Run 9. Outer Step 14. Inner Step 0. Peers 1.

config.json

@@ -2,261 +2,261 @@
   "_name_or_path": "r2911/d_p6",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
-    "10": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
+    "10": "FAIL",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
-    "106": "FAIL",
-    "107": "NON_PARTICIPATING",
+    "104": "FAIL",
+    "105": "FAIL",
+    "106": "NOT_ALIVE",
+    "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
     "109": "FAIL",
-    "11": "FAIL",
-    "110": "FAIL",
+    "11": "NON_PARTICIPATING",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "FAIL",
-    "113": "FAIL",
+    "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "FAIL",
-    "118": "FAIL",
+    "117": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
+    "120": "FAIL",
     "121": "NOT_ALIVE",
-    "122": "NON_PARTICIPATING",
+    "122": "FAIL",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
-    "126": "FAIL",
-    "127": "NON_PARTICIPATING",
-    "128": "FAIL",
+    "126": "NOT_ALIVE",
+    "127": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "FAIL",
     "130": "FAIL",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NOT_ALIVE",
+    "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "135": "FAIL",
     "136": "NOT_ALIVE",
-    "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
+    "137": "NON_PARTICIPATING",
+    "138": "FAIL",
     "139": "NOT_ALIVE",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "FAIL",
-    "143": "FAIL",
+    "142": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
-    "147": "FAIL",
+    "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
-    "15": "NOT_ALIVE",
+    "149": "NOT_ALIVE",
+    "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
     "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
-    "155": "FAIL",
+    "154": "FAIL",
+    "155": "NON_PARTICIPATING",
     "156": "FAIL",
-    "157": "NON_PARTICIPATING",
-    "158": "FAIL",
+    "157": "NOT_ALIVE",
+    "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "FAIL",
-    "164": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
-    "166": "FAIL",
-    "167": "NON_PARTICIPATING",
+    "166": "NON_PARTICIPATING",
+    "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
     "169": "FAIL",
-    "17": "NOT_ALIVE",
-    "170": "FAIL",
+    "17": "NON_PARTICIPATING",
+    "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
-    "174": "NOT_ALIVE",
+    "173": "FAIL",
+    "174": "NON_PARTICIPATING",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "FAIL",
     "178": "NOT_ALIVE",
-    "179": "FAIL",
+    "179": "NON_PARTICIPATING",
     "18": "NON_PARTICIPATING",
-    "180": "NON_PARTICIPATING",
+    "180": "FAIL",
     "181": "NON_PARTICIPATING",
-    "182": "NON_PARTICIPATING",
-    "183": "FAIL",
-    "184": "NOT_ALIVE",
-    "185": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
+    "183": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
+    "185": "NOT_ALIVE",
     "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
-    "189": "NON_PARTICIPATING",
+    "189": "FAIL",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
     "191": "NON_PARTICIPATING",
     "192": "NON_PARTICIPATING",
-    "193": "FAIL",
+    "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
-    "195": "NOT_ALIVE",
-    "196": "FAIL",
-    "197": "NON_PARTICIPATING",
-    "198": "NON_PARTICIPATING",
-    "199": "NOT_ALIVE",
+    "195": "FAIL",
+    "196": "NON_PARTICIPATING",
+    "197": "NOT_ALIVE",
+    "198": "NOT_ALIVE",
+    "199": "FAIL",
     "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
     "200": "NON_PARTICIPATING",
     "201": "FAIL",
     "202": "FAIL",
-    "203": "FAIL",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "FAIL",
+    "205": "NOT_ALIVE",
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "FAIL",
     "209": "NON_PARTICIPATING",
     "21": "FAIL",
-    "210": "NON_PARTICIPATING",
-    "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
+    "210": "NOT_ALIVE",
+    "211": "FAIL",
+    "212": "FAIL",
     "213": "NOT_ALIVE",
-    "214": "FAIL",
-    "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
+    "214": "NON_PARTICIPATING",
+    "215": "FAIL",
+    "216": "FAIL",
     "217": "NON_PARTICIPATING",
-    "218": "FAIL",
-    "219": "FAIL",
+    "218": "NON_PARTICIPATING",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "FAIL",
+    "220": "NON_PARTICIPATING",
     "221": "NON_PARTICIPATING",
     "222": "NON_PARTICIPATING",
-    "223": "FAIL",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "FAIL",
     "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "227": "FAIL",
     "228": "NON_PARTICIPATING",
-    "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "229": "NON_PARTICIPATING",
+    "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
     "233": "NON_PARTICIPATING",
-    "234": "NOT_ALIVE",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "FAIL",
+    "236": "NON_PARTICIPATING",
     "237": "FAIL",
-    "238": "NON_PARTICIPATING",
-    "239": "FAIL",
+    "238": "FAIL",
+    "239": "NON_PARTICIPATING",
     "24": "FAIL",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
-    "242": "NOT_ALIVE",
-    "243": "FAIL",
+    "242": "NON_PARTICIPATING",
+    "243": "NON_PARTICIPATING",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
-    "247": "NOT_ALIVE",
+    "247": "FAIL",
     "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
-    "254": "NON_PARTICIPATING",
-    "255": "NON_PARTICIPATING",
-    "26": "NOT_ALIVE",
-    "27": "NOT_ALIVE",
+    "254": "NOT_ALIVE",
+    "255": "FAIL",
+    "26": "NON_PARTICIPATING",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
-    "29": "NON_PARTICIPATING",
-    "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
+    "29": "FAIL",
+    "3": "NOT_ALIVE",
+    "30": "FAIL",
     "31": "FAIL",
-    "32": "NOT_ALIVE",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
-    "35": "FAIL",
-    "36": "FAIL",
+    "35": "NON_PARTICIPATING",
+    "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
     "38": "FAIL",
     "39": "NON_PARTICIPATING",
     "4": "NON_PARTICIPATING",
-    "40": "NON_PARTICIPATING",
+    "40": "FAIL",
     "41": "NOT_ALIVE",
-    "42": "NOT_ALIVE",
+    "42": "NON_PARTICIPATING",
     "43": "NOT_ALIVE",
-    "44": "NOT_ALIVE",
+    "44": "NON_PARTICIPATING",
     "45": "NON_PARTICIPATING",
-    "46": "NON_PARTICIPATING",
+    "46": "NOT_ALIVE",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
-    "5": "NOT_ALIVE",
+    "5": "FAIL",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
-    "52": "FAIL",
+    "52": "NON_PARTICIPATING",
     "53": "FAIL",
-    "54": "NOT_ALIVE",
+    "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "FAIL",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
-    "6": "NOT_ALIVE",
-    "60": "FAIL",
-    "61": "NON_PARTICIPATING",
-    "62": "NON_PARTICIPATING",
+    "6": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
+    "61": "FAIL",
+    "62": "FAIL",
     "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
-    "66": "NOT_ALIVE",
-    "67": "NOT_ALIVE",
+    "65": "FAIL",
+    "66": "FAIL",
+    "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "NOT_ALIVE",
+    "69": "FAIL",
     "7": "FAIL",
-    "70": "NOT_ALIVE",
+    "70": "FAIL",
     "71": "NOT_ALIVE",
-    "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
-    "74": "NOT_ALIVE",
+    "72": "FAIL",
+    "73": "NOT_ALIVE",
+    "74": "FAIL",
     "75": "NON_PARTICIPATING",
     "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
+    "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
-    "79": "NOT_ALIVE",
-    "8": "NON_PARTICIPATING",
-    "80": "FAIL",
+    "79": "NON_PARTICIPATING",
+    "8": "NOT_ALIVE",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
-    "84": "FAIL",
+    "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
+    "86": "NON_PARTICIPATING",
     "87": "NON_PARTICIPATING",
     "88": "FAIL",
-    "89": "FAIL",
-    "9": "NON_PARTICIPATING",
-    "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
-    "92": "NOT_ALIVE",
+    "89": "NOT_ALIVE",
+    "9": "FAIL",
+    "90": "FAIL",
+    "91": "FAIL",
+    "92": "FAIL",
     "93": "NOT_ALIVE",
-    "94": "FAIL",
-    "95": "NOT_ALIVE",
-    "96": "FAIL",
+    "94": "NON_PARTICIPATING",
+    "95": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
     "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
-    "99": "FAIL"
+    "98": "NOT_ALIVE",
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6457631,
+  "last_allreduce_block": 6458233,
   "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 sha256:61d5f900e5bbb1b2348c5cbb78ae25ec8ef03d5b5ad7e287394ab2b746d834d4
+oid sha256:e6b72c70b16f7429b24356d852694bc119756862e53b56cca568c5681f570713
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a5820c6a762abd773ee55c4426e3e95ab0d7a64fb8d1cd5a1afdde5952ea3bf8
+oid sha256:f39133072bfccb379dc05ceb9d6dd2b596641ea1d5c501450195a3a132193390
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a62da5e79f711cfd30eeb7bc68ff61b58201d7b28c5610150d18ba787301cb17
+oid sha256:917021c7c03b8a9d44f4ec051a29259339d571a06e02e5292a3f543871c5db2b
 size 4400252568