Stage 2 checkpoint: [Step 100/20000] loss=7.5625

README.md

@@ -57,10 +57,10 @@ Key research findings applied:
 
 ## Current Checkpoint
 
-- **Stage**: 1 (layer-alignment)
-- **Step**: 9000
-- **Loss**: 0.0126
-- **Updated**: 2026-03-12T22:48:49.369182+00:00
+- **Stage**: 2 (kl-distillation)
+- **Step**: 15
+- **Loss**: 7.5977
+- **Updated**: 2026-03-12T22:49:07.475098+00:00
 
 ## Languages
 

stage2_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:038e23318f8fc32cb75c0e3a3b376bb947427203c019e26c45b11914cb0e7106
+size 1443376118

stage2_metadata.json

@@ -0,0 +1,14 @@
+{
+  "step": 15,
+  "checkpoint_file": "best.pt",
+  "timestamp": "2026-03-12T22:49:07.475098+00:00",
+  "final_loss": 7.59765625,
+  "best_loss": 7.59765625,
+  "training_config": {
+    "temperature": 2.0,
+    "alpha": 0.7,
+    "seed": 42
+  },
+  "stage": 2,
+  "stage_name": "kl-distillation"
+}