checkpoint step 6500

training-runs/latest/args.json

@@ -8,17 +8,17 @@
   "block_size": 512,
   "batch_size": 4,
   "grad_accum": 8,
-  "max_steps": 20000,
+  "max_steps": 8000,
   "lr": 0.0003,
   "min_lr": 3e-05,
-  "warmup_steps": 100,
+  "warmup_steps": 500,
   "weight_decay": 0.1,
   "grad_clip": 1.0,
-  "dropout": 0.1,
-  "eval_interval": 1000,
+  "eval_interval": 500,
   "eval_batches": 20,
-  "save_interval": 1000,
+  "save_interval": 500,
   "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:ff507fff3e1bdfc9ed4b593352cfeda93eb37d1035dd9daad9e7ac04b695e346
-size 614210639
+oid sha256:d22c0cfeeceeda651217889358a660945039319a13cd3878c2c284630a561b5a
+size 614210767

training-runs/latest/checkpoint_last.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1367b2c471279e15d9abd0ec3469fe121877912bf36e8db7ba5d400100693706
-size 614210639
+oid sha256:e2bae86d8c106c6df2cc8438b0bde8937160fce3bfa6ae0a9cc2a381f931692c
+size 614210767