Run 9. Outer Step 17. Inner Step 0. Peers 1.

config.json

@@ -1,262 +1,262 @@
 {
-  "_name_or_path": "Ahad5625/M1",
+  "_name_or_path": "r2911/d_p6",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "FAIL",
-    "10": "NON_PARTICIPATING",
+    "1": "NON_PARTICIPATING",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "FAIL",
-    "104": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
     "105": "FAIL",
-    "106": "FAIL",
+    "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "NON_PARTICIPATING",
+    "109": "FAIL",
     "11": "NON_PARTICIPATING",
     "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
-    "112": "NOT_ALIVE",
+    "112": "FAIL",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
-    "116": "FAIL",
-    "117": "FAIL",
-    "118": "NOT_ALIVE",
-    "119": "NOT_ALIVE",
+    "116": "NON_PARTICIPATING",
+    "117": "NON_PARTICIPATING",
+    "118": "NON_PARTICIPATING",
+    "119": "FAIL",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
-    "124": "FAIL",
-    "125": "NOT_ALIVE",
-    "126": "NON_PARTICIPATING",
-    "127": "FAIL",
+    "123": "FAIL",
+    "124": "NOT_ALIVE",
+    "125": "NON_PARTICIPATING",
+    "126": "FAIL",
+    "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
     "129": "NOT_ALIVE",
-    "13": "FAIL",
+    "13": "NON_PARTICIPATING",
     "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NOT_ALIVE",
-    "134": "NON_PARTICIPATING",
-    "135": "FAIL",
+    "133": "NON_PARTICIPATING",
+    "134": "FAIL",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NOT_ALIVE",
-    "14": "NON_PARTICIPATING",
+    "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
-    "141": "NOT_ALIVE",
-    "142": "NOT_ALIVE",
-    "143": "FAIL",
+    "141": "NON_PARTICIPATING",
+    "142": "NON_PARTICIPATING",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NOT_ALIVE",
-    "147": "NOT_ALIVE",
-    "148": "NON_PARTICIPATING",
+    "146": "FAIL",
+    "147": "NON_PARTICIPATING",
+    "148": "NOT_ALIVE",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
-    "150": "FAIL",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "FAIL",
     "154": "NON_PARTICIPATING",
     "155": "NON_PARTICIPATING",
-    "156": "FAIL",
-    "157": "NON_PARTICIPATING",
-    "158": "FAIL",
-    "159": "NOT_ALIVE",
-    "16": "NOT_ALIVE",
+    "156": "NON_PARTICIPATING",
+    "157": "FAIL",
+    "158": "NON_PARTICIPATING",
+    "159": "NON_PARTICIPATING",
+    "16": "NON_PARTICIPATING",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "NOT_ALIVE",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
+    "166": "FAIL",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
     "169": "NOT_ALIVE",
-    "17": "NOT_ALIVE",
-    "170": "NOT_ALIVE",
+    "17": "NON_PARTICIPATING",
+    "170": "FAIL",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "FAIL",
-    "174": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
+    "174": "NON_PARTICIPATING",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "FAIL",
+    "177": "NON_PARTICIPATING",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
-    "18": "NON_PARTICIPATING",
+    "18": "FAIL",
     "180": "NON_PARTICIPATING",
     "181": "NON_PARTICIPATING",
-    "182": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "FAIL",
     "186": "NON_PARTICIPATING",
     "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
-    "189": "NOT_ALIVE",
-    "19": "NON_PARTICIPATING",
-    "190": "NOT_ALIVE",
-    "191": "NON_PARTICIPATING",
+    "189": "NON_PARTICIPATING",
+    "19": "FAIL",
+    "190": "NON_PARTICIPATING",
+    "191": "FAIL",
     "192": "NOT_ALIVE",
-    "193": "FAIL",
-    "194": "NOT_ALIVE",
+    "193": "NON_PARTICIPATING",
+    "194": "NON_PARTICIPATING",
     "195": "FAIL",
     "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
-    "198": "NOT_ALIVE",
+    "198": "FAIL",
     "199": "NON_PARTICIPATING",
-    "2": "NON_PARTICIPATING",
+    "2": "FAIL",
     "20": "FAIL",
     "200": "NON_PARTICIPATING",
     "201": "NON_PARTICIPATING",
-    "202": "FAIL",
+    "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
-    "205": "FAIL",
-    "206": "NON_PARTICIPATING",
+    "205": "NON_PARTICIPATING",
+    "206": "FAIL",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "FAIL",
     "21": "NON_PARTICIPATING",
     "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
+    "212": "FAIL",
     "213": "NOT_ALIVE",
-    "214": "NON_PARTICIPATING",
-    "215": "NOT_ALIVE",
-    "216": "FAIL",
+    "214": "FAIL",
+    "215": "FAIL",
+    "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
-    "218": "FAIL",
+    "218": "NON_PARTICIPATING",
     "219": "FAIL",
     "22": "NOT_ALIVE",
-    "220": "FAIL",
-    "221": "FAIL",
+    "220": "NOT_ALIVE",
+    "221": "NON_PARTICIPATING",
     "222": "NON_PARTICIPATING",
     "223": "FAIL",
     "224": "NOT_ALIVE",
-    "225": "FAIL",
-    "226": "NON_PARTICIPATING",
-    "227": "FAIL",
-    "228": "NOT_ALIVE",
+    "225": "NON_PARTICIPATING",
+    "226": "FAIL",
+    "227": "NON_PARTICIPATING",
+    "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
-    "23": "FAIL",
+    "23": "NOT_ALIVE",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
-    "232": "NON_PARTICIPATING",
-    "233": "NOT_ALIVE",
+    "232": "FAIL",
+    "233": "FAIL",
     "234": "NON_PARTICIPATING",
     "235": "FAIL",
     "236": "FAIL",
     "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
-    "24": "NOT_ALIVE",
+    "238": "FAIL",
+    "239": "FAIL",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
-    "246": "FAIL",
+    "245": "NON_PARTICIPATING",
+    "246": "NON_PARTICIPATING",
     "247": "FAIL",
     "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
+    "249": "FAIL",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "FAIL",
     "251": "NOT_ALIVE",
-    "252": "NON_PARTICIPATING",
+    "252": "FAIL",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
-    "255": "FAIL",
-    "26": "NON_PARTICIPATING",
-    "27": "NOT_ALIVE",
-    "28": "NOT_ALIVE",
-    "29": "NON_PARTICIPATING",
+    "255": "NON_PARTICIPATING",
+    "26": "FAIL",
+    "27": "NON_PARTICIPATING",
+    "28": "FAIL",
+    "29": "FAIL",
     "3": "NON_PARTICIPATING",
     "30": "FAIL",
-    "31": "NOT_ALIVE",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
-    "35": "NON_PARTICIPATING",
-    "36": "NON_PARTICIPATING",
+    "35": "FAIL",
+    "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "FAIL",
-    "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
-    "40": "NON_PARTICIPATING",
+    "38": "NON_PARTICIPATING",
+    "39": "FAIL",
+    "4": "FAIL",
+    "40": "FAIL",
     "41": "NOT_ALIVE",
-    "42": "NOT_ALIVE",
+    "42": "NON_PARTICIPATING",
     "43": "NON_PARTICIPATING",
-    "44": "NON_PARTICIPATING",
-    "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
+    "44": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
+    "46": "NOT_ALIVE",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
-    "5": "NON_PARTICIPATING",
+    "5": "NOT_ALIVE",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
     "53": "FAIL",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "FAIL",
-    "57": "FAIL",
+    "56": "NOT_ALIVE",
+    "57": "NON_PARTICIPATING",
     "58": "NON_PARTICIPATING",
     "59": "NOT_ALIVE",
-    "6": "NON_PARTICIPATING",
-    "60": "NOT_ALIVE",
-    "61": "NON_PARTICIPATING",
+    "6": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
+    "61": "NOT_ALIVE",
     "62": "NON_PARTICIPATING",
-    "63": "FAIL",
+    "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
-    "65": "FAIL",
-    "66": "FAIL",
+    "65": "NON_PARTICIPATING",
+    "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
-    "68": "NOT_ALIVE",
-    "69": "FAIL",
-    "7": "NOT_ALIVE",
-    "70": "FAIL",
+    "68": "NON_PARTICIPATING",
+    "69": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
+    "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
     "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
-    "74": "NOT_ALIVE",
+    "74": "FAIL",
     "75": "NOT_ALIVE",
-    "76": "NOT_ALIVE",
+    "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
-    "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "8": "NOT_ALIVE",
+    "80": "FAIL",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
+    "83": "NOT_ALIVE",
     "84": "FAIL",
-    "85": "FAIL",
-    "86": "FAIL",
+    "85": "NOT_ALIVE",
+    "86": "NOT_ALIVE",
     "87": "NON_PARTICIPATING",
-    "88": "FAIL",
-    "89": "FAIL",
-    "9": "FAIL",
+    "88": "NON_PARTICIPATING",
+    "89": "NON_PARTICIPATING",
+    "9": "NON_PARTICIPATING",
     "90": "FAIL",
-    "91": "NOT_ALIVE",
-    "92": "FAIL",
+    "91": "FAIL",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
     "94": "FAIL",
     "95": "NOT_ALIVE",
-    "96": "FAIL",
-    "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
-    "99": "NON_PARTICIPATING"
+    "96": "NON_PARTICIPATING",
+    "97": "NOT_ALIVE",
+    "98": "NOT_ALIVE",
+    "99": "FAIL"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6459469,
+  "last_allreduce_block": 6460132,
   "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 sha256:8c7148e3045a45109dbfc20f679e85b41a4c9d679b676065c7cb1045d60d9ab9
+oid sha256:f5c9d41291e045fc27432b61052f7b1f70beaa4126e4a0bd062f20e19839dace
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b50b5802d233ee8d0e14d72252327387f572150ca695b99372956821a5b06b66
+oid sha256:e932d1ef5818c1e56a3fac4f79a6527191dc4c5e4f699c6500b5f7d3691e27ad
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1612a825b7edaceaf87f618f4583ac3755316e5d46faeb7e0d9939b0e7d7340
+oid sha256:ba208dada0df776e2b9efebf1ebf629e2875dc3b2abf25af6272ffedda9a4359
 size 4400252568