Run 7. Outer Step 109. Inner Step 0. Peers 32.

config.json

@@ -20,17 +20,17 @@
     "112": "NOT_ALIVE",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
+    "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
-    "117": "NOT_ALIVE",
-    "118": "FAIL",
-    "119": "NON_PARTICIPATING",
+    "117": "NON_PARTICIPATING",
+    "118": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
-    "124": "NON_PARTICIPATING",
+    "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
@@ -45,41 +45,41 @@
     "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NON_PARTICIPATING",
-    "138": "NON_PARTICIPATING",
-    "139": "SUCCESS",
+    "138": "SUCCESS",
+    "139": "NOT_ALIVE",
     "14": "SUCCESS",
-    "140": "NON_PARTICIPATING",
+    "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
-    "143": "SUCCESS",
-    "144": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
+    "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
-    "147": "NOT_ALIVE",
+    "146": "NOT_ALIVE",
+    "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "FAIL",
+    "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
-    "154": "NON_PARTICIPATING",
+    "154": "NOT_ALIVE",
     "155": "SUCCESS",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "SUCCESS",
+    "158": "NON_PARTICIPATING",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
     "163": "SUCCESS",
-    "164": "NON_PARTICIPATING",
+    "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
@@ -88,26 +88,26 @@
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "SUCCESS",
     "178": "NOT_ALIVE",
     "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "SUCCESS",
     "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
+    "183": "NOT_ALIVE",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
-    "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
+    "186": "SUCCESS",
+    "187": "SUCCESS",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NON_PARTICIPATING",
     "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
-    "194": "NOT_ALIVE",
+    "193": "SUCCESS",
+    "194": "NON_PARTICIPATING",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
@@ -127,17 +127,17 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NON_PARTICIPATING",
-    "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
+    "211": "SUCCESS",
+    "212": "SUCCESS",
     "213": "NOT_ALIVE",
-    "214": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
     "217": "SUCCESS",
     "218": "NON_PARTICIPATING",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
@@ -153,32 +153,32 @@
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
-    "235": "NOT_ALIVE",
-    "236": "NOT_ALIVE",
+    "235": "NON_PARTICIPATING",
+    "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "SUCCESS",
     "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NOT_ALIVE",
     "246": "SUCCESS",
-    "247": "NON_PARTICIPATING",
+    "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "SUCCESS",
+    "251": "FAIL",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
-    "28": "NON_PARTICIPATING",
+    "28": "SUCCESS",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
     "30": "SUCCESS",
@@ -187,35 +187,35 @@
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
-    "36": "NOT_ALIVE",
+    "36": "SUCCESS",
     "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
-    "41": "NOT_ALIVE",
+    "40": "SUCCESS",
+    "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
     "43": "NON_PARTICIPATING",
     "44": "SUCCESS",
     "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
+    "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
-    "52": "NON_PARTICIPATING",
-    "53": "NON_PARTICIPATING",
+    "52": "NOT_ALIVE",
+    "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
-    "6": "FAIL",
+    "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NOT_ALIVE",
+    "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
@@ -226,13 +226,13 @@
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
-    "77": "NON_PARTICIPATING",
+    "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
@@ -249,7 +249,7 @@
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
-    "92": "NON_PARTICIPATING",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6299087,
+  "last_allreduce_block": 6299753,
   "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:f2f7ff724b0ebd30ad013128451bef7cdc63573cab6c8307f32c4c8f3278738f
+oid sha256:2de35224e1bb55202a794cf83d1da2a7197a8a4015d6dc88c1aacfc4b0975e61
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e64d1d3164d19d5e7eb7d952a3d8d4fdd961bb92017ec7516f1a4b3913728af
+oid sha256:0af002d1c229427985e9e2a1d4a2041a77d6afc28fe82dac24e2b2d391680b7a
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:32e81563ae4a92d76b373361bcf4761b0ecd0429855468c1c1157f2fd049e017
+oid sha256:2c912a1d19df065cdb26c23eb25ba6a0180d05b417ace031b47085c729d5557d
 size 4400252568