Run 7. Outer Step 77. Inner Step 0. Peers 33.

config.json

@@ -1,11 +1,11 @@
 {
-  "_name_or_path": "distributed/llama-1b",
+  "_name_or_path": "strnan01/llama-1b-hot-6",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
-    "100": "NON_PARTICIPATING",
-    "101": "NON_PARTICIPATING",
+    "100": "NOT_ALIVE",
+    "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "NOT_ALIVE",
@@ -23,13 +23,13 @@
     "115": "NOT_ALIVE",
     "116": "NOT_ALIVE",
     "117": "SUCCESS",
-    "118": "NOT_ALIVE",
+    "118": "NON_PARTICIPATING",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
-    "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
+    "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
@@ -39,16 +39,16 @@
     "13": "NOT_ALIVE",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
-    "132": "SUCCESS",
+    "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
+    "134": "NOT_ALIVE",
     "135": "SUCCESS",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
     "139": "SUCCESS",
-    "14": "NOT_ALIVE",
-    "140": "SUCCESS",
+    "14": "SUCCESS",
+    "140": "NON_PARTICIPATING",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "SUCCESS",
@@ -56,57 +56,57 @@
     "145": "SUCCESS",
     "146": "NON_PARTICIPATING",
     "147": "NOT_ALIVE",
-    "148": "NON_PARTICIPATING",
+    "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
-    "154": "NON_PARTICIPATING",
+    "154": "SUCCESS",
     "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NON_PARTICIPATING",
-    "158": "SUCCESS",
-    "159": "NON_PARTICIPATING",
+    "158": "NOT_ALIVE",
+    "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "NOT_ALIVE",
+    "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
-    "165": "NON_PARTICIPATING",
+    "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
-    "169": "SUCCESS",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
-    "171": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
-    "179": "SUCCESS",
+    "178": "SUCCESS",
+    "179": "NON_PARTICIPATING",
     "18": "SUCCESS",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
+    "183": "NOT_ALIVE",
     "184": "NOT_ALIVE",
-    "185": "NON_PARTICIPATING",
-    "186": "NOT_ALIVE",
+    "185": "NOT_ALIVE",
+    "186": "SUCCESS",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
-    "189": "SUCCESS",
+    "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "SUCCESS",
     "192": "SUCCESS",
-    "193": "NOT_ALIVE",
+    "193": "NON_PARTICIPATING",
     "194": "SUCCESS",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
@@ -117,7 +117,7 @@
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
+    "202": "NOT_ALIVE",
     "203": "NON_PARTICIPATING",
     "204": "NON_PARTICIPATING",
     "205": "NOT_ALIVE",
@@ -127,17 +127,17 @@
     "209": "NOT_ALIVE",
     "21": "NOT_ALIVE",
     "210": "NON_PARTICIPATING",
-    "211": "SUCCESS",
+    "211": "NON_PARTICIPATING",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
-    "214": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NON_PARTICIPATING",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
-    "220": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
     "222": "SUCCESS",
     "223": "NOT_ALIVE",
@@ -145,7 +145,7 @@
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
-    "228": "NOT_ALIVE",
+    "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
@@ -156,12 +156,12 @@
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
-    "238": "NOT_ALIVE",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "NOT_ALIVE",
-    "242": "FAIL",
+    "241": "SUCCESS",
+    "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
@@ -170,27 +170,27 @@
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
-    "251": "NOT_ALIVE",
+    "250": "NOT_ALIVE",
+    "251": "SUCCESS",
     "252": "NOT_ALIVE",
-    "253": "NOT_ALIVE",
+    "253": "NON_PARTICIPATING",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "NOT_ALIVE",
+    "27": "NON_PARTICIPATING",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "NOT_ALIVE",
+    "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
-    "39": "NOT_ALIVE",
+    "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
@@ -198,8 +198,8 @@
     "43": "NON_PARTICIPATING",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
-    "46": "NOT_ALIVE",
-    "47": "SUCCESS",
+    "46": "NON_PARTICIPATING",
+    "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
     "49": "NON_PARTICIPATING",
     "5": "NOT_ALIVE",
@@ -214,20 +214,20 @@
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
-    "67": "NOT_ALIVE",
+    "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "SUCCESS",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
-    "72": "NON_PARTICIPATING",
+    "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
@@ -241,7 +241,7 @@
     "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
-    "85": "NOT_ALIVE",
+    "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
     "87": "NOT_ALIVE",
     "88": "NOT_ALIVE",
@@ -256,7 +256,7 @@
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
     "98": "NON_PARTICIPATING",
-    "99": "NOT_ALIVE"
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6274145,
+  "last_allreduce_block": 6274810,
   "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:c7ba3896844939b90d86c3edd6db79ed13c07ac6002e124227f81955b95df172
+oid sha256:77c6a0db2b50644d9520c8277678ed141c251d04ee0b1b74aeff1707aa90d519
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d93899848e71416ddc4c247c433d46c96443afca600ffe1c6234554c9cc1f9c
+oid sha256:d4e816a8ce5d5623db29df674aa8ad7bd79c7f465c95671eea894b041a352cbb
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50628e6cd31fe56e881607ef9ce2df1e3bcf625f8756b651e53e6a748363c089
+oid sha256:1c64743baad0b670acea7ee7ee2b6c65b2fe46fd2625a5d7ea1ed7fd47270b56
 size 4400252568