Run 10. Outer Step 64. Inner Step 0. Peers 63.

config.json

@@ -1,55 +1,55 @@
 {
-  "_name_or_path": "aalv2/c70-h5",
+  "_name_or_path": "aalv2/c71-h21",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NON_PARTICIPATING",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "FAIL",
-    "102": "FAIL",
+    "101": "SUCCESS",
+    "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
-    "106": "FAIL",
-    "107": "NON_PARTICIPATING",
-    "108": "NOT_ALIVE",
-    "109": "FAIL",
+    "105": "SUCCESS",
+    "106": "NON_PARTICIPATING",
+    "107": "NOT_ALIVE",
+    "108": "NON_PARTICIPATING",
+    "109": "SUCCESS",
     "11": "SUCCESS",
-    "110": "NOT_ALIVE",
-    "111": "FAIL",
-    "112": "NOT_ALIVE",
-    "113": "NOT_ALIVE",
-    "114": "FAIL",
+    "110": "NON_PARTICIPATING",
+    "111": "NON_PARTICIPATING",
+    "112": "NON_PARTICIPATING",
+    "113": "NON_PARTICIPATING",
+    "114": "SUCCESS",
     "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
-    "117": "NON_PARTICIPATING",
-    "118": "FAIL",
+    "116": "SUCCESS",
+    "117": "SUCCESS",
+    "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
-    "12": "NOT_ALIVE",
+    "12": "NON_PARTICIPATING",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "SUCCESS",
-    "123": "FAIL",
-    "124": "NON_PARTICIPATING",
-    "125": "FAIL",
-    "126": "FAIL",
+    "123": "SUCCESS",
+    "124": "SUCCESS",
+    "125": "SUCCESS",
+    "126": "SUCCESS",
     "127": "NON_PARTICIPATING",
-    "128": "SUCCESS",
+    "128": "NOT_ALIVE",
     "129": "SUCCESS",
-    "13": "NON_PARTICIPATING",
+    "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "NOT_ALIVE",
+    "134": "NON_PARTICIPATING",
+    "135": "NON_PARTICIPATING",
     "136": "NON_PARTICIPATING",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NOT_ALIVE",
+    "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
-    "141": "NON_PARTICIPATING",
+    "141": "NOT_ALIVE",
     "142": "SUCCESS",
     "143": "NOT_ALIVE",
     "144": "NON_PARTICIPATING",
@@ -57,120 +57,120 @@
     "146": "SUCCESS",
     "147": "SUCCESS",
     "148": "NON_PARTICIPATING",
-    "149": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "SUCCESS",
-    "155": "NOT_ALIVE",
+    "155": "NON_PARTICIPATING",
     "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
-    "159": "FAIL",
+    "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
-    "161": "NOT_ALIVE",
-    "162": "NOT_ALIVE",
-    "163": "FAIL",
+    "160": "NON_PARTICIPATING",
+    "161": "NON_PARTICIPATING",
+    "162": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
-    "17": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
+    "17": "SUCCESS",
     "170": "SUCCESS",
-    "171": "NON_PARTICIPATING",
+    "171": "SUCCESS",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "FAIL",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "NOT_ALIVE",
     "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
-    "18": "NOT_ALIVE",
+    "18": "NON_PARTICIPATING",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "FAIL",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
-    "188": "NOT_ALIVE",
+    "184": "SUCCESS",
+    "185": "NOT_ALIVE",
+    "186": "SUCCESS",
+    "187": "NOT_ALIVE",
+    "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
-    "19": "NOT_ALIVE",
-    "190": "NOT_ALIVE",
+    "19": "NON_PARTICIPATING",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "NOT_ALIVE",
-    "198": "FAIL",
+    "197": "NON_PARTICIPATING",
+    "198": "SUCCESS",
     "199": "SUCCESS",
-    "2": "NOT_ALIVE",
+    "2": "NON_PARTICIPATING",
     "20": "SUCCESS",
-    "200": "NON_PARTICIPATING",
-    "201": "NOT_ALIVE",
-    "202": "NOT_ALIVE",
-    "203": "NOT_ALIVE",
+    "200": "SUCCESS",
+    "201": "NON_PARTICIPATING",
+    "202": "NON_PARTICIPATING",
+    "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "SUCCESS",
-    "206": "FAIL",
+    "205": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
     "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
+    "212": "SUCCESS",
     "213": "NOT_ALIVE",
-    "214": "NOT_ALIVE",
+    "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NOT_ALIVE",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
-    "221": "FAIL",
-    "222": "NON_PARTICIPATING",
-    "223": "FAIL",
+    "220": "NOT_ALIVE",
+    "221": "SUCCESS",
+    "222": "SUCCESS",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
-    "227": "FAIL",
-    "228": "NON_PARTICIPATING",
-    "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
-    "230": "FAIL",
-    "231": "NOT_ALIVE",
+    "227": "NON_PARTICIPATING",
+    "228": "NOT_ALIVE",
+    "229": "NON_PARTICIPATING",
+    "23": "NON_PARTICIPATING",
+    "230": "NON_PARTICIPATING",
+    "231": "NON_PARTICIPATING",
     "232": "SUCCESS",
     "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
-    "235": "FAIL",
-    "236": "NOT_ALIVE",
+    "234": "NOT_ALIVE",
+    "235": "NON_PARTICIPATING",
+    "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
     "239": "NON_PARTICIPATING",
-    "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
-    "242": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
+    "240": "NON_PARTICIPATING",
+    "241": "SUCCESS",
+    "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
-    "246": "FAIL",
+    "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
     "248": "NON_PARTICIPATING",
     "249": "NOT_ALIVE",
     "25": "NOT_ALIVE",
-    "250": "NOT_ALIVE",
+    "250": "SUCCESS",
     "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
@@ -178,25 +178,25 @@
     "255": "SUCCESS",
     "26": "SUCCESS",
     "27": "NOT_ALIVE",
-    "28": "NON_PARTICIPATING",
-    "29": "FAIL",
-    "3": "FAIL",
+    "28": "NOT_ALIVE",
+    "29": "SUCCESS",
+    "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
-    "31": "NON_PARTICIPATING",
-    "32": "NOT_ALIVE",
+    "31": "SUCCESS",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "NOT_ALIVE",
-    "35": "SUCCESS",
-    "36": "NOT_ALIVE",
+    "34": "NON_PARTICIPATING",
+    "35": "NOT_ALIVE",
+    "36": "NON_PARTICIPATING",
     "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "SUCCESS",
     "40": "NON_PARTICIPATING",
     "41": "NON_PARTICIPATING",
-    "42": "FAIL",
+    "42": "SUCCESS",
     "43": "NOT_ALIVE",
-    "44": "FAIL",
+    "44": "NOT_ALIVE",
     "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
@@ -204,58 +204,58 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "SUCCESS",
-    "51": "FAIL",
+    "51": "NON_PARTICIPATING",
     "52": "SUCCESS",
     "53": "NON_PARTICIPATING",
-    "54": "NOT_ALIVE",
+    "54": "NON_PARTICIPATING",
     "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
-    "57": "FAIL",
-    "58": "NOT_ALIVE",
-    "59": "FAIL",
+    "57": "SUCCESS",
+    "58": "NON_PARTICIPATING",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
-    "60": "FAIL",
-    "61": "FAIL",
+    "60": "SUCCESS",
+    "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
-    "66": "NON_PARTICIPATING",
+    "66": "SUCCESS",
     "67": "NON_PARTICIPATING",
-    "68": "NOT_ALIVE",
+    "68": "NON_PARTICIPATING",
     "69": "NON_PARTICIPATING",
-    "7": "FAIL",
-    "70": "NON_PARTICIPATING",
+    "7": "NON_PARTICIPATING",
+    "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
-    "72": "NOT_ALIVE",
-    "73": "NOT_ALIVE",
-    "74": "FAIL",
-    "75": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
+    "73": "NON_PARTICIPATING",
+    "74": "NOT_ALIVE",
+    "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
-    "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
+    "77": "NON_PARTICIPATING",
+    "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NON_PARTICIPATING",
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "FAIL",
+    "82": "SUCCESS",
     "83": "NOT_ALIVE",
     "84": "SUCCESS",
     "85": "NON_PARTICIPATING",
     "86": "SUCCESS",
-    "87": "FAIL",
-    "88": "NOT_ALIVE",
-    "89": "FAIL",
+    "87": "SUCCESS",
+    "88": "NON_PARTICIPATING",
+    "89": "SUCCESS",
     "9": "NON_PARTICIPATING",
-    "90": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
-    "92": "NOT_ALIVE",
+    "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "SUCCESS",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
-    "97": "NOT_ALIVE",
-    "98": "NON_PARTICIPATING",
+    "96": "SUCCESS",
+    "97": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6500846,
+  "last_allreduce_block": 6501715,
   "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 sha256:431d356501c3c9d86fcbe4eea5a0711ec221de0632726c6a5e2459b16d5225a5
+oid sha256:41058fc9f29e88b9dfcfbab86b6449d6c5115e95acfa2e32ec0ce5c9bf2db5c1
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53c592f314872be5ab1070bb63ac9bc63367e48041483f460d9a317fc3473062
+oid sha256:ffe8b3707f517b8ae87fe38774d8d7ea288c35df36f3d262e4e0ea11a667803e
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d575e2dcd2e2bba56a10a804916dee0376591e84b0cf0a40eaa3d05e49751739
+oid sha256:195141002518506ec08bca371ac98f00acd981cca2a41dd044e2678ae67d27ed
 size 4400252568