Run 7. Outer Step 56. Inner Step 0. Peers 32.

config.json

@@ -9,27 +9,27 @@
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
-    "108": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "NOT_ALIVE",
-    "112": "SUCCESS",
+    "112": "NON_PARTICIPATING",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
-    "118": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
-    "123": "NOT_ALIVE",
+    "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
@@ -44,11 +44,11 @@
     "134": "SUCCESS",
     "135": "NOT_ALIVE",
     "136": "SUCCESS",
-    "137": "NON_PARTICIPATING",
+    "137": "NOT_ALIVE",
     "138": "SUCCESS",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
-    "140": "NON_PARTICIPATING",
+    "140": "SUCCESS",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
@@ -56,7 +56,7 @@
     "145": "NOT_ALIVE",
     "146": "SUCCESS",
     "147": "NOT_ALIVE",
-    "148": "NOT_ALIVE",
+    "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "SUCCESS",
@@ -74,24 +74,24 @@
     "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
     "163": "NON_PARTICIPATING",
-    "164": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "SUCCESS",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "SUCCESS",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
-    "170": "NOT_ALIVE",
+    "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "SUCCESS",
     "179": "SUCCESS",
-    "18": "SUCCESS",
+    "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
@@ -100,14 +100,14 @@
     "185": "NON_PARTICIPATING",
     "186": "SUCCESS",
     "187": "NOT_ALIVE",
-    "188": "NON_PARTICIPATING",
+    "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NON_PARTICIPATING",
     "192": "NON_PARTICIPATING",
-    "193": "NOT_ALIVE",
-    "194": "FAIL",
+    "193": "NON_PARTICIPATING",
+    "194": "NOT_ALIVE",
     "195": "SUCCESS",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
@@ -117,7 +117,7 @@
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
+    "202": "NOT_ALIVE",
     "203": "NON_PARTICIPATING",
     "204": "NON_PARTICIPATING",
     "205": "SUCCESS",
@@ -126,7 +126,7 @@
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
-    "210": "NOT_ALIVE",
+    "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
@@ -139,13 +139,13 @@
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "SUCCESS",
-    "222": "NON_PARTICIPATING",
+    "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
-    "228": "NOT_ALIVE",
+    "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
@@ -156,8 +156,8 @@
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
-    "238": "NOT_ALIVE",
-    "239": "NON_PARTICIPATING",
+    "238": "SUCCESS",
+    "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
@@ -165,7 +165,7 @@
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
-    "246": "SUCCESS",
+    "246": "FAIL",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
@@ -177,7 +177,7 @@
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "NOT_ALIVE",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
@@ -188,14 +188,14 @@
     "34": "NON_PARTICIPATING",
     "35": "NOT_ALIVE",
     "36": "SUCCESS",
-    "37": "NON_PARTICIPATING",
+    "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "SUCCESS",
     "41": "NOT_ALIVE",
     "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
+    "43": "NOT_ALIVE",
     "44": "SUCCESS",
     "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
@@ -205,14 +205,14 @@
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
-    "52": "NOT_ALIVE",
+    "52": "NON_PARTICIPATING",
     "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "NON_PARTICIPATING",
@@ -228,7 +228,7 @@
     "70": "NON_PARTICIPATING",
     "71": "NON_PARTICIPATING",
     "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
@@ -242,20 +242,20 @@
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
-    "86": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
-    "92": "NOT_ALIVE",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
-    "98": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6257064,
+  "last_allreduce_block": 6257816,
   "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 sha256:568eed242c8888e74b620ddbee4c6214129341382fa2ad9840e294a6d5edd92c
+oid sha256:47678fcae7a93026e09e230390e3e20ad5907e5a927f3e7f56cbfd91f70bfc7c
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f94f3995653035ec30b8701c257dfd4690e462868442fd42ab283e30c06a9e0d
+oid sha256:16c49c4ed00cef635fbb78ad362e2c24e1a5c44f6417b02412ea0e0983e804ec
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61b3d511e3bf82a10cc6855fcf07999f6c9c88c65dc614d978906009f40d37a8
+oid sha256:a43fad239a6e36e5368c088011cb09ed8125e30d51290498c36a2c54d7c0da45
 size 4400252568