Run 7. Outer Step 113. Inner Step 0. Peers 33.

config.json

@@ -17,7 +17,7 @@
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "SUCCESS",
-    "112": "FAIL",
+    "112": "NON_PARTICIPATING",
     "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
@@ -28,9 +28,9 @@
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
-    "122": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
     "123": "NOT_ALIVE",
-    "124": "NOT_ALIVE",
+    "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
@@ -41,101 +41,101 @@
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
+    "134": "SUCCESS",
     "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
-    "138": "NOT_ALIVE",
+    "138": "NON_PARTICIPATING",
     "139": "NOT_ALIVE",
     "14": "SUCCESS",
-    "140": "FAIL",
+    "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
-    "143": "NOT_ALIVE",
-    "144": "FAIL",
+    "143": "NON_PARTICIPATING",
+    "144": "SUCCESS",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
     "154": "SUCCESS",
-    "155": "NOT_ALIVE",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NON_PARTICIPATING",
-    "158": "FAIL",
-    "159": "NOT_ALIVE",
+    "158": "NON_PARTICIPATING",
+    "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "NON_PARTICIPATING",
+    "163": "SUCCESS",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
-    "166": "NOT_ALIVE",
+    "166": "SUCCESS",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
-    "169": "NON_PARTICIPATING",
+    "168": "NOT_ALIVE",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
-    "173": "FAIL",
+    "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": "NOT_ALIVE",
-    "181": "SUCCESS",
+    "180": "NON_PARTICIPATING",
+    "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
-    "189": "NOT_ALIVE",
+    "188": "NON_PARTICIPATING",
+    "189": "SUCCESS",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
-    "191": "FAIL",
-    "192": "SUCCESS",
-    "193": "SUCCESS",
-    "194": "SUCCESS",
+    "191": "SUCCESS",
+    "192": "NOT_ALIVE",
+    "193": "NON_PARTICIPATING",
+    "194": "NON_PARTICIPATING",
     "195": "SUCCESS",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
-    "2": "NOT_ALIVE",
+    "2": "SUCCESS",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
-    "204": "NON_PARTICIPATING",
+    "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
-    "21": "NOT_ALIVE",
+    "21": "SUCCESS",
     "210": "NOT_ALIVE",
-    "211": "FAIL",
+    "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
-    "214": "NOT_ALIVE",
+    "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
-    "216": "FAIL",
-    "217": "NON_PARTICIPATING",
-    "218": "NON_PARTICIPATING",
-    "219": "NON_PARTICIPATING",
+    "216": "NON_PARTICIPATING",
+    "217": "SUCCESS",
+    "218": "NOT_ALIVE",
+    "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
@@ -145,10 +145,10 @@
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
-    "228": "FAIL",
+    "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
+    "230": "NON_PARTICIPATING",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
@@ -160,20 +160,20 @@
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "FAIL",
-    "242": "FAIL",
+    "241": "NON_PARTICIPATING",
+    "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
     "246": "NON_PARTICIPATING",
-    "247": "NOT_ALIVE",
+    "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "NOT_ALIVE",
     "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
-    "253": "NON_PARTICIPATING",
+    "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
@@ -187,16 +187,16 @@
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
-    "36": "NOT_ALIVE",
+    "36": "SUCCESS",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
+    "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
-    "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
-    "44": "FAIL",
+    "42": "NOT_ALIVE",
+    "43": "NOT_ALIVE",
+    "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
     "47": "NOT_ALIVE",
@@ -213,30 +213,30 @@
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
-    "6": "NOT_ALIVE",
+    "6": "SUCCESS",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "FAIL",
+    "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
-    "73": "FAIL",
+    "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NOT_ALIVE",
+    "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "NOT_ALIVE",
+    "80": "SUCCESS",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
@@ -256,7 +256,7 @@
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
     "98": "NON_PARTICIPATING",
-    "99": "NOT_ALIVE"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6301741,
+  "last_allreduce_block": 6302427,
   "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:dc769d03d7b452633002adf9b5d83111201394372b8aae0e110513a7a2b0ba80
+oid sha256:be401ac20259b49d714a5f0499b58c37aadc47fcbec6518a7d71e4c1ef3baa2d
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bede6da8272df0da249963a87881294a95e0b7c0bd43e3aef26a44318aba6247
+oid sha256:0e2186a951723730233d72986ff5b16db6b4977880b2b5fe6d4e45a002e95e41
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:56b27b508bdb57afe9e1d542652b905603525fa86fcf4c4817082be3b06f24ef
+oid sha256:cfae0f2e2e92ecbf831049382feb8ecd346014166638f42906a3bb527817dceb
 size 4400252568