Run 10. Outer Step 30. Inner Step 0. Peers 49.

config.json

@@ -2,47 +2,47 @@
   "_name_or_path": "aalv2/c71-h21",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "FAIL",
-    "10": "FAIL",
+    "1": "SUCCESS",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "FAIL",
+    "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
     "105": "NOT_ALIVE",
     "106": "NON_PARTICIPATING",
-    "107": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "FAIL",
-    "11": "FAIL",
+    "109": "NON_PARTICIPATING",
+    "11": "NON_PARTICIPATING",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "FAIL",
-    "113": "FAIL",
-    "114": "FAIL",
-    "115": "NOT_ALIVE",
-    "116": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
+    "113": "SUCCESS",
+    "114": "NON_PARTICIPATING",
+    "115": "NON_PARTICIPATING",
+    "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
-    "118": "FAIL",
-    "119": "NOT_ALIVE",
+    "118": "NON_PARTICIPATING",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
-    "122": "NON_PARTICIPATING",
-    "123": "FAIL",
+    "121": "NOT_ALIVE",
+    "122": "SUCCESS",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
-    "125": "FAIL",
-    "126": "FAIL",
-    "127": "NON_PARTICIPATING",
+    "125": "SUCCESS",
+    "126": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
     "129": "NOT_ALIVE",
-    "13": "FAIL",
+    "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
-    "135": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
@@ -50,22 +50,22 @@
     "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "FAIL",
+    "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
-    "15": "NON_PARTICIPATING",
+    "149": "SUCCESS",
+    "15": "SUCCESS",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
-    "154": "NOT_ALIVE",
-    "155": "NON_PARTICIPATING",
-    "156": "FAIL",
+    "153": "SUCCESS",
+    "154": "NON_PARTICIPATING",
+    "155": "SUCCESS",
+    "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
     "158": "NOT_ALIVE",
     "159": "NON_PARTICIPATING",
@@ -73,190 +73,190 @@
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "FAIL",
+    "163": "SUCCESS",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "NON_PARTICIPATING",
+    "167": "SUCCESS",
     "168": "NON_PARTICIPATING",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NON_PARTICIPATING",
-    "170": "FAIL",
-    "171": "FAIL",
+    "170": "SUCCESS",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NON_PARTICIPATING",
-    "183": "FAIL",
-    "184": "FAIL",
-    "185": "FAIL",
-    "186": "NOT_ALIVE",
+    "183": "NOT_ALIVE",
+    "184": "SUCCESS",
+    "185": "SUCCESS",
+    "186": "SUCCESS",
     "187": "NOT_ALIVE",
-    "188": "NON_PARTICIPATING",
+    "188": "SUCCESS",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
-    "192": "NOT_ALIVE",
+    "192": "NON_PARTICIPATING",
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
     "195": "NOT_ALIVE",
-    "196": "FAIL",
+    "196": "SUCCESS",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
-    "2": "NON_PARTICIPATING",
-    "20": "FAIL",
-    "200": "FAIL",
+    "2": "SUCCESS",
+    "20": "NON_PARTICIPATING",
+    "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "FAIL",
-    "206": "FAIL",
+    "205": "NON_PARTICIPATING",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
-    "209": "NON_PARTICIPATING",
-    "21": "FAIL",
+    "209": "NOT_ALIVE",
+    "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
-    "211": "FAIL",
+    "211": "NOT_ALIVE",
     "212": "NON_PARTICIPATING",
-    "213": "NON_PARTICIPATING",
-    "214": "FAIL",
+    "213": "SUCCESS",
+    "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
-    "219": "FAIL",
+    "219": "NON_PARTICIPATING",
     "22": "NON_PARTICIPATING",
     "220": "NOT_ALIVE",
-    "221": "NON_PARTICIPATING",
-    "222": "FAIL",
+    "221": "SUCCESS",
+    "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
-    "225": "NON_PARTICIPATING",
+    "225": "SUCCESS",
     "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "SUCCESS",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
-    "232": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
+    "232": "SUCCESS",
     "233": "NON_PARTICIPATING",
     "234": "NOT_ALIVE",
-    "235": "FAIL",
+    "235": "SUCCESS",
     "236": "NON_PARTICIPATING",
-    "237": "NON_PARTICIPATING",
-    "238": "FAIL",
+    "237": "SUCCESS",
+    "238": "SUCCESS",
     "239": "NON_PARTICIPATING",
     "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
+    "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "SUCCESS",
     "251": "NOT_ALIVE",
-    "252": "FAIL",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
-    "254": "FAIL",
+    "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "FAIL",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
-    "28": "FAIL",
-    "29": "FAIL",
+    "28": "NOT_ALIVE",
+    "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "FAIL",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "FAIL",
-    "35": "FAIL",
-    "36": "FAIL",
+    "34": "SUCCESS",
+    "35": "NON_PARTICIPATING",
+    "36": "SUCCESS",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
-    "39": "FAIL",
-    "4": "FAIL",
-    "40": "FAIL",
-    "41": "NOT_ALIVE",
+    "39": "NON_PARTICIPATING",
+    "4": "SUCCESS",
+    "40": "SUCCESS",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
-    "44": "FAIL",
-    "45": "FAIL",
-    "46": "FAIL",
-    "47": "FAIL",
+    "44": "SUCCESS",
+    "45": "NON_PARTICIPATING",
+    "46": "NON_PARTICIPATING",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
-    "49": "NOT_ALIVE",
-    "5": "NOT_ALIVE",
+    "49": "NON_PARTICIPATING",
+    "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
-    "52": "NON_PARTICIPATING",
-    "53": "FAIL",
+    "52": "NOT_ALIVE",
+    "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NOT_ALIVE",
-    "57": "FAIL",
+    "56": "NON_PARTICIPATING",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
     "59": "NON_PARTICIPATING",
     "6": "NOT_ALIVE",
     "60": "NOT_ALIVE",
     "61": "NOT_ALIVE",
-    "62": "FAIL",
+    "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
-    "64": "FAIL",
+    "64": "NOT_ALIVE",
     "65": "NOT_ALIVE",
     "66": "NON_PARTICIPATING",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "FAIL",
-    "7": "FAIL",
-    "70": "NON_PARTICIPATING",
-    "71": "NON_PARTICIPATING",
+    "69": "SUCCESS",
+    "7": "NON_PARTICIPATING",
+    "70": "SUCCESS",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
-    "74": "FAIL",
+    "74": "SUCCESS",
     "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
+    "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
-    "8": "FAIL",
+    "8": "SUCCESS",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
-    "84": "FAIL",
+    "84": "SUCCESS",
     "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
-    "87": "FAIL",
-    "88": "NON_PARTICIPATING",
+    "87": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
-    "9": "NON_PARTICIPATING",
-    "90": "FAIL",
-    "91": "FAIL",
-    "92": "NOT_ALIVE",
+    "9": "SUCCESS",
+    "90": "NON_PARTICIPATING",
+    "91": "NON_PARTICIPATING",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "FAIL",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "NON_PARTICIPATING",
-    "99": "NOT_ALIVE"
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6482613,
+  "last_allreduce_block": 6483009,
   "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:d8e4f3363e37e97f4cefb5ea8e6e334c7fb045cef4f3df39d6f9aff2a3aaeae7
+oid sha256:7206cbbbe057472c4c5b6f85880b4bca9c612449fb244739eee019dd4a60b115
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6529c0683531c46da25b9c2d3952f26906738ac64eec66de405d6117b7329ecf
+oid sha256:ac7354edff00e0bb745fa49521d1ba7c68626bb80ee2dfc3e259dd2a8246acbc
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:640c575d8b960fdb0ea60799a62f07a0627efac56e6254edb6a4ae2caae77dcf
+oid sha256:4f968c69730682ea070e67df30c8e524aff7a2458445fdc0ba65482fe73cf250
 size 4400252568