Run 9. Outer Step 7. Inner Step 0. Peers 1.

config.json

@@ -2,47 +2,47 @@
   "_name_or_path": "ytenszer/llamatrain4",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
+    "1": "FAIL",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "FAIL",
+    "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
-    "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
+    "105": "FAIL",
+    "106": "NON_PARTICIPATING",
+    "107": "FAIL",
     "108": "NOT_ALIVE",
-    "109": "NON_PARTICIPATING",
-    "11": "FAIL",
-    "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
+    "109": "FAIL",
+    "11": "NON_PARTICIPATING",
+    "110": "FAIL",
+    "111": "NOT_ALIVE",
     "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
-    "117": "NON_PARTICIPATING",
+    "117": "FAIL",
     "118": "NON_PARTICIPATING",
-    "119": "NOT_ALIVE",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "FAIL",
     "121": "NOT_ALIVE",
-    "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
-    "124": "NOT_ALIVE",
-    "125": "NOT_ALIVE",
-    "126": "NON_PARTICIPATING",
-    "127": "NOT_ALIVE",
+    "122": "FAIL",
+    "123": "FAIL",
+    "124": "NON_PARTICIPATING",
+    "125": "FAIL",
+    "126": "NOT_ALIVE",
+    "127": "NON_PARTICIPATING",
     "128": "NON_PARTICIPATING",
-    "129": "FAIL",
+    "129": "NOT_ALIVE",
     "13": "FAIL",
-    "130": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NON_PARTICIPATING",
+    "133": "NOT_ALIVE",
     "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "135": "FAIL",
     "136": "NOT_ALIVE",
     "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
@@ -50,213 +50,213 @@
     "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NON_PARTICIPATING",
-    "142": "NOT_ALIVE",
+    "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
-    "145": "FAIL",
-    "146": "NON_PARTICIPATING",
+    "145": "NON_PARTICIPATING",
+    "146": "NOT_ALIVE",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "FAIL",
+    "149": "NON_PARTICIPATING",
     "15": "FAIL",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
     "153": "NOT_ALIVE",
-    "154": "NOT_ALIVE",
-    "155": "FAIL",
-    "156": "FAIL",
+    "154": "NON_PARTICIPATING",
+    "155": "NON_PARTICIPATING",
+    "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
-    "158": "FAIL",
+    "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "FAIL",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "NON_PARTICIPATING",
+    "163": "FAIL",
     "164": "NON_PARTICIPATING",
-    "165": "FAIL",
+    "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
     "167": "FAIL",
     "168": "NON_PARTICIPATING",
-    "169": "FAIL",
+    "169": "NON_PARTICIPATING",
     "17": "NON_PARTICIPATING",
-    "170": "NON_PARTICIPATING",
+    "170": "FAIL",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NON_PARTICIPATING",
-    "175": "FAIL",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "FAIL",
-    "178": "NON_PARTICIPATING",
-    "179": "FAIL",
-    "18": "FAIL",
+    "178": "NOT_ALIVE",
+    "179": "NON_PARTICIPATING",
+    "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
-    "181": "FAIL",
-    "182": "NON_PARTICIPATING",
-    "183": "FAIL",
+    "181": "NON_PARTICIPATING",
+    "182": "FAIL",
+    "183": "NON_PARTICIPATING",
     "184": "FAIL",
     "185": "FAIL",
-    "186": "NON_PARTICIPATING",
+    "186": "FAIL",
     "187": "NOT_ALIVE",
-    "188": "FAIL",
-    "189": "FAIL",
-    "19": "FAIL",
-    "190": "NOT_ALIVE",
+    "188": "NOT_ALIVE",
+    "189": "NOT_ALIVE",
+    "19": "NON_PARTICIPATING",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NOT_ALIVE",
-    "194": "NOT_ALIVE",
-    "195": "FAIL",
-    "196": "FAIL",
-    "197": "FAIL",
-    "198": "NOT_ALIVE",
-    "199": "NOT_ALIVE",
-    "2": "FAIL",
-    "20": "NON_PARTICIPATING",
-    "200": "NON_PARTICIPATING",
-    "201": "FAIL",
+    "193": "NON_PARTICIPATING",
+    "194": "FAIL",
+    "195": "NON_PARTICIPATING",
+    "196": "NOT_ALIVE",
+    "197": "NOT_ALIVE",
+    "198": "FAIL",
+    "199": "NON_PARTICIPATING",
+    "2": "NON_PARTICIPATING",
+    "20": "FAIL",
+    "200": "NOT_ALIVE",
+    "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "FAIL",
     "204": "NOT_ALIVE",
     "205": "NON_PARTICIPATING",
-    "206": "NON_PARTICIPATING",
+    "206": "FAIL",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
+    "208": "NON_PARTICIPATING",
+    "209": "FAIL",
     "21": "NON_PARTICIPATING",
     "210": "NON_PARTICIPATING",
-    "211": "NON_PARTICIPATING",
-    "212": "NOT_ALIVE",
+    "211": "FAIL",
+    "212": "FAIL",
     "213": "NON_PARTICIPATING",
-    "214": "NON_PARTICIPATING",
-    "215": "NOT_ALIVE",
-    "216": "FAIL",
-    "217": "NON_PARTICIPATING",
-    "218": "NOT_ALIVE",
+    "214": "NOT_ALIVE",
+    "215": "NON_PARTICIPATING",
+    "216": "NON_PARTICIPATING",
+    "217": "NOT_ALIVE",
+    "218": "NON_PARTICIPATING",
     "219": "FAIL",
     "22": "NON_PARTICIPATING",
-    "220": "FAIL",
+    "220": "NON_PARTICIPATING",
     "221": "FAIL",
-    "222": "NOT_ALIVE",
-    "223": "NON_PARTICIPATING",
+    "222": "NON_PARTICIPATING",
+    "223": "FAIL",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
-    "226": "NON_PARTICIPATING",
+    "226": "NOT_ALIVE",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
-    "23": "NOT_ALIVE",
+    "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
+    "231": "FAIL",
     "232": "NON_PARTICIPATING",
-    "233": "NOT_ALIVE",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
-    "237": "FAIL",
+    "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
     "239": "FAIL",
     "24": "FAIL",
-    "240": "NON_PARTICIPATING",
+    "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
-    "243": "FAIL",
+    "243": "NON_PARTICIPATING",
     "244": "NOT_ALIVE",
-    "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
-    "248": "FAIL",
+    "245": "NOT_ALIVE",
+    "246": "NOT_ALIVE",
+    "247": "FAIL",
+    "248": "NOT_ALIVE",
     "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
-    "251": "NON_PARTICIPATING",
-    "252": "NON_PARTICIPATING",
+    "250": "FAIL",
+    "251": "NOT_ALIVE",
+    "252": "NOT_ALIVE",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "FAIL",
-    "27": "FAIL",
+    "26": "NOT_ALIVE",
+    "27": "NON_PARTICIPATING",
     "28": "FAIL",
     "29": "NON_PARTICIPATING",
-    "3": "FAIL",
-    "30": "FAIL",
-    "31": "NON_PARTICIPATING",
+    "3": "NON_PARTICIPATING",
+    "30": "NON_PARTICIPATING",
+    "31": "FAIL",
     "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "FAIL",
     "35": "NON_PARTICIPATING",
     "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "NOT_ALIVE",
+    "38": "FAIL",
     "39": "NON_PARTICIPATING",
-    "4": "FAIL",
-    "40": "NON_PARTICIPATING",
+    "4": "NON_PARTICIPATING",
+    "40": "FAIL",
     "41": "NOT_ALIVE",
-    "42": "NON_PARTICIPATING",
+    "42": "FAIL",
     "43": "FAIL",
     "44": "NON_PARTICIPATING",
-    "45": "FAIL",
+    "45": "NON_PARTICIPATING",
     "46": "NOT_ALIVE",
-    "47": "NOT_ALIVE",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
     "5": "NON_PARTICIPATING",
-    "50": "FAIL",
-    "51": "NON_PARTICIPATING",
+    "50": "NON_PARTICIPATING",
+    "51": "FAIL",
     "52": "NON_PARTICIPATING",
-    "53": "NOT_ALIVE",
+    "53": "FAIL",
     "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
-    "57": "FAIL",
+    "57": "NON_PARTICIPATING",
     "58": "NON_PARTICIPATING",
-    "59": "NON_PARTICIPATING",
-    "6": "NON_PARTICIPATING",
+    "59": "NOT_ALIVE",
+    "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
-    "62": "NON_PARTICIPATING",
+    "61": "NOT_ALIVE",
+    "62": "FAIL",
     "63": "NON_PARTICIPATING",
     "64": "NON_PARTICIPATING",
-    "65": "FAIL",
+    "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
-    "68": "NON_PARTICIPATING",
-    "69": "NON_PARTICIPATING",
-    "7": "FAIL",
-    "70": "NOT_ALIVE",
+    "68": "FAIL",
+    "69": "FAIL",
+    "7": "NON_PARTICIPATING",
+    "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
-    "72": "FAIL",
+    "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
     "74": "FAIL",
     "75": "NON_PARTICIPATING",
-    "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
+    "76": "FAIL",
+    "77": "NON_PARTICIPATING",
     "78": "FAIL",
     "79": "NON_PARTICIPATING",
-    "8": "NOT_ALIVE",
-    "80": "NON_PARTICIPATING",
+    "8": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
+    "86": "NOT_ALIVE",
     "87": "NON_PARTICIPATING",
-    "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
+    "89": "NON_PARTICIPATING",
     "9": "NON_PARTICIPATING",
-    "90": "FAIL",
+    "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
-    "92": "NON_PARTICIPATING",
+    "92": "FAIL",
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
     "95": "NON_PARTICIPATING",
-    "96": "NON_PARTICIPATING",
-    "97": "NOT_ALIVE",
-    "98": "NON_PARTICIPATING",
-    "99": "NON_PARTICIPATING"
+    "96": "NOT_ALIVE",
+    "97": "FAIL",
+    "98": "FAIL",
+    "99": "FAIL"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6451623,
+  "last_allreduce_block": 6452261,
   "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:f3fdbbb8e2daf2666b60350c2e0fb47141f706476b81ca81bf0cb3adcbfc8a87
+oid sha256:a35476e45535d9ad8c55d737c4dbcdec389160c74129099eb630daf553a28b58
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca134aca5f396b67ecd463d6757219768cdee3487592aa069076cfbb5295ef02
+oid sha256:204b80822e06005479f64f9aed389d26a29b4e84d4921cdd015819670593a9c7
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3904743ef97a0d27bdde45ab8b93df655340a45516504d8277583d56f4a1a1d0
+oid sha256:c6509903451dbafd12b16925bb0cda3faa3dbcd037435aa582545ba06e49002a
 size 4400252568