checkpoint step 7000

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:fffad76797f136f45e2ad3e23c0067298c4771871b4c79c8355ba55f377958f5
-size 614210767
+oid sha256:ff507fff3e1bdfc9ed4b593352cfeda93eb37d1035dd9daad9e7ac04b695e346
+size 614210639

training-runs/latest/checkpoint_last.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ce1a2bde5cef7d90fc38ec67f79beb894de4b76abc8c34d886f716821f36b0c
-size 614210767
+oid sha256:640e207d04dc004f5df23dd9d50cf5d579964364e61b38a0dcff2b98ca3adb88
+size 614210639