Run 7. Outer Step 89. Inner Step 0. Peers 31.

config.json

@@ -1,14 +1,14 @@
 {
-  "_name_or_path": "AussieAck/D24_model",
+  "_name_or_path": "strnan01/llama-1b-hot-6",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
-    "100": "NOT_ALIVE",
+    "100": "NON_PARTICIPATING",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
+    "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
@@ -17,12 +17,12 @@
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "NOT_ALIVE",
-    "112": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
-    "117": "NOT_ALIVE",
+    "117": "NON_PARTICIPATING",
     "118": "NOT_ALIVE",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
@@ -35,13 +35,13 @@
     "126": "NON_PARTICIPATING",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
-    "129": "SUCCESS",
+    "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
+    "133": "SUCCESS",
+    "134": "FAIL",
     "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
@@ -52,58 +52,58 @@
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "SUCCESS",
-    "144": "NOT_ALIVE",
+    "144": "SUCCESS",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
-    "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
+    "148": "SUCCESS",
+    "149": "NOT_ALIVE",
     "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
     "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "NON_PARTICIPATING",
+    "158": "SUCCESS",
     "159": "NON_PARTICIPATING",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "SUCCESS",
+    "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "SUCCESS",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
-    "174": "NOT_ALIVE",
-    "175": "NON_PARTICIPATING",
+    "174": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
     "178": "SUCCESS",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
-    "180": "NOT_ALIVE",
+    "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
     "184": "NOT_ALIVE",
-    "185": "NON_PARTICIPATING",
+    "185": "SUCCESS",
     "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "190": "NOT_ALIVE",
     "191": "SUCCESS",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
@@ -112,7 +112,7 @@
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
-    "199": "SUCCESS",
+    "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
@@ -135,13 +135,13 @@
     "216": "SUCCESS",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
-    "219": "NON_PARTICIPATING",
+    "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
-    "224": "NON_PARTICIPATING",
+    "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
@@ -160,7 +160,7 @@
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "SUCCESS",
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
@@ -176,26 +176,26 @@
     "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
-    "27": "SUCCESS",
+    "26": "NON_PARTICIPATING",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "NOT_ALIVE",
+    "3": "SUCCESS",
     "30": "NOT_ALIVE",
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "SUCCESS",
-    "36": "SUCCESS",
+    "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "SUCCESS",
+    "40": "FAIL",
     "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
+    "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
@@ -206,13 +206,13 @@
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
     "52": "SUCCESS",
-    "53": "NON_PARTICIPATING",
+    "53": "SUCCESS",
     "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": "SUCCESS",
@@ -226,14 +226,14 @@
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
-    "73": "SUCCESS",
+    "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
-    "78": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "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": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6282836,
+  "last_allreduce_block": 6283583,
   "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 sha256:19ec80c54938d911b9163b31b3ce429f04f18a8c3c30964830cd1961ffbe23e5
+oid sha256:e39a8204fa904c32b0710e780091bdac6914485fa4d559328a8c16349a114230
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c4670d59c903b4dab120c4ae1f6902445456db850b93bad0cf3de304271e974
+oid sha256:a7bf06e52b0c25b6e6e00b6af08e9da444c0ca69a6913101477e5655d9eabbb9
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa4e6cedc807655430cadad47d14d789107278da7781cf9ce8ba3de2d6fdfa54
+oid sha256:9e60f849cc6a3d53f69f5412e2df0bf132b105f88feaa5e7ae1204713fa8c404
 size 4400252568