Run 2. Outer Step 9. Inner Step 0.

config.json

@@ -3,18 +3,18 @@
   "activation_function": "gelu_new",
   "all_reduce_scores": {
     "0": "NON_PARTICIPATING",
-    "1": "NON_PARTICIPATING",
+    "1": "SUCCESS",
     "10": "NON_PARTICIPATING",
     "100": "NON_PARTICIPATING",
-    "101": "NON_PARTICIPATING",
-    "102": "NON_PARTICIPATING",
+    "101": "SUCCESS",
+    "102": "SUCCESS",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
-    "109": "NON_PARTICIPATING",
+    "109": "SUCCESS",
     "11": "NON_PARTICIPATING",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
@@ -75,12 +75,12 @@
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
-    "164": "NON_PARTICIPATING",
+    "164": "SUCCESS",
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NON_PARTICIPATING",
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
@@ -89,7 +89,7 @@
     "174": "NON_PARTICIPATING",
     "175": "NON_PARTICIPATING",
     "176": "NON_PARTICIPATING",
-    "177": "NON_PARTICIPATING",
+    "177": "SUCCESS",
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "NON_PARTICIPATING",
@@ -107,7 +107,7 @@
     "190": "NON_PARTICIPATING",
     "191": "NON_PARTICIPATING",
     "192": "SUCCESS",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NON_PARTICIPATING",
     "195": "NON_PARTICIPATING",
     "196": "NON_PARTICIPATING",
@@ -115,7 +115,7 @@
     "198": "NON_PARTICIPATING",
     "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
+    "20": "SUCCESS",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NON_PARTICIPATING",
@@ -137,16 +137,16 @@
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
     "219": "NON_PARTICIPATING",
-    "22": "NON_PARTICIPATING",
+    "22": "SUCCESS",
     "220": "NON_PARTICIPATING",
-    "221": "NON_PARTICIPATING",
+    "221": "SUCCESS",
     "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NON_PARTICIPATING",
-    "225": "NON_PARTICIPATING",
+    "225": "SUCCESS",
     "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
-    "228": "SUCCESS",
+    "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
     "23": "SUCCESS",
     "230": "NON_PARTICIPATING",
@@ -184,21 +184,21 @@
     "3": "NON_PARTICIPATING",
     "30": "NON_PARTICIPATING",
     "31": "NON_PARTICIPATING",
-    "32": "NON_PARTICIPATING",
+    "32": "SUCCESS",
     "33": "NON_PARTICIPATING",
     "34": "NON_PARTICIPATING",
-    "35": "NON_PARTICIPATING",
+    "35": "SUCCESS",
     "36": "SUCCESS",
     "37": "NON_PARTICIPATING",
     "38": "NON_PARTICIPATING",
-    "39": "SUCCESS",
+    "39": "NON_PARTICIPATING",
     "4": "NON_PARTICIPATING",
     "40": "NON_PARTICIPATING",
     "41": "NON_PARTICIPATING",
-    "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
+    "42": "SUCCESS",
+    "43": "SUCCESS",
     "44": "NON_PARTICIPATING",
-    "45": "SUCCESS",
+    "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
@@ -238,7 +238,7 @@
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
     "80": "NON_PARTICIPATING",
-    "81": "NON_PARTICIPATING",
+    "81": "SUCCESS",
     "82": "NON_PARTICIPATING",
     "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
@@ -267,21 +267,15 @@
     "AutoConfig": "distributed/optimized-gpt2-500m--configuration_gpt_optimized.GPTOptimConfig",
     "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
   },
-  "block_list": [
-    5338021,
-    5338025,
-    5338029,
-    5338033,
-    5338037
-  ],
+  "block_list": [],
   "block_size": 1024,
   "bos_token_id": 50256,
   "embd_pdrop": 0.1,
   "eos_token_id": 50256,
   "initializer_range": 0.02,
-  "inner_step": 375,
+  "inner_step": 0,
   "inner_steps": 0,
-  "last_allreduce_block": 5332469,
+  "last_allreduce_block": 5351170,
   "layer_norm_epsilon": 1e-05,
   "model_type": "gpt_optimized",
   "n_embd": 1280,

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1b1eaeb78c91b7aa0dfce940caf6bc8b1826416d639b815269454f64b29ea66
+oid sha256:57a7560b7d6dcff2f9b986cdf1519151986f63582928104d82fbbcf373d695a7
 size 8081782026

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3948ef8042932de49b3d5e8ee87fd23f70fb088bda1c43190d1a44c6d2bb8402
+oid sha256:30a3c2a1f583992135596ada96883cba467a485b6ef53f21ba48c2355f414587
 size 4040701744