Run 8. Outer Step 48. Inner Step 5.

config.json

@@ -5,13 +5,13 @@
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
     "100": "NON_PARTICIPATING",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "NON_PARTICIPATING",
     "103": "NOT_ALIVE",
     "104": "NON_PARTICIPATING",
     "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
-    "107": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
     "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
@@ -32,27 +32,27 @@
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
-    "126": "NON_PARTICIPATING",
-    "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
+    "126": "FAIL",
+    "127": "NON_PARTICIPATING",
+    "128": "SUCCESS",
+    "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
     "130": "NON_PARTICIPATING",
-    "131": "NON_PARTICIPATING",
+    "131": "NOT_ALIVE",
     "132": "SUCCESS",
-    "133": "FAIL",
+    "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "SUCCESS",
     "136": "SUCCESS",
-    "137": "NON_PARTICIPATING",
-    "138": "NOT_ALIVE",
+    "137": "FAIL",
+    "138": "NON_PARTICIPATING",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
-    "143": "NON_PARTICIPATING",
-    "144": "NOT_ALIVE",
+    "143": "NOT_ALIVE",
+    "144": "NON_PARTICIPATING",
     "145": "SUCCESS",
     "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
@@ -73,30 +73,30 @@
     "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
-    "163": "FAIL",
-    "164": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
+    "164": "FAIL",
     "165": "SUCCESS",
     "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
-    "168": "NON_PARTICIPATING",
+    "168": "NOT_ALIVE",
     "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "SUCCESS",
-    "171": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
     "173": "NOT_ALIVE",
     "174": "SUCCESS",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "SUCCESS",
     "178": "NON_PARTICIPATING",
     "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "FAIL",
     "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
-    "184": "FAIL",
+    "183": "NOT_ALIVE",
+    "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
     "186": "FAIL",
     "187": "NOT_ALIVE",
@@ -104,42 +104,42 @@
     "189": "SUCCESS",
     "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
-    "191": "FAIL",
-    "192": "FAIL",
+    "191": "SUCCESS",
+    "192": "NON_PARTICIPATING",
     "193": "NOT_ALIVE",
     "194": "NON_PARTICIPATING",
     "195": "NON_PARTICIPATING",
     "196": "SUCCESS",
     "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
+    "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
     "202": "FAIL",
-    "203": "FAIL",
+    "203": "NOT_ALIVE",
     "204": "NON_PARTICIPATING",
-    "205": "FAIL",
-    "206": "SUCCESS",
+    "205": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
+    "209": "NOT_ALIVE",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "NON_PARTICIPATING",
+    "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
     "215": "FAIL",
-    "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "216": "SUCCESS",
+    "217": "SUCCESS",
     "218": "FAIL",
     "219": "FAIL",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "FAIL",
     "221": "SUCCESS",
-    "222": "NON_PARTICIPATING",
+    "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "SUCCESS",
@@ -148,8 +148,8 @@
     "228": "SUCCESS",
     "229": "NON_PARTICIPATING",
     "23": "NOT_ALIVE",
-    "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "SUCCESS",
@@ -160,18 +160,18 @@
     "239": "NON_PARTICIPATING",
     "24": "SUCCESS",
     "240": "NON_PARTICIPATING",
-    "241": "NOT_ALIVE",
-    "242": "NOT_ALIVE",
+    "241": "NON_PARTICIPATING",
+    "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NON_PARTICIPATING",
-    "249": "FAIL",
+    "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "NOT_ALIVE",
+    "251": "NON_PARTICIPATING",
     "252": "SUCCESS",
     "253": "SUCCESS",
     "254": "NOT_ALIVE",
@@ -183,24 +183,24 @@
     "3": "FAIL",
     "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
-    "32": "NOT_ALIVE",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NON_PARTICIPATING",
     "36": "FAIL",
     "37": "NOT_ALIVE",
     "38": "SUCCESS",
-    "39": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
     "41": "FAIL",
     "42": "NOT_ALIVE",
     "43": "NON_PARTICIPATING",
-    "44": "NOT_ALIVE",
+    "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -221,15 +221,15 @@
     "64": "NOT_ALIVE",
     "65": "NOT_ALIVE",
     "66": "NON_PARTICIPATING",
-    "67": "NOT_ALIVE",
-    "68": "NON_PARTICIPATING",
-    "69": "NOT_ALIVE",
+    "67": "NON_PARTICIPATING",
+    "68": "NOT_ALIVE",
+    "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
-    "74": "NON_PARTICIPATING",
+    "74": "SUCCESS",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
@@ -237,25 +237,25 @@
     "79": "SUCCESS",
     "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
-    "81": "NON_PARTICIPATING",
-    "82": "NOT_ALIVE",
+    "81": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
-    "84": "FAIL",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "FAIL",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
-    "90": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
     "97": "FAIL",
-    "98": "FAIL",
+    "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,19 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6372642,
-    6372657,
-    6372673,
-    6372692
+    6372812,
+    6372825
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 52,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6372098,
+  "last_allreduce_block": 6372758,
   "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 sha256:00efdcd585535c98829040e883d76bc3753100114956273c0cc75fb3b6dfd2b9
+oid sha256:f3641367b4d838c071fa215df3c95786fe86d833c3971db9060d271b251f47bc
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99451ae9113249a8e8da0e3aa2de656a699935a66ebfb6dc0559d049148da6f0
+oid sha256:1f21a6772106f14c611e39f244c7d844ad19e1493a5243ea1e14dd1f70d0a88d
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6914727796f9c3dcac8c7e0c72f4a99309c51f16f9a8eb61e03c19cbce10e082
+oid sha256:f545ab8a10611e647eb5b70f8bf875890cf0409e8ecf635318e571757272ce04
 size 4400216536