Run 4. Outer Step 15. Inner Step 0.

config.json

@@ -12,7 +12,7 @@
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
-    "107": "SUCCESS",
+    "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
     "11": "NON_PARTICIPATING",
@@ -20,17 +20,17 @@
     "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
-    "114": "NON_PARTICIPATING",
+    "114": "SUCCESS",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "SUCCESS",
+    "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
     "12": "SUCCESS",
-    "120": "SUCCESS",
+    "120": "FAIL",
     "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
-    "123": "SUCCESS",
+    "123": "FAIL",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
@@ -51,13 +51,13 @@
     "14": "NON_PARTICIPATING",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
+    "142": "SUCCESS",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
-    "146": "NON_PARTICIPATING",
+    "146": "SUCCESS",
     "147": "SUCCESS",
-    "148": "NON_PARTICIPATING",
+    "148": "SUCCESS",
     "149": "SUCCESS",
     "15": "SUCCESS",
     "150": "SUCCESS",
@@ -69,23 +69,23 @@
     "156": "SUCCESS",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
-    "159": "NON_PARTICIPATING",
+    "159": "FAIL",
     "16": "SUCCESS",
-    "160": "SUCCESS",
+    "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
-    "167": "SUCCESS",
-    "168": "SUCCESS",
+    "167": "FAIL",
+    "168": "NON_PARTICIPATING",
     "169": "NON_PARTICIPATING",
     "17": "SUCCESS",
     "170": "SUCCESS",
     "171": "NON_PARTICIPATING",
     "172": "SUCCESS",
-    "173": "NON_PARTICIPATING",
+    "173": "SUCCESS",
     "174": "NON_PARTICIPATING",
     "175": "NON_PARTICIPATING",
     "176": "NON_PARTICIPATING",
@@ -105,11 +105,11 @@
     "189": "NON_PARTICIPATING",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
-    "191": "FAIL",
+    "191": "NON_PARTICIPATING",
     "192": "SUCCESS",
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
-    "195": "SUCCESS",
+    "195": "NON_PARTICIPATING",
     "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
     "198": "NON_PARTICIPATING",
@@ -150,9 +150,9 @@
     "229": "NON_PARTICIPATING",
     "23": "SUCCESS",
     "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
+    "231": "SUCCESS",
     "232": "NON_PARTICIPATING",
-    "233": "NON_PARTICIPATING",
+    "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
@@ -186,7 +186,7 @@
     "31": "NON_PARTICIPATING",
     "32": "SUCCESS",
     "33": "NON_PARTICIPATING",
-    "34": "SUCCESS",
+    "34": "NON_PARTICIPATING",
     "35": "NON_PARTICIPATING",
     "36": "NON_PARTICIPATING",
     "37": "SUCCESS",
@@ -210,7 +210,7 @@
     "53": "NON_PARTICIPATING",
     "54": "NON_PARTICIPATING",
     "55": "SUCCESS",
-    "56": "NON_PARTICIPATING",
+    "56": "SUCCESS",
     "57": "NON_PARTICIPATING",
     "58": "SUCCESS",
     "59": "NON_PARTICIPATING",
@@ -218,8 +218,8 @@
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "SUCCESS",
-    "63": "SUCCESS",
-    "64": "SUCCESS",
+    "63": "FAIL",
+    "64": "FAIL",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "SUCCESS",
@@ -268,20 +268,20 @@
     "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
   },
   "block_list": [
-    5513238,
-    5513242,
-    5513245,
-    5513249,
-    5513254
+    5513295,
+    5513382,
+    5513386,
+    5513389,
+    5513394
   ],
   "block_size": 1024,
   "bos_token_id": 50256,
   "embd_pdrop": 0.1,
   "eos_token_id": 50256,
   "initializer_range": 0.02,
-  "inner_step": 61,
+  "inner_step": 0,
   "inner_steps": 0,
-  "last_allreduce_block": 5511564,
+  "last_allreduce_block": 5513324,
   "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:fbe6369a737d19d958d61902070ea6a665ecc0ed784fd0c9483f31ce7ffef6a0
+oid sha256:2858fb00dd689471659443cf636a883184746fdb5e83cf1fc0888e4430f0d62b
 size 8081782503

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:786593c67c8a73f9d28d7c2f3e508116a02c91fc9d20a9cc30f6d1b98ef09915
+oid sha256:1f5216e63ea5fe48c8089322986ba91ccc7301143167773cbc707dabdd9f0914
 size 4040701744