Run 9. Outer Step 8. Inner Step 0. Peers 1.

config.json

@@ -3,104 +3,104 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "FAIL",
-    "10": "NON_PARTICIPATING",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "NON_PARTICIPATING",
+    "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
-    "105": "FAIL",
+    "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
-    "107": "FAIL",
+    "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "FAIL",
+    "109": "NON_PARTICIPATING",
     "11": "NON_PARTICIPATING",
-    "110": "FAIL",
-    "111": "NOT_ALIVE",
-    "112": "NON_PARTICIPATING",
+    "110": "NON_PARTICIPATING",
+    "111": "NON_PARTICIPATING",
+    "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
-    "117": "FAIL",
+    "116": "FAIL",
+    "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
-    "120": "FAIL",
+    "120": "NON_PARTICIPATING",
     "121": "NOT_ALIVE",
-    "122": "FAIL",
-    "123": "FAIL",
+    "122": "NON_PARTICIPATING",
+    "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
-    "125": "FAIL",
-    "126": "NOT_ALIVE",
-    "127": "NON_PARTICIPATING",
-    "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
-    "13": "FAIL",
-    "130": "NOT_ALIVE",
+    "125": "NOT_ALIVE",
+    "126": "NON_PARTICIPATING",
+    "127": "FAIL",
+    "128": "FAIL",
+    "129": "NON_PARTICIPATING",
+    "13": "NON_PARTICIPATING",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NON_PARTICIPATING",
-    "135": "FAIL",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
-    "137": "NON_PARTICIPATING",
+    "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
     "139": "NOT_ALIVE",
-    "14": "NON_PARTICIPATING",
+    "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
-    "141": "NON_PARTICIPATING",
+    "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
-    "146": "NOT_ALIVE",
-    "147": "NON_PARTICIPATING",
-    "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
-    "15": "FAIL",
+    "146": "NON_PARTICIPATING",
+    "147": "FAIL",
+    "148": "FAIL",
+    "149": "FAIL",
+    "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NON_PARTICIPATING",
-    "153": "NOT_ALIVE",
+    "152": "NOT_ALIVE",
+    "153": "NON_PARTICIPATING",
     "154": "NON_PARTICIPATING",
-    "155": "NON_PARTICIPATING",
-    "156": "NON_PARTICIPATING",
-    "157": "NON_PARTICIPATING",
-    "158": "NON_PARTICIPATING",
-    "159": "NON_PARTICIPATING",
-    "16": "FAIL",
-    "160": "NON_PARTICIPATING",
+    "155": "FAIL",
+    "156": "FAIL",
+    "157": "FAIL",
+    "158": "NOT_ALIVE",
+    "159": "FAIL",
+    "16": "NON_PARTICIPATING",
+    "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "FAIL",
-    "164": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
+    "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
-    "167": "FAIL",
+    "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
-    "17": "NON_PARTICIPATING",
-    "170": "FAIL",
+    "169": "FAIL",
+    "17": "FAIL",
+    "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
-    "174": "NON_PARTICIPATING",
-    "175": "NON_PARTICIPATING",
+    "173": "FAIL",
+    "174": "NOT_ALIVE",
+    "175": "FAIL",
     "176": "NOT_ALIVE",
-    "177": "FAIL",
-    "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
-    "18": "NON_PARTICIPATING",
-    "180": "NON_PARTICIPATING",
+    "177": "NON_PARTICIPATING",
+    "178": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
+    "18": "NOT_ALIVE",
+    "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "FAIL",
+    "182": "NON_PARTICIPATING",
     "183": "NON_PARTICIPATING",
     "184": "FAIL",
-    "185": "FAIL",
+    "185": "NOT_ALIVE",
     "186": "FAIL",
     "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
+    "188": "FAIL",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
@@ -109,153 +109,153 @@
     "193": "NON_PARTICIPATING",
     "194": "FAIL",
     "195": "NON_PARTICIPATING",
-    "196": "NOT_ALIVE",
-    "197": "NOT_ALIVE",
+    "196": "NON_PARTICIPATING",
+    "197": "FAIL",
     "198": "FAIL",
-    "199": "NON_PARTICIPATING",
-    "2": "NON_PARTICIPATING",
+    "199": "FAIL",
+    "2": "FAIL",
     "20": "FAIL",
     "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
-    "203": "FAIL",
+    "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "NON_PARTICIPATING",
+    "205": "NOT_ALIVE",
     "206": "FAIL",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
-    "209": "FAIL",
-    "21": "NON_PARTICIPATING",
-    "210": "NON_PARTICIPATING",
-    "211": "FAIL",
+    "208": "FAIL",
+    "209": "NON_PARTICIPATING",
+    "21": "FAIL",
+    "210": "FAIL",
+    "211": "NON_PARTICIPATING",
     "212": "FAIL",
     "213": "NON_PARTICIPATING",
-    "214": "NOT_ALIVE",
-    "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
-    "217": "NOT_ALIVE",
-    "218": "NON_PARTICIPATING",
-    "219": "FAIL",
+    "214": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
+    "216": "FAIL",
+    "217": "FAIL",
+    "218": "FAIL",
+    "219": "NON_PARTICIPATING",
     "22": "NON_PARTICIPATING",
-    "220": "NON_PARTICIPATING",
+    "220": "FAIL",
     "221": "FAIL",
-    "222": "NON_PARTICIPATING",
+    "222": "FAIL",
     "223": "FAIL",
     "224": "NOT_ALIVE",
-    "225": "NON_PARTICIPATING",
-    "226": "NOT_ALIVE",
-    "227": "NON_PARTICIPATING",
+    "225": "FAIL",
+    "226": "NON_PARTICIPATING",
+    "227": "FAIL",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
-    "231": "FAIL",
-    "232": "NON_PARTICIPATING",
+    "231": "NON_PARTICIPATING",
+    "232": "FAIL",
     "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
+    "234": "FAIL",
     "235": "NON_PARTICIPATING",
-    "236": "NON_PARTICIPATING",
-    "237": "NON_PARTICIPATING",
+    "236": "NOT_ALIVE",
+    "237": "FAIL",
     "238": "NON_PARTICIPATING",
     "239": "FAIL",
     "24": "FAIL",
-    "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
-    "243": "NON_PARTICIPATING",
+    "240": "NON_PARTICIPATING",
+    "241": "FAIL",
+    "242": "NOT_ALIVE",
+    "243": "FAIL",
     "244": "NOT_ALIVE",
-    "245": "NOT_ALIVE",
-    "246": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
+    "246": "NON_PARTICIPATING",
     "247": "FAIL",
-    "248": "NOT_ALIVE",
+    "248": "NON_PARTICIPATING",
     "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "FAIL",
-    "251": "NOT_ALIVE",
-    "252": "NOT_ALIVE",
-    "253": "NON_PARTICIPATING",
-    "254": "NON_PARTICIPATING",
+    "250": "NON_PARTICIPATING",
+    "251": "NON_PARTICIPATING",
+    "252": "NON_PARTICIPATING",
+    "253": "NOT_ALIVE",
+    "254": "NOT_ALIVE",
     "255": "NON_PARTICIPATING",
     "26": "NOT_ALIVE",
-    "27": "NON_PARTICIPATING",
-    "28": "FAIL",
+    "27": "FAIL",
+    "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
     "3": "NON_PARTICIPATING",
     "30": "NON_PARTICIPATING",
-    "31": "FAIL",
-    "32": "NON_PARTICIPATING",
+    "31": "NOT_ALIVE",
+    "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "FAIL",
-    "35": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
+    "35": "FAIL",
     "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "FAIL",
-    "39": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
+    "39": "FAIL",
     "4": "NON_PARTICIPATING",
-    "40": "FAIL",
+    "40": "NOT_ALIVE",
     "41": "NOT_ALIVE",
-    "42": "FAIL",
+    "42": "NON_PARTICIPATING",
     "43": "FAIL",
-    "44": "NON_PARTICIPATING",
+    "44": "FAIL",
     "45": "NON_PARTICIPATING",
-    "46": "NOT_ALIVE",
+    "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
-    "48": "NON_PARTICIPATING",
+    "48": "NOT_ALIVE",
     "49": "NON_PARTICIPATING",
     "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
-    "51": "FAIL",
+    "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
-    "53": "FAIL",
-    "54": "NON_PARTICIPATING",
+    "53": "NON_PARTICIPATING",
+    "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
     "57": "NON_PARTICIPATING",
-    "58": "NON_PARTICIPATING",
-    "59": "NOT_ALIVE",
-    "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "NOT_ALIVE",
-    "62": "FAIL",
+    "58": "NOT_ALIVE",
+    "59": "NON_PARTICIPATING",
+    "6": "NON_PARTICIPATING",
+    "60": "FAIL",
+    "61": "FAIL",
+    "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
     "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
-    "68": "FAIL",
-    "69": "FAIL",
-    "7": "NON_PARTICIPATING",
+    "68": "NOT_ALIVE",
+    "69": "NON_PARTICIPATING",
+    "7": "FAIL",
     "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
-    "72": "NON_PARTICIPATING",
+    "72": "FAIL",
     "73": "NON_PARTICIPATING",
-    "74": "FAIL",
+    "74": "NON_PARTICIPATING",
     "75": "NON_PARTICIPATING",
-    "76": "FAIL",
-    "77": "NON_PARTICIPATING",
-    "78": "FAIL",
+    "76": "NON_PARTICIPATING",
+    "77": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
+    "84": "NOT_ALIVE",
     "85": "NON_PARTICIPATING",
-    "86": "NOT_ALIVE",
-    "87": "NON_PARTICIPATING",
+    "86": "NON_PARTICIPATING",
+    "87": "FAIL",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NON_PARTICIPATING",
     "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
-    "92": "FAIL",
-    "93": "NON_PARTICIPATING",
-    "94": "NON_PARTICIPATING",
-    "95": "NON_PARTICIPATING",
-    "96": "NOT_ALIVE",
-    "97": "FAIL",
-    "98": "FAIL",
+    "92": "NOT_ALIVE",
+    "93": "NOT_ALIVE",
+    "94": "FAIL",
+    "95": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
+    "97": "NON_PARTICIPATING",
+    "98": "NON_PARTICIPATING",
     "99": "FAIL"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6452261,
+  "last_allreduce_block": 6452845,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a35476e45535d9ad8c55d737c4dbcdec389160c74129099eb630daf553a28b58
+oid sha256:5623d7071d738f02e654a5406c811030ff3dbd199f1338aa9c9b0317b47faa86
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:204b80822e06005479f64f9aed389d26a29b4e84d4921cdd015819670593a9c7
+oid sha256:3ddcaf0ae7ce71e9b502a64659b5d4c9180d891d81b0282ed986558284ae0302
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6509903451dbafd12b16925bb0cda3faa3dbcd037435aa582545ba06e49002a
+oid sha256:f0783474703bc64e7389bd528fe56cdc7272ac4d356d3cf948f9037370c813aa
 size 4400252568