Run 10. Outer Step 113. Inner Step 0. Peers 60.

config.json

@@ -5,37 +5,37 @@
     "1": "NON_PARTICIPATING",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "SUCCESS",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
     "105": "SUCCESS",
     "106": "NON_PARTICIPATING",
-    "107": "FAIL",
+    "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
-    "109": "NOT_ALIVE",
+    "109": "NON_PARTICIPATING",
     "11": "SUCCESS",
     "110": "NOT_ALIVE",
-    "111": "NOT_ALIVE",
+    "111": "SUCCESS",
     "112": "SUCCESS",
-    "113": "NON_PARTICIPATING",
+    "113": "FAIL",
     "114": "NON_PARTICIPATING",
-    "115": "SUCCESS",
+    "115": "NOT_ALIVE",
     "116": "SUCCESS",
     "117": "SUCCESS",
     "118": "SUCCESS",
     "119": "SUCCESS",
-    "12": "NON_PARTICIPATING",
+    "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "SUCCESS",
-    "122": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
     "123": "SUCCESS",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
-    "126": "SUCCESS",
+    "126": "NOT_ALIVE",
     "127": "NON_PARTICIPATING",
-    "128": "NOT_ALIVE",
-    "129": "SUCCESS",
+    "128": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
     "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
@@ -43,20 +43,20 @@
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NON_PARTICIPATING",
-    "136": "SUCCESS",
+    "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
+    "139": "SUCCESS",
     "14": "NOT_ALIVE",
-    "140": "SUCCESS",
-    "141": "NON_PARTICIPATING",
+    "140": "NON_PARTICIPATING",
+    "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "NON_PARTICIPATING",
+    "148": "NOT_ALIVE",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
@@ -68,21 +68,21 @@
     "156": "NON_PARTICIPATING",
     "157": "SUCCESS",
     "158": "NOT_ALIVE",
-    "159": "NON_PARTICIPATING",
+    "159": "SUCCESS",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
     "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
-    "163": "SUCCESS",
+    "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "SUCCESS",
+    "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
-    "168": "NON_PARTICIPATING",
+    "168": "SUCCESS",
     "169": "NOT_ALIVE",
-    "17": "NON_PARTICIPATING",
+    "17": "NOT_ALIVE",
     "170": "SUCCESS",
-    "171": "NON_PARTICIPATING",
+    "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NOT_ALIVE",
     "174": "NOT_ALIVE",
@@ -94,19 +94,19 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
-    "185": "SUCCESS",
-    "186": "NOT_ALIVE",
+    "185": "NOT_ALIVE",
+    "186": "SUCCESS",
     "187": "SUCCESS",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "190": "SUCCESS",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "SUCCESS",
+    "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
@@ -115,7 +115,7 @@
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NON_PARTICIPATING",
-    "200": "NOT_ALIVE",
+    "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
@@ -128,36 +128,36 @@
     "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
     "211": "SUCCESS",
-    "212": "NON_PARTICIPATING",
+    "212": "SUCCESS",
     "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
-    "217": "FAIL",
+    "217": "NOT_ALIVE",
     "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "SUCCESS",
     "222": "NOT_ALIVE",
-    "223": "NON_PARTICIPATING",
+    "223": "SUCCESS",
     "224": "NOT_ALIVE",
-    "225": "NOT_ALIVE",
-    "226": "SUCCESS",
+    "225": "NON_PARTICIPATING",
+    "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
     "23": "SUCCESS",
     "230": "NON_PARTICIPATING",
-    "231": "SUCCESS",
+    "231": "NOT_ALIVE",
     "232": "SUCCESS",
     "233": "NOT_ALIVE",
-    "234": "NOT_ALIVE",
+    "234": "NON_PARTICIPATING",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "NOT_ALIVE",
     "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
+    "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
     "241": "NOT_ALIVE",
@@ -165,20 +165,20 @@
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "NOT_ALIVE",
     "248": "SUCCESS",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "SUCCESS",
     "251": "NOT_ALIVE",
-    "252": "NOT_ALIVE",
+    "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
-    "254": "NON_PARTICIPATING",
+    "254": "NOT_ALIVE",
     "255": "SUCCESS",
     "26": "SUCCESS",
     "27": "SUCCESS",
-    "28": "SUCCESS",
+    "28": "NOT_ALIVE",
     "29": "NON_PARTICIPATING",
     "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
@@ -187,49 +187,49 @@
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "SUCCESS",
-    "36": "NOT_ALIVE",
+    "36": "NON_PARTICIPATING",
     "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
+    "40": "SUCCESS",
     "41": "NOT_ALIVE",
-    "42": "NON_PARTICIPATING",
+    "42": "SUCCESS",
     "43": "NOT_ALIVE",
-    "44": "NON_PARTICIPATING",
-    "45": "NOT_ALIVE",
-    "46": "NOT_ALIVE",
-    "47": "NON_PARTICIPATING",
+    "44": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
+    "46": "SUCCESS",
+    "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NON_PARTICIPATING",
     "51": "NOT_ALIVE",
-    "52": "NON_PARTICIPATING",
-    "53": "SUCCESS",
+    "52": "SUCCESS",
+    "53": "NOT_ALIVE",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NON_PARTICIPATING",
+    "56": "NOT_ALIVE",
     "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
-    "6": "NOT_ALIVE",
+    "59": "NOT_ALIVE",
+    "6": "NON_PARTICIPATING",
     "60": "NON_PARTICIPATING",
-    "61": "NOT_ALIVE",
+    "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
-    "63": "FAIL",
+    "63": "SUCCESS",
     "64": "SUCCESS",
-    "65": "NON_PARTICIPATING",
+    "65": "NOT_ALIVE",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NON_PARTICIPATING",
     "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
+    "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
-    "74": "NOT_ALIVE",
+    "74": "SUCCESS",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
@@ -238,22 +238,22 @@
     "8": "SUCCESS",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
     "84": "SUCCESS",
-    "85": "NON_PARTICIPATING",
-    "86": "SUCCESS",
+    "85": "NOT_ALIVE",
+    "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NOT_ALIVE",
     "89": "SUCCESS",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "SUCCESS",
+    "91": "NOT_ALIVE",
     "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "96": "SUCCESS",
     "97": "NON_PARTICIPATING",
     "98": "SUCCESS",
     "99": "SUCCESS"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6526381,
+  "last_allreduce_block": 6526780,
   "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:db78007a26a6805487d6027d3aa2218fbc7a8292c2e4ac8122b64f819cf015b7
+oid sha256:675c26f938341413d03c1d32d301a4b351daf73970f6396b6c7d76f5bab32fd0
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ecb8cb52a34a68dcb76d771a17474abeffe125a76a6143620f489c2dd3f0e2f8
+oid sha256:f8de240d879372853fc81c6d52ce1caf75bde1d19645eb5f91176a7fd4fa8a00
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1a0f3ffd8f558f617c12a46d4bfd556ad241ef90567470dd662e43582bc2a63
+oid sha256:c96a3d97277d6f8c6f57312a23c8021bfaee2c29f534cbcedb82acffee021e62
 size 4400252568