Run 10. Outer Step 24. Inner Step 5.

config.json

@@ -3,7 +3,7 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "SUCCESS",
-    "10": "SUCCESS",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "SUCCESS",
@@ -11,12 +11,12 @@
     "104": "NOT_ALIVE",
     "105": "SUCCESS",
     "106": "NON_PARTICIPATING",
-    "107": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
     "109": "SUCCESS",
     "11": "NON_PARTICIPATING",
     "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
+    "111": "SUCCESS",
     "112": "NON_PARTICIPATING",
     "113": "SUCCESS",
     "114": "SUCCESS",
@@ -24,53 +24,53 @@
     "116": "NOT_ALIVE",
     "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
-    "119": "NOT_ALIVE",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
-    "125": "SUCCESS",
+    "125": "NON_PARTICIPATING",
     "126": "SUCCESS",
     "127": "NON_PARTICIPATING",
-    "128": "NON_PARTICIPATING",
+    "128": "NOT_ALIVE",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
-    "134": "NON_PARTICIPATING",
+    "134": "NOT_ALIVE",
     "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
-    "139": "NON_PARTICIPATING",
+    "138": "NOT_ALIVE",
+    "139": "NOT_ALIVE",
     "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "SUCCESS",
+    "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
     "146": "SUCCESS",
     "147": "NOT_ALIVE",
     "148": "NON_PARTICIPATING",
-    "149": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
     "15": "SUCCESS",
     "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
-    "153": "SUCCESS",
+    "153": "NOT_ALIVE",
     "154": "SUCCESS",
-    "155": "NON_PARTICIPATING",
+    "155": "SUCCESS",
     "156": "SUCCESS",
     "157": "NON_PARTICIPATING",
     "158": "NOT_ALIVE",
     "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "SUCCESS",
@@ -78,19 +78,19 @@
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "NOT_ALIVE",
     "169": "SUCCESS",
     "17": "NON_PARTICIPATING",
-    "170": "SUCCESS",
+    "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
     "174": "NON_PARTICIPATING",
-    "175": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "SUCCESS",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
     "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
@@ -98,12 +98,12 @@
     "183": "SUCCESS",
     "184": "SUCCESS",
     "185": "NOT_ALIVE",
-    "186": "NOT_ALIVE",
+    "186": "SUCCESS",
     "187": "NOT_ALIVE",
     "188": "SUCCESS",
     "189": "NOT_ALIVE",
-    "19": "NOT_ALIVE",
-    "190": "NOT_ALIVE",
+    "19": "NON_PARTICIPATING",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
     "193": "SUCCESS",
@@ -113,59 +113,59 @@
     "197": "NOT_ALIVE",
     "198": "SUCCESS",
     "199": "NON_PARTICIPATING",
-    "2": "NOT_ALIVE",
+    "2": "NON_PARTICIPATING",
     "20": "SUCCESS",
-    "200": "FAIL",
+    "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
     "205": "SUCCESS",
-    "206": "FAIL",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
-    "21": "NON_PARTICIPATING",
+    "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
     "212": "SUCCESS",
     "213": "NON_PARTICIPATING",
-    "214": "SUCCESS",
-    "215": "NOT_ALIVE",
+    "214": "NON_PARTICIPATING",
+    "215": "NON_PARTICIPATING",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
-    "219": "NOT_ALIVE",
+    "219": "SUCCESS",
     "22": "NON_PARTICIPATING",
     "220": "NOT_ALIVE",
     "221": "SUCCESS",
     "222": "SUCCESS",
-    "223": "NOT_ALIVE",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
-    "225": "NON_PARTICIPATING",
+    "225": "NOT_ALIVE",
     "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "227": "SUCCESS",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
-    "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
+    "233": "FAIL",
+    "234": "NOT_ALIVE",
     "235": "SUCCESS",
-    "236": "NON_PARTICIPATING",
+    "236": "NOT_ALIVE",
     "237": "SUCCESS",
-    "238": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
-    "246": "NON_PARTICIPATING",
+    "245": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
     "249": "NON_PARTICIPATING",
@@ -174,30 +174,30 @@
     "251": "NOT_ALIVE",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
-    "254": "NOT_ALIVE",
+    "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "SUCCESS",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "SUCCESS",
-    "29": "NOT_ALIVE",
+    "29": "SUCCESS",
     "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
+    "30": "NOT_ALIVE",
     "31": "NON_PARTICIPATING",
-    "32": "NOT_ALIVE",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
-    "35": "SUCCESS",
+    "34": "FAIL",
+    "35": "NON_PARTICIPATING",
     "36": "SUCCESS",
     "37": "NOT_ALIVE",
-    "38": "SUCCESS",
-    "39": "SUCCESS",
+    "38": "FAIL",
+    "39": "NOT_ALIVE",
     "4": "SUCCESS",
-    "40": "NOT_ALIVE",
+    "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "SUCCESS",
-    "45": "SUCCESS",
+    "45": "NON_PARTICIPATING",
     "46": "SUCCESS",
     "47": "SUCCESS",
     "48": "NON_PARTICIPATING",
@@ -205,23 +205,23 @@
     "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
-    "52": "NOT_ALIVE",
-    "53": "NON_PARTICIPATING",
-    "54": "NOT_ALIVE",
-    "55": "NON_PARTICIPATING",
+    "52": "NON_PARTICIPATING",
+    "53": "NOT_ALIVE",
+    "54": "NON_PARTICIPATING",
+    "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
     "57": "NOT_ALIVE",
-    "58": "NOT_ALIVE",
+    "58": "NON_PARTICIPATING",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "SUCCESS",
+    "60": "NON_PARTICIPATING",
     "61": "NOT_ALIVE",
-    "62": "NON_PARTICIPATING",
-    "63": "NOT_ALIVE",
+    "62": "NOT_ALIVE",
+    "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
     "65": "NOT_ALIVE",
     "66": "NON_PARTICIPATING",
-    "67": "NOT_ALIVE",
+    "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "SUCCESS",
     "7": "SUCCESS",
@@ -230,32 +230,32 @@
     "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
     "74": "SUCCESS",
-    "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "75": "NON_PARTICIPATING",
+    "76": "NOT_ALIVE",
+    "77": "NON_PARTICIPATING",
+    "78": "NON_PARTICIPATING",
+    "79": "NOT_ALIVE",
     "8": "SUCCESS",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
-    "83": "NOT_ALIVE",
+    "82": "NOT_ALIVE",
+    "83": "NON_PARTICIPATING",
     "84": "SUCCESS",
     "85": "NON_PARTICIPATING",
-    "86": "NOT_ALIVE",
+    "86": "NON_PARTICIPATING",
     "87": "SUCCESS",
     "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
+    "89": "NON_PARTICIPATING",
     "9": "SUCCESS",
-    "90": "NOT_ALIVE",
-    "91": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
+    "91": "SUCCESS",
     "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "SUCCESS",
-    "97": "NOT_ALIVE",
-    "98": "NON_PARTICIPATING",
+    "97": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6480245,
-    6480256
+    6480309,
+    6480323
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 41,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6479877,
+  "last_allreduce_block": 6480247,
   "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 sha256:24040ce21c5913443ed691ba38c330203e3f36c565c700a337776bbccaf9a245
+oid sha256:61f10934a24b2e43752424c34bec79156742ad818bfacf93e7781d8f610c8d7c
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ed8a864ef9727ec349dc2200aaa5c430edbc268ca230364daec2fe0eb3a99e4
+oid sha256:e576339b936c76f65aa55fa0728dd1e04354dfa98c446bee7ea845ca8668dd90
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a624eb3e58f6fffa30c0d855d5c44e77dc4a31d6e8ce1ed92424932c8415d47
+oid sha256:794207d05b8bf04ef1bcd3627adaa5d04d1bb3be7b7c759c1e2e0e8ec0529260
 size 4400216536