Training in progress, step 36500, checkpoint

last-checkpoint/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcfacbeb4f4eddb175b2e0f9ee4f9f5fcd3804eb9c67bebaf63f8ba868cd0c6f
+oid sha256:651010c952c4f2e5438cf065b7526e8b7ed7591516d3bc5f9700bebff456e394
 size 132187888

last-checkpoint/ar_diffusion_info.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb54fb35a407195fa0393ccddae444ffa53bbd93544621c32110b554f176ae82
+oid sha256:ee0e0adf2d9b4aaacdcafc95dc9f871bf71cf1d2a2f1053d39ef57319d6a1bfe
 size 1800

last-checkpoint/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d19f177598244598d1f9fc2b6f1af6c8f2b072a1487abb8927d38b4c9181ebf2
+oid sha256:444bcb69ebd6dade1e1ffb612d0702662876875378acdb30d35fc8fc0385d21c
 size 264665786

last-checkpoint/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f04bd3a9c50ae82a14c7b2fa792411be370b9b5b60aeb5bfd024183874a6a1d
+oid sha256:e8d399a9966448d8b21883a1ed6c9a0ed112cf6f2c15eec1d90ff9a1b08984a7
 size 14244

last-checkpoint/scaler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23a098e227a7f13ed947fa28dc0329c514f46192f6bb6910a1d5556540e69df6
+oid sha256:727a5807be63831a649cf3a788d2dbe9fdce0ad9219d92aa0ebe11b8f8712f61
 size 988

last-checkpoint/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:906c43e30b9edc8f130b3f77317e01f206c6e5f1267c22ca4899b9c5968ba8ca
+oid sha256:0cf91a495853427b5f841ab1be3c35766ad0edbc12ff06e2c15160a47811ed22
 size 1064

last-checkpoint/trainer_state.json

@@ -1,10 +1,10 @@
 {
-  "best_global_step": 31000,
-  "best_metric": 0.6043956279754639,
-  "best_model_checkpoint": "./qwen3-4b-ar-diffusion-checkpoints/checkpoint-31000",
-  "epoch": 2.7690177678640104,
+  "best_global_step": 36500,
+  "best_metric": 0.6020208597183228,
+  "best_model_checkpoint": "./qwen3-4b-ar-diffusion-checkpoints/checkpoint-36500",
+  "epoch": 2.8074763479732328,
   "eval_steps": 250,
-  "global_step": 36000,
+  "global_step": 36500,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -6200,6 +6200,92 @@
       "eval_samples_per_second": 22.639,
       "eval_steps_per_second": 5.66,
       "step": 36000
+    },
+    {
+      "epoch": 2.772863625874933,
+      "grad_norm": 0.6253750920295715,
+      "learning_rate": 1.5422174895462695e-05,
+      "loss": 0.5864,
+      "step": 36050
+    },
+    {
+      "epoch": 2.776709483885855,
+      "grad_norm": 0.9502193927764893,
+      "learning_rate": 1.516245487364621e-05,
+      "loss": 0.5955,
+      "step": 36100
+    },
+    {
+      "epoch": 2.780555341896777,
+      "grad_norm": 1.109303593635559,
+      "learning_rate": 1.4902734851829727e-05,
+      "loss": 0.6285,
+      "step": 36150
+    },
+    {
+      "epoch": 2.784401199907699,
+      "grad_norm": 0.8590667843818665,
+      "learning_rate": 1.4643014830013247e-05,
+      "loss": 0.6036,
+      "step": 36200
+    },
+    {
+      "epoch": 2.7882470579186216,
+      "grad_norm": 1.0944328308105469,
+      "learning_rate": 1.4383294808196765e-05,
+      "loss": 0.581,
+      "step": 36250
+    },
+    {
+      "epoch": 2.7882470579186216,
+      "eval_loss": 0.6036717295646667,
+      "eval_runtime": 21.3396,
+      "eval_samples_per_second": 23.431,
+      "eval_steps_per_second": 5.858,
+      "step": 36250
+    },
+    {
+      "epoch": 2.792092915929544,
+      "grad_norm": 0.8459743857383728,
+      "learning_rate": 1.4123574786380284e-05,
+      "loss": 0.5993,
+      "step": 36300
+    },
+    {
+      "epoch": 2.795938773940466,
+      "grad_norm": 1.1745880842208862,
+      "learning_rate": 1.3863854764563799e-05,
+      "loss": 0.6209,
+      "step": 36350
+    },
+    {
+      "epoch": 2.7997846319513884,
+      "grad_norm": 0.8465165495872498,
+      "learning_rate": 1.3604134742747318e-05,
+      "loss": 0.6049,
+      "step": 36400
+    },
+    {
+      "epoch": 2.803630489962311,
+      "grad_norm": 1.0009855031967163,
+      "learning_rate": 1.3344414720930836e-05,
+      "loss": 0.6104,
+      "step": 36450
+    },
+    {
+      "epoch": 2.8074763479732328,
+      "grad_norm": 0.963095486164093,
+      "learning_rate": 1.3084694699114356e-05,
+      "loss": 0.623,
+      "step": 36500
+    },
+    {
+      "epoch": 2.8074763479732328,
+      "eval_loss": 0.6020208597183228,
+      "eval_runtime": 22.1394,
+      "eval_samples_per_second": 22.584,
+      "eval_steps_per_second": 5.646,
+      "step": 36500
     }
   ],
   "logging_steps": 50,