step10000: step 10000, 640000 examples, eval {'val_loss': 1.4349, 'val_wp_mae': 0.0981, 'top_move_acc': 0.19666666666666666, 'step': 10000}

README.md

@@ -9,7 +9,7 @@ tags: [chess, laya, decision-model]
 for each legal move, a Laya `score` question predicts the side to move's win chance over 10 levels.
 The engine plays the move with the highest expected win chance (no search).
 
-- checkpoint: `step8904` (step 8904, 569,856 training examples)
-- latest eval: `{'val_loss': 1.4704, 'val_wp_mae': 0.1043, 'top_move_acc': 0.21666666666666667, 'step': 8000}`
+- checkpoint: `step10000` (step 10000, 640,000 training examples)
+- latest eval: `{'val_loss': 1.4349, 'val_wp_mae': 0.0981, 'top_move_acc': 0.19666666666666666, 'step': 10000}`
 - encoding / levels / prompt template: see `chess_meta.json`
 - code: github.com/devroopsaha744/LayaChess

chess_meta.json

@@ -1,8 +1,8 @@
 {
- "tag": "step8904",
+ "tag": "step10000",
  "final": false,
- "step": 8904,
- "consumed": 569856,
+ "step": 10000,
+ "consumed": 640000,
  "n_levels": 10,
  "crit": [
   "0-10%",
@@ -45,98 +45,6 @@
   "SEED": 0
  },
  "history": [
-  {
-   "step": 6500,
-   "loss": 1.5242
-  },
-  {
-   "step": 6550,
-   "loss": 1.533
-  },
-  {
-   "step": 6600,
-   "loss": 1.5114
-  },
-  {
-   "step": 6650,
-   "loss": 1.5443
-  },
-  {
-   "step": 6700,
-   "loss": 1.5239
-  },
-  {
-   "step": 6750,
-   "loss": 1.4953
-  },
-  {
-   "step": 6800,
-   "loss": 1.539
-  },
-  {
-   "step": 6850,
-   "loss": 1.5144
-  },
-  {
-   "step": 6900,
-   "loss": 1.5075
-  },
-  {
-   "step": 6950,
-   "loss": 1.5172
-  },
-  {
-   "step": 7000,
-   "loss": 1.5296
-  },
-  {
-   "step": 7050,
-   "loss": 1.4984
-  },
-  {
-   "step": 7100,
-   "loss": 1.5098
-  },
-  {
-   "step": 7150,
-   "loss": 1.5205
-  },
-  {
-   "step": 7200,
-   "loss": 1.5052
-  },
-  {
-   "step": 7250,
-   "loss": 1.5121
-  },
-  {
-   "step": 7300,
-   "loss": 1.5161
-  },
-  {
-   "step": 7350,
-   "loss": 1.5277
-  },
-  {
-   "step": 7400,
-   "loss": 1.5275
-  },
-  {
-   "step": 7450,
-   "loss": 1.4838
-  },
-  {
-   "step": 7500,
-   "loss": 1.4762
-  },
-  {
-   "step": 7550,
-   "loss": 1.4789
-  },
-  {
-   "step": 7600,
-   "loss": 1.4958
-  },
   {
    "step": 7650,
    "loss": 1.5131
@@ -246,6 +154,100 @@
   {
    "step": 8900,
    "loss": 1.4476
+  },
+  {
+   "step": 8950,
+   "loss": 1.4855
+  },
+  {
+   "step": 9000,
+   "loss": 1.4658
+  },
+  {
+   "step": 9050,
+   "loss": 1.4446
+  },
+  {
+   "step": 9100,
+   "loss": 1.463
+  },
+  {
+   "step": 9150,
+   "loss": 1.4703
+  },
+  {
+   "step": 9200,
+   "loss": 1.4479
+  },
+  {
+   "step": 9250,
+   "loss": 1.4321
+  },
+  {
+   "step": 9300,
+   "loss": 1.45
+  },
+  {
+   "step": 9350,
+   "loss": 1.4566
+  },
+  {
+   "step": 9400,
+   "loss": 1.4516
+  },
+  {
+   "step": 9450,
+   "loss": 1.4543
+  },
+  {
+   "step": 9500,
+   "loss": 1.459
+  },
+  {
+   "step": 9550,
+   "loss": 1.4481
+  },
+  {
+   "step": 9600,
+   "loss": 1.4224
+  },
+  {
+   "step": 9650,
+   "loss": 1.4161
+  },
+  {
+   "step": 9700,
+   "loss": 1.4238
+  },
+  {
+   "step": 9750,
+   "loss": 1.4439
+  },
+  {
+   "step": 9800,
+   "loss": 1.4393
+  },
+  {
+   "step": 9850,
+   "loss": 1.4532
+  },
+  {
+   "step": 9900,
+   "loss": 1.4452
+  },
+  {
+   "step": 9950,
+   "loss": 1.4522
+  },
+  {
+   "step": 10000,
+   "loss": 1.4421
+  },
+  {
+   "val_loss": 1.4349,
+   "val_wp_mae": 0.0981,
+   "top_move_acc": 0.19666666666666666,
+   "step": 10000
   }
  ]
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea9d548376ed8146bf5b304895cfae9f8ac824820df2fb6e33593a70b0f53398
+oid sha256:0d91c7ca1dc98b19b4dc1279258d7f55c42cab0b33ec846fb3530ee32ae059ac
 size 842609220