Run 6. Outer Step 96. Inner Step 5.

config.json

@@ -17,18 +17,18 @@
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "SUCCESS",
-    "112": "NON_PARTICIPATING",
+    "112": "SUCCESS",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NON_PARTICIPATING",
-    "119": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "SUCCESS",
-    "122": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
@@ -42,59 +42,59 @@
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
     "134": "NOT_ALIVE",
-    "135": "NOT_ALIVE",
-    "136": "NON_PARTICIPATING",
+    "135": "SUCCESS",
+    "136": "SUCCESS",
     "137": "NON_PARTICIPATING",
-    "138": "SUCCESS",
-    "139": "NON_PARTICIPATING",
+    "138": "NON_PARTICIPATING",
+    "139": "SUCCESS",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
-    "143": "SUCCESS",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
-    "148": "NON_PARTICIPATING",
-    "149": "NOT_ALIVE",
+    "148": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
-    "151": "NON_PARTICIPATING",
+    "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
     "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
-    "159": "NON_PARTICIPATING",
+    "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
     "163": "NON_PARTICIPATING",
-    "164": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
-    "166": "SUCCESS",
+    "166": "NON_PARTICIPATING",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
     "174": "NOT_ALIVE",
-    "175": "NOT_ALIVE",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NON_PARTICIPATING",
-    "179": "NOT_ALIVE",
+    "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
-    "182": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
     "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
@@ -104,11 +104,11 @@
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
-    "191": "NON_PARTICIPATING",
+    "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
+    "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
@@ -130,14 +130,14 @@
     "211": "NON_PARTICIPATING",
     "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
-    "214": "NOT_ALIVE",
-    "215": "NON_PARTICIPATING",
+    "214": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
-    "219": "NON_PARTICIPATING",
+    "219": "SUCCESS",
     "22": "NOT_ALIVE",
-    "220": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
@@ -145,7 +145,7 @@
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
     "227": "NOT_ALIVE",
-    "228": "NON_PARTICIPATING",
+    "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
@@ -170,13 +170,13 @@
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NOT_ALIVE",
+    "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
-    "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
+    "255": "NOT_ALIVE",
+    "26": "NOT_ALIVE",
     "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
@@ -185,20 +185,20 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
+    "40": "SUCCESS",
     "41": "NOT_ALIVE",
     "42": "NON_PARTICIPATING",
     "43": "NON_PARTICIPATING",
     "44": "NOT_ALIVE",
-    "45": "NON_PARTICIPATING",
-    "46": "SUCCESS",
+    "45": "NOT_ALIVE",
+    "46": "NON_PARTICIPATING",
     "47": "NOT_ALIVE",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
@@ -206,7 +206,7 @@
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
     "52": "NOT_ALIVE",
-    "53": "SUCCESS",
+    "53": "NOT_ALIVE",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
@@ -214,26 +214,26 @@
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
     "61": "NOT_ALIVE",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
-    "64": "NOT_ALIVE",
+    "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "SUCCESS",
+    "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
     "74": "NOT_ALIVE",
-    "75": "NON_PARTICIPATING",
+    "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
-    "78": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
@@ -243,7 +243,7 @@
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
-    "87": "NOT_ALIVE",
+    "87": "SUCCESS",
     "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
@@ -251,7 +251,7 @@
     "91": "NOT_ALIVE",
     "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NON_PARTICIPATING",
     "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6197590,
-    6197606
+    6197669,
+    6197681
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 84,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6196966,
+  "last_allreduce_block": 6197602,
   "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:dbb1db14e9557c2f06b7d5250f155a58bd3e6994453ce0d16cc24fcd07cf8a40
+oid sha256:cd85986a87168e20beec669648aa6e6ed531fa213de6f054ccafb1812e977d5a
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e227eb3c4c2921c09c02266d4471f17440915a6676f169502db96aaefaa2bca
+oid sha256:95220c50c22ea2f55a096399a84115422b104aa46408479ff9f3911544e8dcb3
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04dd05181a07bbc1af159faf03f1f468a10d9a9ad9355c72be8de51b72ce1c3b
+oid sha256:567ce6d90c11e16fecc168f699abcac07a2acf501c85926ad19b054d2f229277
 size 4400216536