Run 7. Outer Step 92. Inner Step 0. Peers 33.

config.json

@@ -8,8 +8,8 @@
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
@@ -20,7 +20,7 @@
     "112": "NON_PARTICIPATING",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
     "117": "SUCCESS",
     "118": "NOT_ALIVE",
@@ -29,10 +29,10 @@
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "123": "SUCCESS",
     "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
-    "126": "NON_PARTICIPATING",
+    "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
@@ -41,79 +41,79 @@
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "SUCCESS",
-    "134": "SUCCESS",
+    "134": "NOT_ALIVE",
     "135": "NOT_ALIVE",
-    "136": "NON_PARTICIPATING",
+    "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "SUCCESS",
     "139": "SUCCESS",
     "14": "NOT_ALIVE",
-    "140": "NON_PARTICIPATING",
+    "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "SUCCESS",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
+    "146": "NOT_ALIVE",
     "147": "NOT_ALIVE",
     "148": "SUCCESS",
-    "149": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
     "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NON_PARTICIPATING",
-    "158": "SUCCESS",
+    "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
-    "163": "SUCCESS",
+    "162": "NOT_ALIVE",
+    "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
-    "170": "NOT_ALIVE",
+    "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
-    "174": "NON_PARTICIPATING",
+    "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "SUCCESS",
     "179": "NON_PARTICIPATING",
-    "18": "SUCCESS",
-    "180": "NOT_ALIVE",
+    "18": "NOT_ALIVE",
+    "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
+    "183": "NOT_ALIVE",
     "184": "NOT_ALIVE",
-    "185": "NON_PARTICIPATING",
+    "185": "SUCCESS",
     "186": "SUCCESS",
     "187": "NOT_ALIVE",
-    "188": "NON_PARTICIPATING",
+    "188": "NOT_ALIVE",
     "189": "SUCCESS",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "SUCCESS",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
-    "194": "NOT_ALIVE",
+    "194": "NON_PARTICIPATING",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
     "199": "NOT_ALIVE",
-    "2": "NOT_ALIVE",
+    "2": "SUCCESS",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
@@ -130,12 +130,12 @@
     "211": "NON_PARTICIPATING",
     "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
-    "214": "SUCCESS",
+    "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "SUCCESS",
     "217": "SUCCESS",
     "218": "NON_PARTICIPATING",
-    "219": "SUCCESS",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
@@ -148,24 +148,24 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "SUCCESS",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
-    "237": "NOT_ALIVE",
+    "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "SUCCESS",
-    "242": "SUCCESS",
+    "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
-    "246": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
@@ -177,12 +177,12 @@
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "SUCCESS",
+    "27": "NON_PARTICIPATING",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "SUCCESS",
-    "31": "NOT_ALIVE",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
@@ -190,12 +190,12 @@
     "36": "SUCCESS",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
+    "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
@@ -205,15 +205,15 @@
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
-    "52": "NON_PARTICIPATING",
-    "53": "NOT_ALIVE",
+    "52": "SUCCESS",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
-    "6": "NOT_ALIVE",
+    "6": "SUCCESS",
     "60": "NON_PARTICIPATING",
     "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
@@ -232,14 +232,14 @@
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
-    "77": "NOT_ALIVE",
+    "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
-    "81": "NON_PARTICIPATING",
+    "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
-    "83": "NON_PARTICIPATING",
+    "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
@@ -249,13 +249,13 @@
     "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": "NOT_ALIVE"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6285056,
+  "last_allreduce_block": 6285781,
   "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 sha256:51f3e394405d964c5e08de4c882bcdbb62ab0cf3ffacc8b3bb6c6cc3dd34d9a1
+oid sha256:760479b2b5713df6fcb2c0e4af3a732100a31d1aa4732e29af9bc98ae52a7d4c
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ab97d4aed73d955129f1dcae9a4d186ccddd7e77fe8afed0e05f70158bbc466
+oid sha256:86006cae2d41f7800d27e88f2970997faaca2d1d0a91cc92dec6b454b9d5c58b
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aed5162d321651903bd4a005de99dac0b14d1df6a4f59db197deb5e9ce5b10dd
+oid sha256:1619defc30fd549421734a80fca637d019e7f4ed10bc41d61defe9f8ff5c671e
 size 4400252568