checkpoint step 10000

training-runs/latest/args.json

@@ -8,17 +8,17 @@
   "block_size": 512,
   "batch_size": 4,
   "grad_accum": 8,
-  "max_steps": 8000,
+  "max_steps": 20000,
   "lr": 0.0003,
   "min_lr": 3e-05,
-  "warmup_steps": 500,
+  "warmup_steps": 100,
   "weight_decay": 0.1,
   "grad_clip": 1.0,
-  "eval_interval": 500,
+  "dropout": 0.1,
+  "eval_interval": 1000,
   "eval_batches": 20,
-  "save_interval": 500,
+  "save_interval": 1000,
   "log_interval": 50,
-  "dropout": 0.1,
   "pcs_a": 0.8309193524478643,
   "pcs_b": 0.0,
   "out_dir": "runs/latest",

training-runs/latest/checkpoint_best.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:453b80ceb21a4db77c595b2566b36540d5646f3845420f2a0e36c4e9850d5255
-size 614210767
+oid sha256:a386e554610ffaa305376bd54a9b6792e38209b45b613d8e5893a42744a2a6bf
+size 614210639

training-runs/latest/checkpoint_last.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d7f2b99a2ce965291543e2e06d82fd20e30a56a401daff5049ab640c5e0f8ee
-size 614210767
+oid sha256:52865902b8eb091a387eef93bc33ad9f9bf492b284e21bf56a2d024763ae21f3
+size 614210639