step 1600 | val_ppl=31.16

README.md

@@ -25,9 +25,9 @@ Small language model (3.0M parameters) trained from scratch.
 | Total parameters | 3.000M |
 
 ## Training
-- Tokens seen: 1,257,062,400
-- Val loss: 3.6765
-- Val PPL: 39.51
+- Tokens seen: 1,676,083,200
+- Val loss: 3.4390
+- Val PPL: 31.16
 
 ## Usage
 ```python

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:17b7ca7a1bf0a45160b8c75808bebaf48847361cfc180a009958d5ba4b17fbb9
+oid sha256:6b754a51f8f46a69c27b4deee3dba32e1d5503b17829868a74e091a378cb1244
 size 12006544

resume/ckpt.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8791a1d26c90560bbd040ee614e276bbe175e4a54220c1428deed4c5799dd958
+oid sha256:4e17f273007a88577c6e746158a82011fe9599f1644f638078649172332b1c5f
 size 36728047

resume/latest_step.txt

@@ -1 +1 @@
-1200
+1600

training_meta.json

@@ -1,8 +1,8 @@
 {
-  "step": 1200,
-  "val_loss": 3.676522359161377,
-  "val_ppl": 39.50875761723507,
+  "step": 1600,
+  "val_loss": 3.438975407333374,
+  "val_ppl": 31.15502060396692,
   "params_M": 3.0,
-  "pushed_at": "2026-07-01T11:45:46.092897",
-  "tokens_seen": 1257062400
+  "pushed_at": "2026-07-01T12:36:19.723327",
+  "tokens_seen": 1676083200
 }