Run 8. Outer Step 127. Inner Step 5.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "zeras141a/a6",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
@@ -12,112 +12,112 @@
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NON_PARTICIPATING",
-    "108": "NOT_ALIVE",
-    "109": "FAIL",
+    "108": "NON_PARTICIPATING",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
-    "111": "FAIL",
+    "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
     "113": "NON_PARTICIPATING",
     "114": "SUCCESS",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "NON_PARTICIPATING",
-    "118": "NOT_ALIVE",
+    "117": "FAIL",
+    "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
-    "124": "NON_PARTICIPATING",
-    "125": "FAIL",
+    "124": "NOT_ALIVE",
+    "125": "SUCCESS",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
+    "127": "NON_PARTICIPATING",
     "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "FAIL",
+    "130": "NON_PARTICIPATING",
     "131": "NON_PARTICIPATING",
-    "132": "NOT_ALIVE",
-    "133": "NON_PARTICIPATING",
+    "132": "NON_PARTICIPATING",
+    "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "NOT_ALIVE",
-    "136": "NON_PARTICIPATING",
+    "136": "NOT_ALIVE",
     "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
     "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
-    "141": "FAIL",
-    "142": "FAIL",
-    "143": "FAIL",
-    "144": "NON_PARTICIPATING",
+    "141": "SUCCESS",
+    "142": "SUCCESS",
+    "143": "SUCCESS",
+    "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "FAIL",
+    "146": "SUCCESS",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
-    "151": "NON_PARTICIPATING",
-    "152": "FAIL",
+    "151": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
     "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
-    "155": "NON_PARTICIPATING",
+    "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "FAIL",
+    "158": "SUCCESS",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
-    "162": "NOT_ALIVE",
+    "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
-    "165": "FAIL",
+    "165": "SUCCESS",
     "166": "SUCCESS",
-    "167": "NON_PARTICIPATING",
-    "168": "FAIL",
+    "167": "SUCCESS",
+    "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
-    "171": "FAIL",
-    "172": "NON_PARTICIPATING",
+    "171": "NON_PARTICIPATING",
+    "172": "NOT_ALIVE",
     "173": "NOT_ALIVE",
     "174": "NON_PARTICIPATING",
-    "175": "FAIL",
+    "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "NON_PARTICIPATING",
-    "178": "NOT_ALIVE",
+    "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "NON_PARTICIPATING",
+    "182": "SUCCESS",
     "183": "SUCCESS",
-    "184": "FAIL",
-    "185": "FAIL",
+    "184": "NOT_ALIVE",
+    "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
-    "189": "FAIL",
+    "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "190": "SUCCESS",
     "191": "NON_PARTICIPATING",
     "192": "SUCCESS",
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
-    "196": "FAIL",
+    "195": "SUCCESS",
+    "196": "NON_PARTICIPATING",
     "197": "NOT_ALIVE",
-    "198": "FAIL",
-    "199": "NOT_ALIVE",
+    "198": "SUCCESS",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
     "200": "SUCCESS",
     "201": "SUCCESS",
-    "202": "NON_PARTICIPATING",
+    "202": "FAIL",
     "203": "SUCCESS",
     "204": "NOT_ALIVE",
     "205": "NON_PARTICIPATING",
@@ -126,137 +126,137 @@
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "FAIL",
-    "211": "SUCCESS",
+    "210": "SUCCESS",
+    "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
-    "213": "FAIL",
-    "214": "SUCCESS",
+    "213": "NON_PARTICIPATING",
+    "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "FAIL",
-    "217": "FAIL",
-    "218": "SUCCESS",
+    "216": "SUCCESS",
+    "217": "NON_PARTICIPATING",
+    "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "SUCCESS",
+    "220": "NON_PARTICIPATING",
     "221": "NON_PARTICIPATING",
-    "222": "NON_PARTICIPATING",
-    "223": "FAIL",
+    "222": "NOT_ALIVE",
+    "223": "NON_PARTICIPATING",
     "224": "NON_PARTICIPATING",
-    "225": "NON_PARTICIPATING",
+    "225": "SUCCESS",
     "226": "NON_PARTICIPATING",
-    "227": "NOT_ALIVE",
-    "228": "FAIL",
+    "227": "NON_PARTICIPATING",
+    "228": "SUCCESS",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
-    "234": "FAIL",
+    "233": "NON_PARTICIPATING",
+    "234": "SUCCESS",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
     "237": "SUCCESS",
-    "238": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
     "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
+    "240": "NON_PARTICIPATING",
     "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
-    "243": "NON_PARTICIPATING",
+    "242": "NOT_ALIVE",
+    "243": "NOT_ALIVE",
     "244": "NON_PARTICIPATING",
     "245": "SUCCESS",
     "246": "NON_PARTICIPATING",
-    "247": "FAIL",
-    "248": "NON_PARTICIPATING",
-    "249": "FAIL",
+    "247": "SUCCESS",
+    "248": "SUCCESS",
+    "249": "SUCCESS",
     "25": "SUCCESS",
-    "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
+    "250": "NON_PARTICIPATING",
+    "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
-    "253": "NON_PARTICIPATING",
+    "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "NON_PARTICIPATING",
-    "28": "FAIL",
-    "29": "NON_PARTICIPATING",
+    "28": "NON_PARTICIPATING",
+    "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
-    "30": "NON_PARTICIPATING",
-    "31": "NOT_ALIVE",
+    "30": "SUCCESS",
+    "31": "NON_PARTICIPATING",
     "32": "NON_PARTICIPATING",
     "33": "NON_PARTICIPATING",
     "34": "NON_PARTICIPATING",
     "35": "NON_PARTICIPATING",
-    "36": "NON_PARTICIPATING",
-    "37": "NOT_ALIVE",
-    "38": "NON_PARTICIPATING",
+    "36": "SUCCESS",
+    "37": "NON_PARTICIPATING",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
-    "42": "NOT_ALIVE",
-    "43": "FAIL",
+    "42": "NON_PARTICIPATING",
+    "43": "SUCCESS",
     "44": "NOT_ALIVE",
-    "45": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
-    "47": "FAIL",
+    "47": "SUCCESS",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NON_PARTICIPATING",
-    "52": "FAIL",
+    "51": "SUCCESS",
+    "52": "SUCCESS",
     "53": "NON_PARTICIPATING",
     "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
+    "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
-    "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "FAIL",
+    "6": "SUCCESS",
+    "60": "SUCCESS",
+    "61": "SUCCESS",
     "62": "NOT_ALIVE",
-    "63": "NON_PARTICIPATING",
-    "64": "NON_PARTICIPATING",
-    "65": "NOT_ALIVE",
+    "63": "SUCCESS",
+    "64": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
-    "68": "FAIL",
+    "68": "NON_PARTICIPATING",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
-    "70": "NON_PARTICIPATING",
+    "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
     "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
     "74": "SUCCESS",
     "75": "NON_PARTICIPATING",
-    "76": "FAIL",
+    "76": "SUCCESS",
     "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
-    "82": "SUCCESS",
+    "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "NON_PARTICIPATING",
-    "87": "FAIL",
-    "88": "FAIL",
+    "86": "SUCCESS",
+    "87": "SUCCESS",
+    "88": "SUCCESS",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
-    "90": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
     "91": "SUCCESS",
     "92": "NON_PARTICIPATING",
-    "93": "NON_PARTICIPATING",
+    "93": "SUCCESS",
     "94": "SUCCESS",
     "95": "NON_PARTICIPATING",
     "96": "NON_PARTICIPATING",
-    "97": "NON_PARTICIPATING",
+    "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "FAIL"
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6443776,
-    6443794
+    6443860,
+    6443882
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6443186,
+  "last_allreduce_block": 6443839,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

gradients.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b85fb0ad974e5f4d1365dd5e26360fa255395d184c5be53401ef3d02704820ec
+oid sha256:acd1db2cc4b430f4e6895592229d224d7b339655702feac7b62c161c2c472226
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f894412397bdd68730d703a9ebf9fe03f1de4c85d26aeb3d178f2e4e0dfd00fa
+oid sha256:a236df3fd22069d9016d32ff4a006bf0d8a4c52fdba7eef546d80b91ef1f2f07
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5d8082b7878c62abd37a8d1025b9938f83827e5678b4e7c660ba249eef63a2a
+oid sha256:a2c018cb5d01723d6eac6031640ccbd705c2b85e4bb6bd61bf72d429da2318a7
 size 4400216536