Run 8. Outer Step 96. Inner Step 5.

config.json

@@ -1,261 +1,261 @@
 {
-  "_name_or_path": "zeras141a/tata",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
-    "102": "NON_PARTICIPATING",
-    "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
-    "106": "SUCCESS",
+    "101": "NOT_ALIVE",
+    "102": "NOT_ALIVE",
+    "103": "NON_PARTICIPATING",
+    "104": "NON_PARTICIPATING",
+    "105": "NON_PARTICIPATING",
+    "106": "NON_PARTICIPATING",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
-    "109": "SUCCESS",
+    "109": "FAIL",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "110": "NON_PARTICIPATING",
     "111": "NOT_ALIVE",
-    "112": "NON_PARTICIPATING",
-    "113": "NON_PARTICIPATING",
-    "114": "NON_PARTICIPATING",
+    "112": "NOT_ALIVE",
+    "113": "SUCCESS",
+    "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
-    "116": "NOT_ALIVE",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
+    "120": "FAIL",
     "121": "NON_PARTICIPATING",
-    "122": "NON_PARTICIPATING",
-    "123": "NOT_ALIVE",
-    "124": "SUCCESS",
-    "125": "NOT_ALIVE",
+    "122": "NOT_ALIVE",
+    "123": "NON_PARTICIPATING",
+    "124": "FAIL",
+    "125": "NON_PARTICIPATING",
     "126": "SUCCESS",
     "127": "NOT_ALIVE",
-    "128": "NON_PARTICIPATING",
-    "129": "SUCCESS",
+    "128": "NOT_ALIVE",
+    "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
-    "130": "NON_PARTICIPATING",
-    "131": "NON_PARTICIPATING",
-    "132": "SUCCESS",
-    "133": "SUCCESS",
+    "130": "SUCCESS",
+    "131": "NOT_ALIVE",
+    "132": "NON_PARTICIPATING",
+    "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
-    "135": "SUCCESS",
-    "136": "NON_PARTICIPATING",
-    "137": "NOT_ALIVE",
+    "135": "NON_PARTICIPATING",
+    "136": "NOT_ALIVE",
+    "137": "SUCCESS",
     "138": "NOT_ALIVE",
-    "139": "SUCCESS",
-    "14": "NOT_ALIVE",
-    "140": "SUCCESS",
-    "141": "NON_PARTICIPATING",
-    "142": "NOT_ALIVE",
+    "139": "NOT_ALIVE",
+    "14": "SUCCESS",
+    "140": "NOT_ALIVE",
+    "141": "NOT_ALIVE",
+    "142": "SUCCESS",
     "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
-    "145": "SUCCESS",
+    "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "NON_PARTICIPATING",
-    "148": "NOT_ALIVE",
+    "147": "SUCCESS",
+    "148": "SUCCESS",
     "149": "SUCCESS",
-    "15": "NOT_ALIVE",
+    "15": "NON_PARTICIPATING",
     "150": "NOT_ALIVE",
-    "151": "NON_PARTICIPATING",
-    "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
-    "154": "SUCCESS",
+    "151": "NOT_ALIVE",
+    "152": "NON_PARTICIPATING",
+    "153": "NON_PARTICIPATING",
+    "154": "NOT_ALIVE",
     "155": "SUCCESS",
-    "156": "SUCCESS",
-    "157": "NON_PARTICIPATING",
-    "158": "NON_PARTICIPATING",
+    "156": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
+    "158": "SUCCESS",
     "159": "SUCCESS",
-    "16": "NOT_ALIVE",
+    "16": "SUCCESS",
     "160": "SUCCESS",
     "161": "SUCCESS",
-    "162": "NON_PARTICIPATING",
-    "163": "NOT_ALIVE",
-    "164": "NON_PARTICIPATING",
-    "165": "NON_PARTICIPATING",
+    "162": "SUCCESS",
+    "163": "SUCCESS",
+    "164": "NOT_ALIVE",
+    "165": "SUCCESS",
     "166": "NOT_ALIVE",
-    "167": "NOT_ALIVE",
-    "168": "SUCCESS",
-    "169": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
+    "168": "NON_PARTICIPATING",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
-    "170": "SUCCESS",
+    "170": "FAIL",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "SUCCESS",
-    "174": "NON_PARTICIPATING",
+    "173": "NON_PARTICIPATING",
+    "174": "NOT_ALIVE",
     "175": "SUCCESS",
     "176": "NOT_ALIVE",
-    "177": "SUCCESS",
-    "178": "NON_PARTICIPATING",
-    "179": "NON_PARTICIPATING",
+    "177": "NON_PARTICIPATING",
+    "178": "NOT_ALIVE",
+    "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
+    "180": "NOT_ALIVE",
     "181": "NON_PARTICIPATING",
-    "182": "NOT_ALIVE",
-    "183": "SUCCESS",
+    "182": "SUCCESS",
+    "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
-    "188": "SUCCESS",
-    "189": "NOT_ALIVE",
+    "185": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
+    "187": "NOT_ALIVE",
+    "188": "NON_PARTICIPATING",
+    "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "NOT_ALIVE",
-    "191": "SUCCESS",
-    "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
-    "194": "SUCCESS",
-    "195": "NON_PARTICIPATING",
-    "196": "NOT_ALIVE",
+    "190": "NON_PARTICIPATING",
+    "191": "NON_PARTICIPATING",
+    "192": "NOT_ALIVE",
+    "193": "NOT_ALIVE",
+    "194": "NOT_ALIVE",
+    "195": "NOT_ALIVE",
+    "196": "NON_PARTICIPATING",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
-    "199": "SUCCESS",
+    "199": "NON_PARTICIPATING",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
-    "203": "NOT_ALIVE",
-    "204": "SUCCESS",
-    "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "202": "NOT_ALIVE",
+    "203": "NON_PARTICIPATING",
+    "204": "FAIL",
+    "205": "SUCCESS",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "SUCCESS",
     "209": "NOT_ALIVE",
-    "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
-    "211": "SUCCESS",
+    "21": "NON_PARTICIPATING",
+    "210": "SUCCESS",
+    "211": "NON_PARTICIPATING",
     "212": "NOT_ALIVE",
     "213": "SUCCESS",
     "214": "NOT_ALIVE",
-    "215": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
     "216": "SUCCESS",
     "217": "SUCCESS",
-    "218": "SUCCESS",
+    "218": "NON_PARTICIPATING",
     "219": "NON_PARTICIPATING",
-    "22": "SUCCESS",
-    "220": "FAIL",
-    "221": "NOT_ALIVE",
+    "22": "NON_PARTICIPATING",
+    "220": "SUCCESS",
+    "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
-    "224": "NON_PARTICIPATING",
-    "225": "SUCCESS",
-    "226": "NOT_ALIVE",
-    "227": "SUCCESS",
+    "224": "NOT_ALIVE",
+    "225": "NON_PARTICIPATING",
+    "226": "NON_PARTICIPATING",
+    "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
-    "229": "NON_PARTICIPATING",
-    "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
-    "232": "NOT_ALIVE",
+    "229": "NOT_ALIVE",
+    "23": "SUCCESS",
+    "230": "SUCCESS",
+    "231": "NOT_ALIVE",
+    "232": "SUCCESS",
     "233": "NOT_ALIVE",
     "234": "SUCCESS",
-    "235": "NON_PARTICIPATING",
+    "235": "SUCCESS",
     "236": "NON_PARTICIPATING",
-    "237": "NOT_ALIVE",
-    "238": "SUCCESS",
-    "239": "NOT_ALIVE",
+    "237": "SUCCESS",
+    "238": "NON_PARTICIPATING",
+    "239": "NON_PARTICIPATING",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
-    "243": "NON_PARTICIPATING",
+    "242": "NOT_ALIVE",
+    "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
-    "246": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "SUCCESS",
-    "248": "SUCCESS",
-    "249": "NOT_ALIVE",
+    "248": "NON_PARTICIPATING",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "SUCCESS",
-    "251": "SUCCESS",
-    "252": "NON_PARTICIPATING",
+    "250": "FAIL",
+    "251": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "SUCCESS",
-    "254": "NON_PARTICIPATING",
-    "255": "SUCCESS",
+    "254": "NOT_ALIVE",
+    "255": "NON_PARTICIPATING",
     "26": "NOT_ALIVE",
-    "27": "SUCCESS",
-    "28": "NON_PARTICIPATING",
-    "29": "NOT_ALIVE",
-    "3": "NOT_ALIVE",
+    "27": "NON_PARTICIPATING",
+    "28": "NOT_ALIVE",
+    "29": "NON_PARTICIPATING",
+    "3": "NON_PARTICIPATING",
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
-    "33": "NON_PARTICIPATING",
-    "34": "NON_PARTICIPATING",
-    "35": "NOT_ALIVE",
-    "36": "NOT_ALIVE",
+    "33": "NOT_ALIVE",
+    "34": "NOT_ALIVE",
+    "35": "SUCCESS",
+    "36": "SUCCESS",
     "37": "NOT_ALIVE",
-    "38": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
     "39": "NOT_ALIVE",
-    "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
-    "41": "NOT_ALIVE",
-    "42": "SUCCESS",
-    "43": "SUCCESS",
-    "44": "SUCCESS",
-    "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
+    "4": "NON_PARTICIPATING",
+    "40": "NOT_ALIVE",
+    "41": "SUCCESS",
+    "42": "NOT_ALIVE",
+    "43": "FAIL",
+    "44": "NON_PARTICIPATING",
+    "45": "NON_PARTICIPATING",
+    "46": "NOT_ALIVE",
     "47": "NON_PARTICIPATING",
-    "48": "SUCCESS",
-    "49": "NON_PARTICIPATING",
+    "48": "NON_PARTICIPATING",
+    "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "FAIL",
-    "52": "NON_PARTICIPATING",
-    "53": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
+    "52": "NOT_ALIVE",
+    "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "SUCCESS",
+    "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
-    "61": "SUCCESS",
+    "60": "NON_PARTICIPATING",
+    "61": "NON_PARTICIPATING",
     "62": "SUCCESS",
-    "63": "NON_PARTICIPATING",
+    "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
-    "67": "NON_PARTICIPATING",
-    "68": "NON_PARTICIPATING",
-    "69": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
+    "68": "NOT_ALIVE",
+    "69": "SUCCESS",
     "7": "NOT_ALIVE",
-    "70": "NOT_ALIVE",
+    "70": "FAIL",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
-    "73": "NON_PARTICIPATING",
-    "74": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
+    "74": "FAIL",
     "75": "NOT_ALIVE",
-    "76": "SUCCESS",
+    "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
-    "81": "NON_PARTICIPATING",
+    "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "NON_PARTICIPATING",
-    "87": "NON_PARTICIPATING",
-    "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
-    "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
+    "87": "NOT_ALIVE",
+    "88": "SUCCESS",
+    "89": "NON_PARTICIPATING",
+    "9": "NON_PARTICIPATING",
+    "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
-    "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
-    "95": "SUCCESS",
-    "96": "NON_PARTICIPATING",
-    "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
+    "93": "SUCCESS",
+    "94": "SUCCESS",
+    "95": "NOT_ALIVE",
+    "96": "SUCCESS",
+    "97": "SUCCESS",
+    "98": "SUCCESS",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6418974,
-    6418983
+    6419050,
+    6419072
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 530,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6412880,
+  "last_allreduce_block": 6419009,
   "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 sha256:99a883e70735cd8b9491badfcb1ec31188147cb33a9a99a5ae9add371748aa35
+oid sha256:c04e984c99fda68623a1a523a56861f523ae401d32590a413ecdd90f9a2110bd
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1b8e08e0bfea18fae91b8c5ac3bb02b1d21ce8f74a119aec7a0d97102123bf5
+oid sha256:db836a55370e80f92febd244c4095feb343ea510c4b3b01af4e5d6c92c9f5166
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9aff07c26a4789821f6ee9e2748efa6a40b8bef34e389596d5626514c9fa7bd2
+oid sha256:063cbe77a0f8e8e2e9636f52a046a5db83e7ea04e694f75efd1e0231c2246d60
 size 4400216536