Run 11. Outer Step 311. Inner Step 9.

config.json

@@ -8,19 +8,19 @@
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
+    "104": "SUCCESS",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
+    "107": "SUCCESS",
     "108": "NON_PARTICIPATING",
-    "109": "NOT_ALIVE",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
     "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
@@ -35,21 +35,21 @@
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
     "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
-    "13": "SUCCESS",
-    "130": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
+    "13": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
-    "135": "SUCCESS",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "NOT_ALIVE",
-    "141": "SUCCESS",
+    "140": "SUCCESS",
+    "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
     "143": "SUCCESS",
     "144": "NOT_ALIVE",
@@ -58,11 +58,11 @@
     "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": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -70,7 +70,7 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
@@ -96,10 +96,10 @@
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
-    "184": "NON_PARTICIPATING",
+    "184": "SUCCESS",
     "185": "NON_PARTICIPATING",
-    "186": "NOT_ALIVE",
-    "187": "SUCCESS",
+    "186": "NON_PARTICIPATING",
+    "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -110,8 +110,8 @@
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "NOT_ALIVE",
-    "198": "NOT_ALIVE",
+    "197": "SUCCESS",
+    "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
@@ -121,7 +121,7 @@
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
@@ -130,7 +130,7 @@
     "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NOT_ALIVE",
+    "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
@@ -151,16 +151,16 @@
     "230": "SUCCESS",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
+    "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
+    "236": "NOT_ALIVE",
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "FAIL",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "NOT_ALIVE",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -171,24 +171,24 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
-    "252": "NON_PARTICIPATING",
+    "251": "NOT_ALIVE",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "SUCCESS",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "SUCCESS",
+    "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
@@ -204,42 +204,42 @@
     "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": "SUCCESS",
+    "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "SUCCESS",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
-    "61": "NON_PARTICIPATING",
+    "60": "NON_PARTICIPATING",
+    "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
-    "63": "SUCCESS",
+    "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "FAIL",
     "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
     "74": "NOT_ALIVE",
     "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
-    "79": "NOT_ALIVE",
-    "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
-    "81": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
+    "79": "NON_PARTICIPATING",
+    "8": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
+    "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
+    "83": "NOT_ALIVE",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -249,9 +249,9 @@
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
-    "92": "NON_PARTICIPATING",
+    "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "SUCCESS",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NON_PARTICIPATING",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6705874,
-    6705896
+    6705974,
+    6705994
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 90,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6705165,
+  "last_allreduce_block": 6705905,
   "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 sha256:99899105908478267b579bf30d2fecf425c8afe3e18a6a3fc858129c7bafd42d
+oid sha256:a6171d4de02863f4a6783266c71576792bf946465d9650936450d67f72e22a6c
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9901d18a86ef868316f21b0720b1851135c181abf61733c6933adaeb399ae0cd
+oid sha256:ab56b274227ee7a41a47eb2a415bf76163780da842ddaa345a7bb2000425027b
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3aaefde436df971a9cc4690edec31776dddec7f46df314972cd82356a932e7d
+oid sha256:2d655f962143a54c019e586c88b41b0f595a5714b4a58efa21dafc988c7ec2f0
 size 4400216536