Run 11. Outer Step 52. Inner Step 9.

config.json

@@ -22,27 +22,27 @@
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
     "116": "NOT_ALIVE",
-    "117": "SUCCESS",
-    "118": "NOT_ALIVE",
+    "117": "NOT_ALIVE",
+    "118": "SUCCESS",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
     "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
-    "124": "NOT_ALIVE",
-    "125": "SUCCESS",
-    "126": "SUCCESS",
+    "124": "SUCCESS",
+    "125": "NOT_ALIVE",
+    "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
-    "13": "NOT_ALIVE",
-    "130": "SUCCESS",
+    "13": "NON_PARTICIPATING",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
-    "135": "NOT_ALIVE",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -57,17 +57,17 @@
     "146": "NOT_ALIVE",
     "147": "NOT_ALIVE",
     "148": "NOT_ALIVE",
-    "149": "SUCCESS",
+    "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
-    "158": "NOT_ALIVE",
+    "157": "NON_PARTICIPATING",
+    "158": "SUCCESS",
     "159": "SUCCESS",
     "16": "SUCCESS",
     "160": "NOT_ALIVE",
@@ -76,16 +76,16 @@
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "SUCCESS",
+    "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "SUCCESS",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
-    "174": "NOT_ALIVE",
+    "174": "NON_PARTICIPATING",
     "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
@@ -98,29 +98,29 @@
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
     "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
-    "190": "NOT_ALIVE",
+    "190": "SUCCESS",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
-    "197": "SUCCESS",
+    "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
-    "20": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "NOT_ALIVE",
+    "205": "NON_PARTICIPATING",
     "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
@@ -150,9 +150,9 @@
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
-    "232": "NOT_ALIVE",
+    "232": "NON_PARTICIPATING",
     "233": "NOT_ALIVE",
-    "234": "SUCCESS",
+    "234": "NOT_ALIVE",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "SUCCESS",
@@ -165,7 +165,7 @@
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NOT_ALIVE",
-    "246": "NON_PARTICIPATING",
+    "246": "NOT_ALIVE",
     "247": "NOT_ALIVE",
     "248": "SUCCESS",
     "249": "NOT_ALIVE",
@@ -176,12 +176,12 @@
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "NOT_ALIVE",
     "27": "SUCCESS",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
-    "30": "NON_PARTICIPATING",
+    "30": "SUCCESS",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
@@ -198,50 +198,50 @@
     "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
-    "46": "NOT_ALIVE",
+    "46": "SUCCESS",
     "47": "NOT_ALIVE",
     "48": "SUCCESS",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "SUCCESS",
-    "53": "SUCCESS",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "SUCCESS",
-    "58": "NOT_ALIVE",
+    "58": "NON_PARTICIPATING",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
-    "61": "SUCCESS",
+    "60": "NON_PARTICIPATING",
+    "61": "NOT_ALIVE",
     "62": "SUCCESS",
     "63": "NON_PARTICIPATING",
-    "64": "NOT_ALIVE",
+    "64": "NON_PARTICIPATING",
     "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
     "69": "SUCCESS",
-    "7": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
-    "74": "NOT_ALIVE",
+    "74": "NON_PARTICIPATING",
     "75": "SUCCESS",
     "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
     "79": "SUCCESS",
-    "8": "SUCCESS",
+    "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "NOT_ALIVE",
+    "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
+    "85": "SUCCESS",
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
@@ -253,9 +253,9 @@
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "96": "SUCCESS",
     "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
+    "98": "SUCCESS",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6570894,
-    6570926
+    6571012,
+    6571040
   ],
   "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": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6570588,
+  "last_allreduce_block": 6570982,
   "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:cd7dcbdd35f61135ffe87d6f9d369365bafe4f0c3279b7386ea3124902982008
+oid sha256:ff778b957f0952fa6390852452b3129d17006ca8156d6d19a3954a9a1cdd6069
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:666b3a9a1f250a3bbc46d1f804c54eab62dacf2c610cbb7eadab88adb924f732
+oid sha256:34f6ada4fa20b69f5232aee40130dd4ba0c6fd2338e90a28c67c7cc9af724ab5
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6a24b598aa4ddf95f3af99f40f6b5dbcc13a78039efc2d514c5ba1b1a3f0b804
+oid sha256:54ad59a53012b0e2405373f31f0dbc4a93e78c9e92372e43aa58aeb86a551359
 size 4400216536