Run 11. Outer Step 340. Inner Step 9.

config.json

@@ -3,25 +3,25 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NON_PARTICIPATING",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
+    "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
-    "108": "NOT_ALIVE",
+    "107": "SUCCESS",
+    "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "NOT_ALIVE",
+    "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
-    "116": "SUCCESS",
+    "115": "NON_PARTICIPATING",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
@@ -29,26 +29,26 @@
     "120": "NOT_ALIVE",
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "FAIL",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
+    "127": "SUCCESS",
     "128": "SUCCESS",
     "129": "NOT_ALIVE",
-    "13": "NOT_ALIVE",
+    "13": "SUCCESS",
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "134": "NOT_ALIVE",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
-    "14": "NOT_ALIVE",
-    "140": "SUCCESS",
+    "14": "NON_PARTICIPATING",
+    "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
@@ -62,7 +62,7 @@
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -77,19 +77,19 @@
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "NON_PARTICIPATING",
+    "167": "SUCCESS",
     "168": "NOT_ALIVE",
-    "169": "SUCCESS",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "NOT_ALIVE",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "FAIL",
+    "178": "SUCCESS",
     "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -98,7 +98,7 @@
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -115,7 +115,7 @@
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
@@ -132,35 +132,35 @@
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
+    "216": "NOT_ALIVE",
     "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "NOT_ALIVE",
+    "221": "SUCCESS",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "SUCCESS",
+    "226": "NON_PARTICIPATING",
     "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "SUCCESS",
-    "231": "SUCCESS",
+    "230": "NOT_ALIVE",
+    "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "SUCCESS",
-    "238": "NOT_ALIVE",
+    "237": "NOT_ALIVE",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "NOT_ALIVE",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -172,23 +172,23 @@
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "FAIL",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NON_PARTICIPATING",
+    "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "NOT_ALIVE",
+    "37": "SUCCESS",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
@@ -210,7 +210,7 @@
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
@@ -219,7 +219,7 @@
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
@@ -230,12 +230,12 @@
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NON_PARTICIPATING",
+    "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
-    "8": "SUCCESS",
+    "79": "SUCCESS",
+    "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
@@ -243,13 +243,13 @@
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
-    "87": "SUCCESS",
+    "87": "NON_PARTICIPATING",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "SUCCESS",
-    "92": "SUCCESS",
+    "91": "NOT_ALIVE",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6718116,
-    6718146
+    6718311,
+    6718341
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6717092,
+  "last_allreduce_block": 6718281,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

gradients.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:081b011e61b21878471a7f080e974cc04b82e7c985cef95d0cc027c00172d031
+oid sha256:7c92efafe9485db10858ad1e1ee9f373ff0accc58f8e88f6585033cfbd762d17
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5ec67549714cc756940bb84a14d91911924dabcbb24e9c29c8acce2269c8c79c
+oid sha256:df12d427e6caebbab7d4dcbd36490bc057570f1e899470b1fa8348f0c7469dcf
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7df0f9840cd5fe99a085a638c9d81ca8394924dafc72f5c0840c15b8c5de264c
+oid sha256:d114c2c155c6d77a574e7bd3f44b7524e61302afce0073e327d6c87620f84e24
 size 4400216536