Run 10. Outer Step 28. Inner Step 0. Peers 39.

config.json

@@ -2,69 +2,69 @@
   "_name_or_path": "strnan01/llama-1b-hot-6",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "SUCCESS",
+    "1": "FAIL",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "FAIL",
-    "103": "NON_PARTICIPATING",
+    "102": "SUCCESS",
+    "103": "SUCCESS",
     "104": "NOT_ALIVE",
-    "105": "FAIL",
-    "106": "NON_PARTICIPATING",
+    "105": "NOT_ALIVE",
+    "106": "NOT_ALIVE",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "FAIL",
-    "11": "FAIL",
+    "109": "NOT_ALIVE",
+    "11": "SUCCESS",
     "110": "NOT_ALIVE",
-    "111": "NON_PARTICIPATING",
-    "112": "FAIL",
+    "111": "SUCCESS",
+    "112": "NON_PARTICIPATING",
     "113": "SUCCESS",
-    "114": "NON_PARTICIPATING",
+    "114": "SUCCESS",
     "115": "NON_PARTICIPATING",
-    "116": "FAIL",
+    "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
+    "120": "NOT_ALIVE",
     "121": "NON_PARTICIPATING",
-    "122": "NON_PARTICIPATING",
-    "123": "FAIL",
+    "122": "NOT_ALIVE",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
-    "127": "NON_PARTICIPATING",
-    "128": "NOT_ALIVE",
-    "129": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
     "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NON_PARTICIPATING",
-    "134": "NON_PARTICIPATING",
+    "133": "NOT_ALIVE",
+    "134": "NOT_ALIVE",
     "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
-    "14": "NON_PARTICIPATING",
+    "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "SUCCESS",
-    "143": "NOT_ALIVE",
+    "142": "FAIL",
+    "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
-    "146": "FAIL",
-    "147": "NOT_ALIVE",
+    "146": "SUCCESS",
+    "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
-    "15": "FAIL",
-    "150": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
+    "15": "NOT_ALIVE",
+    "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
-    "152": "NON_PARTICIPATING",
-    "153": "FAIL",
+    "152": "NOT_ALIVE",
+    "153": "SUCCESS",
     "154": "SUCCESS",
-    "155": "NON_PARTICIPATING",
+    "155": "NOT_ALIVE",
     "156": "FAIL",
     "157": "NON_PARTICIPATING",
     "158": "NOT_ALIVE",
@@ -72,62 +72,62 @@
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "NOT_ALIVE",
     "163": "SUCCESS",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "FAIL",
+    "169": "SUCCESS",
     "17": "NON_PARTICIPATING",
-    "170": "SUCCESS",
-    "171": "FAIL",
+    "170": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
-    "174": "NON_PARTICIPATING",
-    "175": "NON_PARTICIPATING",
+    "174": "NOT_ALIVE",
+    "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
-    "18": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
+    "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
-    "184": "FAIL",
-    "185": "SUCCESS",
-    "186": "NOT_ALIVE",
+    "184": "SUCCESS",
+    "185": "FAIL",
+    "186": "FAIL",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NOT_ALIVE",
-    "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
-    "196": "SUCCESS",
+    "193": "NON_PARTICIPATING",
+    "194": "NON_PARTICIPATING",
+    "195": "NOT_ALIVE",
+    "196": "FAIL",
     "197": "NOT_ALIVE",
     "198": "SUCCESS",
     "199": "NON_PARTICIPATING",
-    "2": "FAIL",
-    "20": "NOT_ALIVE",
-    "200": "FAIL",
-    "201": "NON_PARTICIPATING",
+    "2": "NOT_ALIVE",
+    "20": "SUCCESS",
+    "200": "SUCCESS",
+    "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "SUCCESS",
-    "206": "NOT_ALIVE",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
     "21": "FAIL",
     "210": "NON_PARTICIPATING",
-    "211": "NOT_ALIVE",
+    "211": "SUCCESS",
     "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
@@ -135,128 +135,128 @@
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
-    "219": "FAIL",
-    "22": "NON_PARTICIPATING",
+    "219": "SUCCESS",
+    "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "SUCCESS",
-    "222": "FAIL",
+    "221": "FAIL",
+    "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
-    "227": "NOT_ALIVE",
+    "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "FAIL",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
-    "232": "NON_PARTICIPATING",
-    "233": "FAIL",
-    "234": "SUCCESS",
-    "235": "SUCCESS",
-    "236": "NON_PARTICIPATING",
-    "237": "NON_PARTICIPATING",
-    "238": "SUCCESS",
+    "232": "NOT_ALIVE",
+    "233": "NON_PARTICIPATING",
+    "234": "FAIL",
+    "235": "FAIL",
+    "236": "NOT_ALIVE",
+    "237": "FAIL",
+    "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
-    "24": "NON_PARTICIPATING",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
+    "241": "FAIL",
+    "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
-    "248": "NON_PARTICIPATING",
+    "248": "NOT_ALIVE",
     "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "SUCCESS",
+    "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
-    "252": "FAIL",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
-    "254": "NON_PARTICIPATING",
+    "254": "SUCCESS",
     "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
-    "29": "FAIL",
-    "3": "SUCCESS",
-    "30": "NON_PARTICIPATING",
-    "31": "NOT_ALIVE",
+    "29": "SUCCESS",
+    "3": "NOT_ALIVE",
+    "30": "NOT_ALIVE",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "FAIL",
-    "35": "FAIL",
-    "36": "FAIL",
+    "34": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
+    "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
-    "38": "SUCCESS",
+    "38": "FAIL",
     "39": "FAIL",
-    "4": "SUCCESS",
-    "40": "FAIL",
+    "4": "FAIL",
+    "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "SUCCESS",
     "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
-    "47": "NON_PARTICIPATING",
+    "46": "FAIL",
+    "47": "FAIL",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
-    "5": "NON_PARTICIPATING",
+    "5": "SUCCESS",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
-    "53": "SUCCESS",
-    "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
+    "53": "FAIL",
+    "54": "NOT_ALIVE",
+    "55": "NOT_ALIVE",
     "56": "NON_PARTICIPATING",
-    "57": "NON_PARTICIPATING",
-    "58": "NON_PARTICIPATING",
+    "57": "SUCCESS",
+    "58": "NOT_ALIVE",
     "59": "NON_PARTICIPATING",
     "6": "NOT_ALIVE",
     "60": "NOT_ALIVE",
     "61": "NOT_ALIVE",
-    "62": "NOT_ALIVE",
+    "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
-    "64": "NON_PARTICIPATING",
+    "64": "SUCCESS",
     "65": "NOT_ALIVE",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
     "69": "SUCCESS",
-    "7": "SUCCESS",
+    "7": "FAIL",
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
-    "73": "NON_PARTICIPATING",
-    "74": "SUCCESS",
-    "75": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
+    "74": "FAIL",
+    "75": "NOT_ALIVE",
     "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
-    "78": "NON_PARTICIPATING",
-    "79": "NON_PARTICIPATING",
-    "8": "FAIL",
+    "78": "NOT_ALIVE",
+    "79": "NOT_ALIVE",
+    "8": "SUCCESS",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
-    "83": "NON_PARTICIPATING",
+    "83": "NOT_ALIVE",
     "84": "SUCCESS",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
+    "86": "NOT_ALIVE",
     "87": "NON_PARTICIPATING",
-    "88": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
-    "9": "SUCCESS",
-    "90": "FAIL",
+    "9": "FAIL",
+    "90": "NON_PARTICIPATING",
     "91": "FAIL",
-    "92": "NON_PARTICIPATING",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "SUCCESS",
     "95": "NOT_ALIVE",
     "96": "SUCCESS",
-    "97": "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": 6481823,
+  "last_allreduce_block": 6482230,
   "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 sha256:335092b081cd5b414cb4283eea1f3ffd0e6b8b0bdc814effb3f878786b992715
+oid sha256:b84458a817f8a97587d0ce1121c8a602faed9aa1b6d0a77c781d1acedeb47252
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ed97d9d8675128411cd40e201870f36a071aede4e395bec9443848c37da56c8
+oid sha256:0cc5e6ebfa3765109127e5ef6e29ff7bebb1c6cbc1fa4a4993392f5f923ef700
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e118c7ced212753926da6f62837847330877595908e8e88f8c8981b4417a22f2
+oid sha256:05f93b18456639d1c16208f1af4dcb063c34294be5ea99e369391f3adc5abd2e
 size 4400252568