Run 9. Outer Step 19. Inner Step 99.

config.json

@@ -2,72 +2,72 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
-    "10": "NOT_ALIVE",
+    "1": "FAIL",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
-    "106": "FAIL",
+    "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "FAIL",
-    "11": "FAIL",
-    "110": "FAIL",
+    "109": "NOT_ALIVE",
+    "11": "NOT_ALIVE",
+    "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
-    "112": "FAIL",
-    "113": "FAIL",
+    "112": "NOT_ALIVE",
+    "113": "NOT_ALIVE",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "FAIL",
+    "117": "NON_PARTICIPATING",
     "118": "FAIL",
-    "119": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
+    "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
-    "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
-    "124": "NON_PARTICIPATING",
+    "122": "FAIL",
+    "123": "FAIL",
+    "124": "FAIL",
     "125": "NON_PARTICIPATING",
-    "126": "FAIL",
+    "126": "NON_PARTICIPATING",
     "127": "NON_PARTICIPATING",
-    "128": "FAIL",
-    "129": "NON_PARTICIPATING",
-    "13": "FAIL",
-    "130": "FAIL",
+    "128": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
+    "13": "NOT_ALIVE",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "134": "FAIL",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
+    "138": "NOT_ALIVE",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "FAIL",
+    "142": "NON_PARTICIPATING",
     "143": "FAIL",
     "144": "NOT_ALIVE",
-    "145": "NON_PARTICIPATING",
-    "146": "NON_PARTICIPATING",
-    "147": "FAIL",
-    "148": "NON_PARTICIPATING",
+    "145": "NOT_ALIVE",
+    "146": "NOT_ALIVE",
+    "147": "NOT_ALIVE",
+    "148": "NOT_ALIVE",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
-    "155": "FAIL",
-    "156": "FAIL",
+    "153": "FAIL",
+    "154": "FAIL",
+    "155": "NOT_ALIVE",
+    "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
-    "158": "FAIL",
+    "158": "NOT_ALIVE",
     "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
     "160": "NOT_ALIVE",
@@ -78,9 +78,9 @@
     "165": "NON_PARTICIPATING",
     "166": "FAIL",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "FAIL",
     "169": "FAIL",
-    "17": "NOT_ALIVE",
+    "17": "FAIL",
     "170": "FAIL",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
@@ -88,174 +88,174 @@
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "FAIL",
     "178": "NOT_ALIVE",
-    "179": "FAIL",
+    "179": "NOT_ALIVE",
     "18": "NON_PARTICIPATING",
-    "180": "NON_PARTICIPATING",
-    "181": "NON_PARTICIPATING",
-    "182": "NON_PARTICIPATING",
-    "183": "FAIL",
-    "184": "NOT_ALIVE",
-    "185": "NON_PARTICIPATING",
+    "180": "FAIL",
+    "181": "NOT_ALIVE",
+    "182": "FAIL",
+    "183": "NON_PARTICIPATING",
+    "184": "FAIL",
+    "185": "NOT_ALIVE",
     "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
-    "188": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
+    "188": "NOT_ALIVE",
     "189": "NON_PARTICIPATING",
-    "19": "NON_PARTICIPATING",
+    "19": "FAIL",
     "190": "NON_PARTICIPATING",
-    "191": "NON_PARTICIPATING",
-    "192": "NON_PARTICIPATING",
+    "191": "NOT_ALIVE",
+    "192": "FAIL",
     "193": "FAIL",
     "194": "NON_PARTICIPATING",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "FAIL",
     "197": "NON_PARTICIPATING",
-    "198": "NON_PARTICIPATING",
-    "199": "NOT_ALIVE",
+    "198": "NOT_ALIVE",
+    "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
-    "200": "NON_PARTICIPATING",
-    "201": "FAIL",
-    "202": "FAIL",
-    "203": "FAIL",
+    "20": "NOT_ALIVE",
+    "200": "FAIL",
+    "201": "NON_PARTICIPATING",
+    "202": "NOT_ALIVE",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "FAIL",
-    "206": "NOT_ALIVE",
+    "205": "NON_PARTICIPATING",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
-    "21": "FAIL",
-    "210": "NON_PARTICIPATING",
-    "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
+    "21": "NOT_ALIVE",
+    "210": "NOT_ALIVE",
+    "211": "NON_PARTICIPATING",
+    "212": "NON_PARTICIPATING",
     "213": "NOT_ALIVE",
-    "214": "FAIL",
+    "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
+    "216": "FAIL",
     "217": "NON_PARTICIPATING",
-    "218": "FAIL",
-    "219": "FAIL",
+    "218": "NOT_ALIVE",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "FAIL",
     "221": "NON_PARTICIPATING",
     "222": "NON_PARTICIPATING",
     "223": "FAIL",
     "224": "NOT_ALIVE",
-    "225": "FAIL",
-    "226": "NON_PARTICIPATING",
+    "225": "NOT_ALIVE",
+    "226": "NOT_ALIVE",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "FAIL",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
-    "232": "NON_PARTICIPATING",
+    "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
-    "234": "NOT_ALIVE",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "FAIL",
-    "237": "FAIL",
+    "236": "NOT_ALIVE",
+    "237": "NOT_ALIVE",
     "238": "NON_PARTICIPATING",
-    "239": "FAIL",
-    "24": "FAIL",
+    "239": "NOT_ALIVE",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
-    "243": "FAIL",
+    "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "FAIL",
-    "246": "NON_PARTICIPATING",
-    "247": "NOT_ALIVE",
-    "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "246": "FAIL",
+    "247": "NON_PARTICIPATING",
+    "248": "FAIL",
+    "249": "FAIL",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "FAIL",
     "251": "NOT_ALIVE",
-    "252": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "NON_PARTICIPATING",
-    "254": "NON_PARTICIPATING",
+    "254": "FAIL",
     "255": "NON_PARTICIPATING",
-    "26": "NOT_ALIVE",
-    "27": "NOT_ALIVE",
+    "26": "FAIL",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
     "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
-    "31": "FAIL",
+    "30": "FAIL",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
     "35": "FAIL",
     "36": "FAIL",
     "37": "NOT_ALIVE",
-    "38": "FAIL",
-    "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
+    "38": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
+    "4": "FAIL",
     "40": "NON_PARTICIPATING",
     "41": "NOT_ALIVE",
-    "42": "NOT_ALIVE",
-    "43": "NOT_ALIVE",
-    "44": "NOT_ALIVE",
-    "45": "NON_PARTICIPATING",
-    "46": "NON_PARTICIPATING",
-    "47": "NON_PARTICIPATING",
+    "42": "FAIL",
+    "43": "NON_PARTICIPATING",
+    "44": "FAIL",
+    "45": "NOT_ALIVE",
+    "46": "NOT_ALIVE",
+    "47": "NOT_ALIVE",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
-    "5": "NOT_ALIVE",
-    "50": "NON_PARTICIPATING",
+    "5": "NON_PARTICIPATING",
+    "50": "FAIL",
     "51": "NON_PARTICIPATING",
     "52": "FAIL",
-    "53": "FAIL",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
-    "55": "NON_PARTICIPATING",
-    "56": "NOT_ALIVE",
-    "57": "FAIL",
+    "55": "NOT_ALIVE",
+    "56": "FAIL",
+    "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "FAIL",
-    "61": "NON_PARTICIPATING",
+    "60": "NON_PARTICIPATING",
+    "61": "FAIL",
     "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
-    "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
-    "66": "NOT_ALIVE",
-    "67": "NOT_ALIVE",
+    "63": "FAIL",
+    "64": "FAIL",
+    "65": "NOT_ALIVE",
+    "66": "NON_PARTICIPATING",
+    "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "NOT_ALIVE",
-    "7": "FAIL",
-    "70": "NOT_ALIVE",
+    "69": "FAIL",
+    "7": "NOT_ALIVE",
+    "70": "FAIL",
     "71": "NOT_ALIVE",
     "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NON_PARTICIPATING",
+    "75": "NOT_ALIVE",
     "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
-    "79": "NOT_ALIVE",
+    "79": "FAIL",
     "8": "NON_PARTICIPATING",
-    "80": "FAIL",
+    "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
-    "84": "FAIL",
+    "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
-    "87": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
+    "87": "FAIL",
     "88": "FAIL",
-    "89": "FAIL",
+    "89": "NON_PARTICIPATING",
     "9": "NON_PARTICIPATING",
     "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
-    "92": "NOT_ALIVE",
+    "92": "FAIL",
     "93": "NOT_ALIVE",
-    "94": "FAIL",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "FAIL",
-    "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
+    "97": "FAIL",
+    "98": "NOT_ALIVE",
     "99": "FAIL"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6458800,
-    6458815
+    6461996,
+    6462012
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 144,
+  "inner_step": 99,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6457631,
+  "last_allreduce_block": 6461450,
   "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 sha256:175131ec8d010a7e361cce0c04f28aed81ddc1434bf038f6d98f974453db7d5b
+oid sha256:5efb95b49fc11969f0c4370b6838f7a1f8be4f1aece7cb4b45142473dad1ee77
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2426979b93298b3f01a32d7575b15f0f4a0cda458ab80164964fb627ac57187
+oid sha256:841cecbec5959e258eac6d10ded23044e58564ac85004a7b05e82a8054383a79
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:248dd204071715df372adaee09d25f934293c6405ce531e6c613513b80bc5b0b
+oid sha256:d4e7fa1000c4c3025bc2cead275bc3a30ed272769ab241f9fd74a0537b2d0a1b
 size 4400216536