Run 10. Outer Step 70. Inner Step 0. Peers 27.

config.json

@@ -1,145 +1,145 @@
 {
-  "_name_or_path": "strnan01/llama-1b-hot-6",
+  "_name_or_path": "aalv2/c70-h11",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "SUCCESS",
-    "10": "NON_PARTICIPATING",
+    "1": "FAIL",
+    "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
-    "102": "SUCCESS",
+    "101": "SUCCESS",
+    "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "SUCCESS",
-    "106": "NON_PARTICIPATING",
-    "107": "NON_PARTICIPATING",
+    "105": "FAIL",
+    "106": "NOT_ALIVE",
+    "107": "SUCCESS",
     "108": "NON_PARTICIPATING",
-    "109": "NON_PARTICIPATING",
-    "11": "SUCCESS",
-    "110": "NOT_ALIVE",
-    "111": "SUCCESS",
-    "112": "NON_PARTICIPATING",
+    "109": "FAIL",
+    "11": "NON_PARTICIPATING",
+    "110": "FAIL",
+    "111": "FAIL",
+    "112": "SUCCESS",
     "113": "NOT_ALIVE",
     "114": "SUCCESS",
-    "115": "NON_PARTICIPATING",
-    "116": "SUCCESS",
-    "117": "SUCCESS",
-    "118": "SUCCESS",
-    "119": "NON_PARTICIPATING",
+    "115": "NOT_ALIVE",
+    "116": "NON_PARTICIPATING",
+    "117": "NON_PARTICIPATING",
+    "118": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NON_PARTICIPATING",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "FAIL",
     "122": "NON_PARTICIPATING",
-    "123": "SUCCESS",
+    "123": "NON_PARTICIPATING",
     "124": "SUCCESS",
-    "125": "SUCCESS",
-    "126": "SUCCESS",
+    "125": "NON_PARTICIPATING",
+    "126": "FAIL",
     "127": "NON_PARTICIPATING",
-    "128": "NOT_ALIVE",
-    "129": "NON_PARTICIPATING",
-    "13": "NON_PARTICIPATING",
+    "128": "NON_PARTICIPATING",
+    "129": "SUCCESS",
+    "13": "FAIL",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NON_PARTICIPATING",
-    "136": "NON_PARTICIPATING",
+    "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
-    "139": "NOT_ALIVE",
+    "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
-    "142": "SUCCESS",
+    "142": "FAIL",
     "143": "NOT_ALIVE",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "SUCCESS",
+    "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "NOT_ALIVE",
-    "149": "NON_PARTICIPATING",
+    "148": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
-    "150": "NON_PARTICIPATING",
+    "150": "FAIL",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "SUCCESS",
-    "154": "SUCCESS",
+    "154": "FAIL",
     "155": "NOT_ALIVE",
-    "156": "NON_PARTICIPATING",
+    "156": "FAIL",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
-    "159": "SUCCESS",
+    "159": "FAIL",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
     "161": "SUCCESS",
     "162": "NOT_ALIVE",
-    "163": "NOT_ALIVE",
+    "163": "SUCCESS",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NON_PARTICIPATING",
-    "167": "NON_PARTICIPATING",
-    "168": "SUCCESS",
+    "167": "FAIL",
+    "168": "NON_PARTICIPATING",
     "169": "NON_PARTICIPATING",
-    "17": "NON_PARTICIPATING",
-    "170": "SUCCESS",
-    "171": "NOT_ALIVE",
+    "17": "NOT_ALIVE",
+    "170": "FAIL",
+    "171": "SUCCESS",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "SUCCESS",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
-    "184": "SUCCESS",
-    "185": "NOT_ALIVE",
-    "186": "NON_PARTICIPATING",
-    "187": "SUCCESS",
-    "188": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
+    "185": "NON_PARTICIPATING",
+    "186": "FAIL",
+    "187": "FAIL",
+    "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "SUCCESS",
+    "193": "FAIL",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
-    "196": "SUCCESS",
+    "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
-    "198": "SUCCESS",
+    "198": "FAIL",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
+    "20": "FAIL",
     "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "206": "FAIL",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
-    "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
+    "21": "NON_PARTICIPATING",
+    "210": "SUCCESS",
     "211": "NON_PARTICIPATING",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
-    "214": "SUCCESS",
+    "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NOT_ALIVE",
-    "219": "NON_PARTICIPATING",
+    "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
-    "220": "NOT_ALIVE",
-    "221": "SUCCESS",
-    "222": "SUCCESS",
+    "220": "NON_PARTICIPATING",
+    "221": "NOT_ALIVE",
+    "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
@@ -148,15 +148,15 @@
     "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
     "23": "NOT_ALIVE",
-    "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
+    "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
     "233": "NON_PARTICIPATING",
-    "234": "SUCCESS",
-    "235": "NON_PARTICIPATING",
-    "236": "NON_PARTICIPATING",
-    "237": "NON_PARTICIPATING",
-    "238": "SUCCESS",
+    "234": "NON_PARTICIPATING",
+    "235": "NOT_ALIVE",
+    "236": "NOT_ALIVE",
+    "237": "SUCCESS",
+    "238": "NON_PARTICIPATING",
     "239": "NON_PARTICIPATING",
     "24": "NON_PARTICIPATING",
     "240": "NON_PARTICIPATING",
@@ -165,97 +165,97 @@
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "246": "FAIL",
     "247": "NOT_ALIVE",
-    "248": "SUCCESS",
+    "248": "FAIL",
     "249": "NOT_ALIVE",
     "25": "NOT_ALIVE",
-    "250": "SUCCESS",
+    "250": "FAIL",
     "251": "NOT_ALIVE",
-    "252": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NON_PARTICIPATING",
     "255": "SUCCESS",
-    "26": "SUCCESS",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
-    "28": "NOT_ALIVE",
+    "28": "FAIL",
     "29": "NON_PARTICIPATING",
-    "3": "SUCCESS",
+    "3": "FAIL",
     "30": "NON_PARTICIPATING",
-    "31": "SUCCESS",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
-    "35": "SUCCESS",
+    "35": "NON_PARTICIPATING",
     "36": "NON_PARTICIPATING",
-    "37": "NON_PARTICIPATING",
-    "38": "NOT_ALIVE",
+    "37": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
     "39": "NON_PARTICIPATING",
-    "4": "SUCCESS",
-    "40": "NOT_ALIVE",
+    "4": "NON_PARTICIPATING",
+    "40": "NON_PARTICIPATING",
     "41": "NON_PARTICIPATING",
-    "42": "SUCCESS",
+    "42": "NON_PARTICIPATING",
     "43": "NOT_ALIVE",
-    "44": "NON_PARTICIPATING",
+    "44": "FAIL",
     "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
-    "47": "NON_PARTICIPATING",
+    "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
-    "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
-    "52": "NON_PARTICIPATING",
-    "53": "NON_PARTICIPATING",
+    "50": "FAIL",
+    "51": "NOT_ALIVE",
+    "52": "SUCCESS",
+    "53": "FAIL",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
     "57": "SUCCESS",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
-    "6": "NOT_ALIVE",
-    "60": "SUCCESS",
-    "61": "NON_PARTICIPATING",
-    "62": "NOT_ALIVE",
-    "63": "NON_PARTICIPATING",
-    "64": "NOT_ALIVE",
+    "6": "NON_PARTICIPATING",
+    "60": "FAIL",
+    "61": "FAIL",
+    "62": "FAIL",
+    "63": "FAIL",
+    "64": "FAIL",
     "65": "NON_PARTICIPATING",
-    "66": "NON_PARTICIPATING",
+    "66": "SUCCESS",
     "67": "NON_PARTICIPATING",
-    "68": "NOT_ALIVE",
+    "68": "NON_PARTICIPATING",
     "69": "NON_PARTICIPATING",
-    "7": "NON_PARTICIPATING",
+    "7": "SUCCESS",
     "70": "NON_PARTICIPATING",
     "71": "NON_PARTICIPATING",
     "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
-    "74": "SUCCESS",
+    "74": "NOT_ALIVE",
     "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
-    "8": "NON_PARTICIPATING",
-    "80": "SUCCESS",
+    "8": "SUCCESS",
+    "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
+    "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
-    "84": "SUCCESS",
-    "85": "NON_PARTICIPATING",
-    "86": "SUCCESS",
-    "87": "SUCCESS",
+    "84": "NON_PARTICIPATING",
+    "85": "NOT_ALIVE",
+    "86": "NON_PARTICIPATING",
+    "87": "NON_PARTICIPATING",
     "88": "NOT_ALIVE",
     "89": "SUCCESS",
-    "9": "NOT_ALIVE",
+    "9": "NON_PARTICIPATING",
     "90": "NOT_ALIVE",
     "91": "NON_PARTICIPATING",
-    "92": "NOT_ALIVE",
+    "92": "FAIL",
     "93": "NOT_ALIVE",
-    "94": "SUCCESS",
+    "94": "FAIL",
     "95": "NOT_ALIVE",
     "96": "SUCCESS",
     "97": "NOT_ALIVE",
-    "98": "SUCCESS",
+    "98": "NOT_ALIVE",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6504102,
+  "last_allreduce_block": 6507277,
   "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 sha256:467c4b422697d1c74c7849f55958c1802d858d63a79246937529065ab11224f4
+oid sha256:dcbe75f978db7d40799588d7c3babc68d3611846fe2c85b4b8b32bd00d8b01ba
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9595c6611f2ae796c194b79dca3c2e6c391dfa1e35a56ca8d7e80f45b11ff9f
+oid sha256:4f457abb5c7de16619e7ebee932e98c3f056f92c7574241b6da6b7dd58377265
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a12f9c31f6880d9d90ddc627ae0cf893ce57b4c52a1e08ac07d39fef16b9d132
+oid sha256:1996934b3113d01f3d8e9dbc32d93a1a8adfd564c914b14b7f1251b845a618b7
 size 4400252568