Training in progress, step 38500, checkpoint

last-checkpoint/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f14125fba192c696e9ea3e36ca9f9429c24cac19e9b8c350f6135ab01f7ad59
+oid sha256:78bb7fd938fc79d04fd8027a0abe1e34028a04e2b63e0cc950fef379ab50d87e
 size 132187888

last-checkpoint/ar_diffusion_info.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea672d9cf3b33b67a7d7775ed77487a43c2ed825e9c82fdac570ac86bb7b82e0
+oid sha256:9d987e9657df7f151748ac91d327188ea5727ed2126f4ab712064c12bbf47258
 size 1800

last-checkpoint/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8bcbfe92a4edd63a8b647a0117b3f07614d7a6540fc40ddbb287c8b91dd837c1
+oid sha256:f1fa81680adbb395cf42af1d0162aa26889a89b2e0a8b194c7f5b489f789ba80
 size 264665786

last-checkpoint/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc11b8250e89187182c146356c8dd099d66b6a75fadfa56262487e142dbe4b90
+oid sha256:ddf4785afde6da18af71573e1d2e7789bcefbfcd7bcf8163ade4a7bd4cb547c8
 size 14244

last-checkpoint/scaler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:102e993b93214454fb03b9ca20f482147915e5a724733ab537e52178df0f84f1
+oid sha256:a79402dde3815fb10cc451e7d2fb9bb473a43858a3148dbc3bddcddce2fe949e
 size 988

last-checkpoint/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:541cb6dea9c80e2359e5134078f161ec83a5de3af8e29b3e9b6b880e5dd9058f
+oid sha256:f3dea87c40748698e240c93bc3eea5482b3f627ab3a1ed391f6cbe2fe5d7da89
 size 1064

last-checkpoint/trainer_state.json

@@ -1,10 +1,10 @@
 {
-  "best_global_step": 37750,
-  "best_metric": 0.5946066975593567,
-  "best_model_checkpoint": "./qwen3-4b-ar-diffusion-checkpoints/checkpoint-37500",
-  "epoch": 2.9228520883008997,
+  "best_global_step": 38500,
+  "best_metric": 0.5919152498245239,
+  "best_model_checkpoint": "./qwen3-4b-ar-diffusion-checkpoints/checkpoint-38500",
+  "epoch": 2.961310668410122,
   "eval_steps": 250,
-  "global_step": 38000,
+  "global_step": 38500,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -6544,6 +6544,92 @@
       "eval_samples_per_second": 22.573,
       "eval_steps_per_second": 5.643,
       "step": 38000
+    },
+    {
+      "epoch": 2.926697946311822,
+      "grad_norm": 0.9351138472557068,
+      "learning_rate": 5.033374022803418e-06,
+      "loss": 0.6027,
+      "step": 38050
+    },
+    {
+      "epoch": 2.9305438043227445,
+      "grad_norm": 0.425923615694046,
+      "learning_rate": 4.773654000986936e-06,
+      "loss": 0.578,
+      "step": 38100
+    },
+    {
+      "epoch": 2.9343896623336665,
+      "grad_norm": 0.728718638420105,
+      "learning_rate": 4.513933979170454e-06,
+      "loss": 0.6103,
+      "step": 38150
+    },
+    {
+      "epoch": 2.938235520344589,
+      "grad_norm": 0.9447925090789795,
+      "learning_rate": 4.254213957353972e-06,
+      "loss": 0.6281,
+      "step": 38200
+    },
+    {
+      "epoch": 2.9420813783555113,
+      "grad_norm": 1.3824400901794434,
+      "learning_rate": 3.994493935537491e-06,
+      "loss": 0.5825,
+      "step": 38250
+    },
+    {
+      "epoch": 2.9420813783555113,
+      "eval_loss": 0.5938913226127625,
+      "eval_runtime": 21.2182,
+      "eval_samples_per_second": 23.565,
+      "eval_steps_per_second": 5.891,
+      "step": 38250
+    },
+    {
+      "epoch": 2.9459272363664333,
+      "grad_norm": 1.073121190071106,
+      "learning_rate": 3.734773913721009e-06,
+      "loss": 0.6468,
+      "step": 38300
+    },
+    {
+      "epoch": 2.9497730943773557,
+      "grad_norm": 1.4922077655792236,
+      "learning_rate": 3.475053891904527e-06,
+      "loss": 0.6511,
+      "step": 38350
+    },
+    {
+      "epoch": 2.953618952388278,
+      "grad_norm": 0.7231102585792542,
+      "learning_rate": 3.215333870088045e-06,
+      "loss": 0.5952,
+      "step": 38400
+    },
+    {
+      "epoch": 2.9574648103992,
+      "grad_norm": 1.1157270669937134,
+      "learning_rate": 2.9556138482715634e-06,
+      "loss": 0.5715,
+      "step": 38450
+    },
+    {
+      "epoch": 2.961310668410122,
+      "grad_norm": 0.7627404928207397,
+      "learning_rate": 2.6958938264550813e-06,
+      "loss": 0.6085,
+      "step": 38500
+    },
+    {
+      "epoch": 2.961310668410122,
+      "eval_loss": 0.5919152498245239,
+      "eval_runtime": 22.1697,
+      "eval_samples_per_second": 22.553,
+      "eval_steps_per_second": 5.638,
+      "step": 38500
     }
   ],
   "logging_steps": 50,