Run 7. Outer Step 88. Inner Step 0. Peers 33.

config.json

@@ -4,11 +4,11 @@
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
-    "100": "NON_PARTICIPATING",
+    "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
@@ -17,12 +17,12 @@
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "NOT_ALIVE",
-    "112": "NON_PARTICIPATING",
+    "112": "NOT_ALIVE",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
-    "117": "NON_PARTICIPATING",
+    "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
@@ -37,13 +37,13 @@
     "128": "NOT_ALIVE",
     "129": "SUCCESS",
     "13": "NOT_ALIVE",
-    "130": "SUCCESS",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NOT_ALIVE",
-    "136": "NOT_ALIVE",
+    "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
     "138": "SUCCESS",
     "139": "SUCCESS",
@@ -54,45 +54,45 @@
     "143": "SUCCESS",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NOT_ALIVE",
+    "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
-    "15": "NOT_ALIVE",
+    "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
     "154": "NON_PARTICIPATING",
-    "155": "NON_PARTICIPATING",
+    "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "NON_PARTICIPATING",
+    "163": "SUCCESS",
     "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
-    "170": "NON_PARTICIPATING",
+    "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
-    "174": "NON_PARTICIPATING",
+    "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
     "178": "SUCCESS",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
+    "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
@@ -100,8 +100,8 @@
     "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
-    "189": "NOT_ALIVE",
+    "188": "NON_PARTICIPATING",
+    "189": "SUCCESS",
     "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
     "191": "SUCCESS",
@@ -121,7 +121,7 @@
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "NOT_ALIVE",
@@ -132,10 +132,10 @@
     "213": "NOT_ALIVE",
     "214": "SUCCESS",
     "215": "NOT_ALIVE",
-    "216": "NON_PARTICIPATING",
-    "217": "SUCCESS",
+    "216": "SUCCESS",
+    "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
-    "219": "SUCCESS",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
@@ -152,7 +152,7 @@
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
-    "234": "NON_PARTICIPATING",
+    "234": "SUCCESS",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
@@ -161,10 +161,10 @@
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "SUCCESS",
-    "242": "NON_PARTICIPATING",
+    "242": "SUCCESS",
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
-    "245": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
     "246": "SUCCESS",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
@@ -173,14 +173,14 @@
     "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
-    "253": "NON_PARTICIPATING",
+    "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "NOT_ALIVE",
     "27": "SUCCESS",
-    "28": "SUCCESS",
+    "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "SUCCESS",
+    "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
@@ -215,7 +215,7 @@
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "61": "SUCCESS",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
@@ -223,14 +223,14 @@
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "SUCCESS",
+    "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
     "74": "NOT_ALIVE",
-    "75": "SUCCESS",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
     "78": "NOT_ALIVE",
@@ -242,21 +242,21 @@
     "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
-    "86": "NOT_ALIVE",
-    "87": "SUCCESS",
+    "86": "NON_PARTICIPATING",
+    "87": "NOT_ALIVE",
     "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
-    "92": "SUCCESS",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
     "98": "NON_PARTICIPATING",
-    "99": "SUCCESS"
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6282086,
+  "last_allreduce_block": 6282836,
   "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:f9f2e42eef176dfd988611f3f3a3e0e2f0c45b13ff3c7f52e12930e2af783f82
+oid sha256:19ec80c54938d911b9163b31b3ce429f04f18a8c3c30964830cd1961ffbe23e5
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc317aada8c32e4478be2d141e6b28fad7052209abc68f10508b8cbbfa1fe8de
+oid sha256:3c4670d59c903b4dab120c4ae1f6902445456db850b93bad0cf3de304271e974
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d7c8796ba8a396529a31b5383e7c6f928ec94410f8ed4696e178e3e6e8bd386
+oid sha256:aa4e6cedc807655430cadad47d14d789107278da7781cf9ce8ba3de2d6fdfa54
 size 4400252568