Run 11. Outer Step 317. Inner Step 9.

config.json

@@ -3,46 +3,46 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NOT_ALIVE",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
+    "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
-    "108": "NOT_ALIVE",
-    "109": "SUCCESS",
+    "107": "SUCCESS",
+    "108": "SUCCESS",
+    "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
+    "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
-    "116": "NON_PARTICIPATING",
+    "115": "SUCCESS",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "123": "NON_PARTICIPATING",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
     "128": "NON_PARTICIPATING",
-    "129": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
     "13": "SUCCESS",
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
-    "135": "SUCCESS",
+    "134": "NOT_ALIVE",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -58,28 +58,28 @@
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "NON_PARTICIPATING",
+    "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NOT_ALIVE",
+    "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "SUCCESS",
     "168": "NOT_ALIVE",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
@@ -94,12 +94,12 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
-    "184": "NON_PARTICIPATING",
+    "184": "SUCCESS",
     "185": "NOT_ALIVE",
-    "186": "NOT_ALIVE",
-    "187": "NON_PARTICIPATING",
+    "186": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -121,7 +121,7 @@
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
@@ -130,12 +130,12 @@
     "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NOT_ALIVE",
-    "215": "NOT_ALIVE",
-    "216": "NON_PARTICIPATING",
-    "217": "SUCCESS",
+    "214": "NON_PARTICIPATING",
+    "215": "NON_PARTICIPATING",
+    "216": "NOT_ALIVE",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
-    "219": "NON_PARTICIPATING",
+    "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
     "221": "SUCCESS",
@@ -156,11 +156,11 @@
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
     "237": "SUCCESS",
-    "238": "SUCCESS",
+    "238": "NON_PARTICIPATING",
     "239": "NOT_ALIVE",
-    "24": "NON_PARTICIPATING",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -172,11 +172,11 @@
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "SUCCESS",
+    "252": "FAIL",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -204,29 +204,29 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "NON_PARTICIPATING",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "SUCCESS",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "SUCCESS",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "NOT_ALIVE",
+    "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "FAIL",
+    "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -235,8 +235,8 @@
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
     "79": "SUCCESS",
-    "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
+    "8": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -245,18 +245,18 @@
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
-    "92": "NOT_ALIVE",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "NOT_ALIVE"
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6708053,
-    6708080
+    6708161,
+    6708190
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -276,7 +276,7 @@
   "initializer_range": 0.02,
   "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6707707,
+  "last_allreduce_block": 6708109,
   "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:a2bb62bfafaec2c257a2414920c04e23edf7feb29995f40bb6a92a579d2c6f53
+oid sha256:5cfbd47d5dacdf340519769810aea24bb59d45914a28600aff5a68e97a8ac293
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9f921e15daa2a60f7ab2717e61ecfa0987ae1fce4ecc20cb841f43c9ec5fcd8
+oid sha256:53fea8e64c944024784ad6cc37355dfca856cbda6c1cb5148e88155e413a7f82
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d66250858545a4da830dee4bbf1a868abceb6334928b437529302e4a864c1ac9
+oid sha256:8a7d65e8902532fca29063ef98bc6bbfa67f014eae44872d7f058894fc0ed1b9
 size 4400216536