Run 10. Outer Step 57. Inner Step 0. Peers 63.

config.json

@@ -2,260 +2,260 @@
   "_name_or_path": "aalv2/c71-h21",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "SUCCESS",
-    "10": "NOT_ALIVE",
+    "1": "NOT_ALIVE",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
-    "102": "FAIL",
+    "101": "SUCCESS",
+    "102": "SUCCESS",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
+    "105": "SUCCESS",
     "106": "SUCCESS",
     "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
-    "109": "FAIL",
+    "109": "SUCCESS",
     "11": "SUCCESS",
     "110": "NON_PARTICIPATING",
-    "111": "FAIL",
-    "112": "FAIL",
-    "113": "NON_PARTICIPATING",
-    "114": "FAIL",
-    "115": "NOT_ALIVE",
-    "116": "FAIL",
-    "117": "NON_PARTICIPATING",
-    "118": "FAIL",
+    "111": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
+    "113": "SUCCESS",
+    "114": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
+    "116": "SUCCESS",
+    "117": "NOT_ALIVE",
+    "118": "SUCCESS",
     "119": "NON_PARTICIPATING",
     "12": "NON_PARTICIPATING",
     "120": "NOT_ALIVE",
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "123": "SUCCESS",
     "124": "NON_PARTICIPATING",
-    "125": "FAIL",
-    "126": "FAIL",
-    "127": "NON_PARTICIPATING",
-    "128": "NON_PARTICIPATING",
-    "129": "SUCCESS",
-    "13": "NON_PARTICIPATING",
+    "125": "NON_PARTICIPATING",
+    "126": "SUCCESS",
+    "127": "NOT_ALIVE",
+    "128": "SUCCESS",
+    "129": "NON_PARTICIPATING",
+    "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NON_PARTICIPATING",
-    "136": "NOT_ALIVE",
+    "136": "NON_PARTICIPATING",
     "137": "NOT_ALIVE",
-    "138": "NOT_ALIVE",
-    "139": "FAIL",
+    "138": "NON_PARTICIPATING",
+    "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
-    "142": "FAIL",
+    "142": "SUCCESS",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
-    "146": "FAIL",
+    "146": "SUCCESS",
     "147": "SUCCESS",
     "148": "NON_PARTICIPATING",
-    "149": "NOT_ALIVE",
-    "15": "NON_PARTICIPATING",
+    "149": "NON_PARTICIPATING",
+    "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "FAIL",
+    "153": "SUCCESS",
     "154": "SUCCESS",
-    "155": "NON_PARTICIPATING",
-    "156": "SUCCESS",
-    "157": "NON_PARTICIPATING",
+    "155": "SUCCESS",
+    "156": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
-    "159": "NON_PARTICIPATING",
+    "159": "SUCCESS",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "NON_PARTICIPATING",
+    "163": "SUCCESS",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "FAIL",
+    "166": "NON_PARTICIPATING",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
     "169": "SUCCESS",
-    "17": "NON_PARTICIPATING",
-    "170": "FAIL",
+    "17": "NOT_ALIVE",
+    "170": "SUCCESS",
     "171": "SUCCESS",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "FAIL",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "FAIL",
+    "177": "NON_PARTICIPATING",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NON_PARTICIPATING",
-    "183": "NOT_ALIVE",
-    "184": "FAIL",
-    "185": "FAIL",
+    "183": "SUCCESS",
+    "184": "SUCCESS",
+    "185": "SUCCESS",
     "186": "SUCCESS",
-    "187": "NOT_ALIVE",
-    "188": "FAIL",
+    "187": "NON_PARTICIPATING",
+    "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
-    "19": "NOT_ALIVE",
+    "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "FAIL",
+    "193": "NON_PARTICIPATING",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
-    "196": "FAIL",
+    "196": "SUCCESS",
     "197": "NON_PARTICIPATING",
-    "198": "NOT_ALIVE",
-    "199": "NOT_ALIVE",
+    "198": "SUCCESS",
+    "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
-    "200": "FAIL",
+    "20": "SUCCESS",
+    "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "202": "NOT_ALIVE",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "SUCCESS",
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
-    "209": "NON_PARTICIPATING",
-    "21": "FAIL",
+    "209": "NOT_ALIVE",
+    "21": "NON_PARTICIPATING",
     "210": "NON_PARTICIPATING",
-    "211": "SUCCESS",
+    "211": "NON_PARTICIPATING",
     "212": "SUCCESS",
     "213": "NON_PARTICIPATING",
-    "214": "FAIL",
+    "214": "SUCCESS",
     "215": "NON_PARTICIPATING",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
-    "218": "NON_PARTICIPATING",
+    "218": "NOT_ALIVE",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
-    "221": "NON_PARTICIPATING",
-    "222": "NON_PARTICIPATING",
+    "221": "SUCCESS",
+    "222": "SUCCESS",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
-    "23": "NON_PARTICIPATING",
+    "229": "SUCCESS",
+    "23": "NOT_ALIVE",
     "230": "SUCCESS",
     "231": "NON_PARTICIPATING",
     "232": "SUCCESS",
-    "233": "NON_PARTICIPATING",
+    "233": "NOT_ALIVE",
     "234": "SUCCESS",
-    "235": "FAIL",
-    "236": "NON_PARTICIPATING",
-    "237": "SUCCESS",
+    "235": "NON_PARTICIPATING",
+    "236": "NOT_ALIVE",
+    "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
+    "239": "NOT_ALIVE",
     "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "FAIL",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
-    "246": "FAIL",
+    "245": "NOT_ALIVE",
+    "246": "SUCCESS",
     "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
-    "25": "SUCCESS",
+    "249": "NOT_ALIVE",
+    "25": "NOT_ALIVE",
     "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
-    "252": "FAIL",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
-    "254": "NOT_ALIVE",
-    "255": "NOT_ALIVE",
-    "26": "FAIL",
+    "254": "NON_PARTICIPATING",
+    "255": "SUCCESS",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
-    "28": "NON_PARTICIPATING",
-    "29": "NOT_ALIVE",
-    "3": "FAIL",
+    "28": "SUCCESS",
+    "29": "NON_PARTICIPATING",
+    "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
-    "31": "NOT_ALIVE",
+    "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
-    "35": "NOT_ALIVE",
+    "35": "SUCCESS",
     "36": "NON_PARTICIPATING",
     "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
+    "4": "SUCCESS",
     "40": "NON_PARTICIPATING",
     "41": "NOT_ALIVE",
     "42": "NON_PARTICIPATING",
     "43": "NOT_ALIVE",
-    "44": "NON_PARTICIPATING",
-    "45": "NON_PARTICIPATING",
+    "44": "NOT_ALIVE",
+    "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NOT_ALIVE",
     "49": "NON_PARTICIPATING",
     "5": "NOT_ALIVE",
-    "50": "NON_PARTICIPATING",
+    "50": "SUCCESS",
     "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
     "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
-    "57": "FAIL",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "FAIL",
-    "61": "NON_PARTICIPATING",
-    "62": "NON_PARTICIPATING",
+    "60": "SUCCESS",
+    "61": "SUCCESS",
+    "62": "NOT_ALIVE",
     "63": "NON_PARTICIPATING",
-    "64": "FAIL",
+    "64": "SUCCESS",
     "65": "NON_PARTICIPATING",
-    "66": "NOT_ALIVE",
+    "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "NON_PARTICIPATING",
-    "7": "NON_PARTICIPATING",
+    "69": "NOT_ALIVE",
+    "7": "SUCCESS",
     "70": "NON_PARTICIPATING",
-    "71": "NON_PARTICIPATING",
-    "72": "NOT_ALIVE",
-    "73": "NON_PARTICIPATING",
-    "74": "NOT_ALIVE",
+    "71": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
+    "74": "SUCCESS",
     "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
     "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
+    "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
-    "84": "FAIL",
+    "84": "SUCCESS",
     "85": "NON_PARTICIPATING",
-    "86": "SUCCESS",
-    "87": "NOT_ALIVE",
+    "86": "NOT_ALIVE",
+    "87": "NON_PARTICIPATING",
     "88": "NOT_ALIVE",
-    "89": "FAIL",
+    "89": "SUCCESS",
     "9": "NON_PARTICIPATING",
-    "90": "FAIL",
+    "90": "SUCCESS",
     "91": "NON_PARTICIPATING",
-    "92": "SUCCESS",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "SUCCESS",
     "95": "NOT_ALIVE",
-    "96": "FAIL",
+    "96": "SUCCESS",
     "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
+    "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6497755,
+  "last_allreduce_block": 6498111,
   "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 sha256:88dcad4b02928c090ef38d21922a83b98c9971abe502aeb79f0fb5fdf627bdbd
+oid sha256:96cf26eed0c93da4abd3ac84013cfe2220b3cf12edd85ddd32601fc280d040c0
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d72e3fc5c2187a89b77186915494ffe0740069c479e30e0d06db057bcfdb94a5
+oid sha256:4bce493d10bbc521d55c1b4425c81ba990b1688e1ea3dc6ac10f840b6cfeee50
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3d562bfd1e8bf0db337a46ff44fe40c62a618d08f966bccee986b1689c23c5e
+oid sha256:7a5f400497020913406e532290000868fe95cdb74d3c163bbc75e39dc5b0519d
 size 4400252568