Run 7. Outer Step 110. Inner Step 0. Peers 33.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "AussieAck/D24_model",
+  "_name_or_path": "AussieAck/D2_model",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "SUCCESS",
@@ -9,7 +9,7 @@
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "NOT_ALIVE",
-    "105": "NON_PARTICIPATING",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
@@ -24,12 +24,12 @@
     "116": "NOT_ALIVE",
     "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
-    "119": "NOT_ALIVE",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
@@ -40,7 +40,7 @@
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "SUCCESS",
+    "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
@@ -51,10 +51,10 @@
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
-    "143": "NON_PARTICIPATING",
+    "143": "SUCCESS",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "NOT_ALIVE",
+    "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
@@ -64,17 +64,17 @@
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
     "154": "NOT_ALIVE",
-    "155": "SUCCESS",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "NON_PARTICIPATING",
+    "158": "SUCCESS",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "SUCCESS",
-    "164": "NOT_ALIVE",
+    "163": "NON_PARTICIPATING",
+    "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
@@ -84,34 +84,34 @@
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
-    "173": "NON_PARTICIPATING",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
     "178": "NOT_ALIVE",
-    "179": "SUCCESS",
+    "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
+    "180": "NOT_ALIVE",
     "181": "SUCCESS",
     "182": "NOT_ALIVE",
-    "183": "NOT_ALIVE",
+    "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
     "186": "SUCCESS",
     "187": "SUCCESS",
-    "188": "NON_PARTICIPATING",
+    "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NON_PARTICIPATING",
-    "192": "NOT_ALIVE",
+    "192": "SUCCESS",
     "193": "SUCCESS",
-    "194": "NON_PARTICIPATING",
+    "194": "SUCCESS",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NOT_ALIVE",
+    "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
@@ -127,7 +127,7 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NON_PARTICIPATING",
-    "211": "SUCCESS",
+    "211": "NON_PARTICIPATING",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
     "214": "NOT_ALIVE",
@@ -137,9 +137,9 @@
     "218": "NON_PARTICIPATING",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "NOT_ALIVE",
     "221": "NOT_ALIVE",
-    "222": "NOT_ALIVE",
+    "222": "SUCCESS",
     "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "NOT_ALIVE",
@@ -152,9 +152,9 @@
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
-    "234": "NON_PARTICIPATING",
+    "234": "SUCCESS",
     "235": "NON_PARTICIPATING",
-    "236": "NON_PARTICIPATING",
+    "236": "NOT_ALIVE",
     "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
@@ -163,15 +163,15 @@
     "241": "SUCCESS",
     "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
-    "244": "NOT_ALIVE",
+    "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
     "246": "SUCCESS",
-    "247": "NOT_ALIVE",
+    "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "FAIL",
+    "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -180,21 +180,21 @@
     "27": "NOT_ALIVE",
     "28": "SUCCESS",
     "29": "NOT_ALIVE",
-    "3": "SUCCESS",
+    "3": "NOT_ALIVE",
     "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
-    "36": "SUCCESS",
-    "37": "NON_PARTICIPATING",
+    "36": "NOT_ALIVE",
+    "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
-    "42": "NON_PARTICIPATING",
+    "42": "NOT_ALIVE",
     "43": "NON_PARTICIPATING",
     "44": "SUCCESS",
     "45": "NOT_ALIVE",
@@ -205,22 +205,22 @@
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
-    "52": "NOT_ALIVE",
-    "53": "SUCCESS",
+    "52": "SUCCESS",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "NOT_ALIVE",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "61": "NOT_ALIVE",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
-    "66": "NON_PARTICIPATING",
+    "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
@@ -232,13 +232,13 @@
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
-    "77": "NOT_ALIVE",
-    "78": "NON_PARTICIPATING",
+    "77": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
@@ -249,7 +249,7 @@
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
-    "92": "NOT_ALIVE",
+    "92": "NON_PARTICIPATING",
     "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": 6299753,
+  "last_allreduce_block": 6300423,
   "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 sha256:2de35224e1bb55202a794cf83d1da2a7197a8a4015d6dc88c1aacfc4b0975e61
+oid sha256:6443d7857be379deb2e3bbc7aab27431d6996acf5e4abc129ed7ddba5eb43a8d
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0af002d1c229427985e9e2a1d4a2041a77d6afc28fe82dac24e2b2d391680b7a
+oid sha256:608f2d0e5d4c33ef45e7104ec5ded13b53ad4ac8c7bbded02169361ca4f3c265
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c912a1d19df065cdb26c23eb25ba6a0180d05b417ace031b47085c729d5557d
+oid sha256:99afddb32c1f65b00670549d13378c2225018cd119350b90e0842f9e20357570
 size 4400252568