Run 11. Outer Step 60. Inner Step 9.

config.json

@@ -8,19 +8,19 @@
     "101": "NOT_ALIVE",
     "102": "SUCCESS",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
+    "104": "SUCCESS",
     "105": "SUCCESS",
-    "106": "SUCCESS",
+    "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
-    "109": "SUCCESS",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
-    "111": "SUCCESS",
+    "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
+    "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
     "117": "SUCCESS",
     "118": "SUCCESS",
@@ -30,13 +30,13 @@
     "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
-    "124": "NOT_ALIVE",
-    "125": "NON_PARTICIPATING",
+    "124": "SUCCESS",
+    "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
-    "13": "NOT_ALIVE",
+    "13": "NON_PARTICIPATING",
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
@@ -46,14 +46,14 @@
     "136": "NOT_ALIVE",
     "137": "NON_PARTICIPATING",
     "138": "NOT_ALIVE",
-    "139": "NOT_ALIVE",
-    "14": "NON_PARTICIPATING",
+    "139": "SUCCESS",
+    "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "SUCCESS",
     "143": "SUCCESS",
     "144": "NOT_ALIVE",
-    "145": "NON_PARTICIPATING",
+    "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "NOT_ALIVE",
     "148": "NOT_ALIVE",
@@ -61,47 +61,47 @@
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
     "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "SUCCESS",
-    "159": "NON_PARTICIPATING",
-    "16": "NON_PARTICIPATING",
+    "159": "SUCCESS",
+    "16": "SUCCESS",
     "160": "NON_PARTICIPATING",
-    "161": "NON_PARTICIPATING",
-    "162": "NON_PARTICIPATING",
-    "163": "NON_PARTICIPATING",
+    "161": "NOT_ALIVE",
+    "162": "NOT_ALIVE",
+    "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "SUCCESS",
     "167": "NOT_ALIVE",
     "168": "SUCCESS",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "SUCCESS",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
-    "18": "NON_PARTICIPATING",
+    "18": "SUCCESS",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
-    "185": "NOT_ALIVE",
-    "186": "NON_PARTICIPATING",
+    "185": "NON_PARTICIPATING",
+    "186": "SUCCESS",
     "187": "NOT_ALIVE",
-    "188": "NON_PARTICIPATING",
-    "189": "NOT_ALIVE",
+    "188": "NOT_ALIVE",
+    "189": "NON_PARTICIPATING",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
@@ -110,8 +110,8 @@
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
-    "197": "SUCCESS",
-    "198": "NOT_ALIVE",
+    "197": "NON_PARTICIPATING",
+    "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NON_PARTICIPATING",
@@ -120,15 +120,15 @@
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "205": "NON_PARTICIPATING",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "NOT_ALIVE",
     "21": "SUCCESS",
-    "210": "NON_PARTICIPATING",
+    "210": "SUCCESS",
     "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
+    "212": "SUCCESS",
     "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
@@ -138,15 +138,15 @@
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
-    "221": "NON_PARTICIPATING",
+    "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
-    "223": "NOT_ALIVE",
+    "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
-    "226": "SUCCESS",
+    "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
@@ -156,7 +156,7 @@
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "SUCCESS",
-    "238": "NON_PARTICIPATING",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
@@ -167,11 +167,11 @@
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
-    "248": "SUCCESS",
+    "248": "NON_PARTICIPATING",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
+    "250": "SUCCESS",
+    "251": "NON_PARTICIPATING",
     "252": "SUCCESS",
     "253": "NON_PARTICIPATING",
     "254": "NOT_ALIVE",
@@ -181,7 +181,7 @@
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
-    "30": "SUCCESS",
+    "30": "NON_PARTICIPATING",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
@@ -189,8 +189,8 @@
     "35": "SUCCESS",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
-    "38": "NON_PARTICIPATING",
-    "39": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
@@ -200,8 +200,8 @@
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "SUCCESS",
-    "49": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
+    "49": "SUCCESS",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
@@ -216,30 +216,30 @@
     "6": "NOT_ALIVE",
     "60": "NOT_ALIVE",
     "61": "SUCCESS",
-    "62": "SUCCESS",
+    "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
     "64": "SUCCESS",
     "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "NON_PARTICIPATING",
+    "69": "SUCCESS",
     "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
     "74": "NON_PARTICIPATING",
-    "75": "SUCCESS",
-    "76": "SUCCESS",
+    "75": "NOT_ALIVE",
+    "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "SUCCESS",
-    "8": "NOT_ALIVE",
+    "79": "NON_PARTICIPATING",
+    "8": "NON_PARTICIPATING",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
@@ -247,10 +247,10 @@
     "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
-    "90": "SUCCESS",
+    "90": "NON_PARTICIPATING",
     "91": "SUCCESS",
     "92": "SUCCESS",
-    "93": "NON_PARTICIPATING",
+    "93": "SUCCESS",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6575795,
-    6575830
+    6575911,
+    6575939
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 54,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6575401,
+  "last_allreduce_block": 6575861,
   "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 sha256:7302ea224798577dcf9373c8585f6aa584bf2612e1a970f7404888f9e26dc24c
+oid sha256:449af937a4f97804f1f9bcfa94f7ed88f4ffe93e29a28fc76d737e4876f41a80
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a40a32aea9da0b63766cf4ccb24fc77aa2c6afb38cded41f141e9dfe10abb474
+oid sha256:e28c46e92028fb4adeff7737ba765d82088f93e3f655f3aba61e773e21ce1880
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:146aa44fc6ce70a959c7325d7567ddba7ee89835d7e4f15f8fcddabda3d5d18a
+oid sha256:ce4b2343d9dfe3f6f67aa31e2da210a28d02aa03d4dafd3b289e495b80ca1bf1
 size 4400216536