checkpoint step 4000

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 sha256:43bc2c8bd1b86f43521dc36f3fdb11b458093d6bc32741df3f522a209c2259ae
-size 614210767
+oid sha256:87800cb1c37c34647491d2a080676a4132d1cae0ddecef1f26bf8bf4735947f7
+size 614210639

training-runs/latest/checkpoint_last.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e2ebaeef7fe3a93002db516e655cf146c65bab322a6b0f991a2e6f6b157ce3d
-size 614210767
+oid sha256:4da0ec6afb5d483cb7f2066f551b3debe0ca62f42d3f6502822d3ef87cda1f13
+size 614210639