch_id": "hill_subseq_h8_T12_s5",
+  "arch": "hill_subseq_h8_T12_s5",
+  "group": 99,
+  "src_class": "SubseqH8",
+  "src_module": "src.architectures.phase5b_subseq_sampling_txcdr",
+  "T": null,
+  "n_layers": null,
+  "k_win": 500,
+  "k_pos": 41,
+  "alpha": 1.0,
+  "gamma": null,
+  "n_scales": null,
+  "seed": 42,
+  "d_in": 2304,
+  "d_sae": 18432,
+  "subject_model": "google/gemma-2-2b",
+  "anchor_layer": 12,
+  "mlc_layers": [
+    10,
+    11,
+    12,
+    13,
+    14
+  ],
+  "phase": "phase7_unification",
+  "run_id": "hill_subseq_h8_T12_s5__seed42"
+}