Run 8. Outer Step 93. Inner Step 5.

config.json

@@ -1,152 +1,152 @@
 {
-  "_name_or_path": "zeras141a/a6",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
-    "100": "NON_PARTICIPATING",
+    "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
-    "102": "NON_PARTICIPATING",
+    "102": "NOT_ALIVE",
     "103": "SUCCESS",
-    "104": "SUCCESS",
-    "105": "SUCCESS",
-    "106": "FAIL",
+    "104": "NOT_ALIVE",
+    "105": "NOT_ALIVE",
+    "106": "SUCCESS",
     "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
-    "109": "NOT_ALIVE",
-    "11": "SUCCESS",
-    "110": "SUCCESS",
+    "109": "NON_PARTICIPATING",
+    "11": "NOT_ALIVE",
+    "110": "NOT_ALIVE",
     "111": "NOT_ALIVE",
     "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "NOT_ALIVE",
-    "116": "SUCCESS",
+    "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NON_PARTICIPATING",
-    "119": "NON_PARTICIPATING",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
-    "120": "FAIL",
+    "120": "NOT_ALIVE",
     "121": "NON_PARTICIPATING",
     "122": "SUCCESS",
     "123": "SUCCESS",
     "124": "SUCCESS",
     "125": "NOT_ALIVE",
     "126": "SUCCESS",
-    "127": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
-    "129": "SUCCESS",
-    "13": "FAIL",
-    "130": "NON_PARTICIPATING",
+    "129": "NON_PARTICIPATING",
+    "13": "NOT_ALIVE",
+    "130": "NOT_ALIVE",
     "131": "NON_PARTICIPATING",
     "132": "SUCCESS",
     "133": "SUCCESS",
-    "134": "SUCCESS",
+    "134": "NOT_ALIVE",
     "135": "SUCCESS",
-    "136": "NON_PARTICIPATING",
-    "137": "FAIL",
+    "136": "NOT_ALIVE",
+    "137": "NOT_ALIVE",
     "138": "SUCCESS",
-    "139": "SUCCESS",
-    "14": "NOT_ALIVE",
+    "139": "NOT_ALIVE",
+    "14": "SUCCESS",
     "140": "SUCCESS",
-    "141": "NON_PARTICIPATING",
-    "142": "FAIL",
+    "141": "NOT_ALIVE",
+    "142": "NOT_ALIVE",
     "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
-    "145": "SUCCESS",
+    "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
+    "149": "SUCCESS",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NON_PARTICIPATING",
-    "152": "SUCCESS",
+    "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
-    "154": "SUCCESS",
+    "154": "NOT_ALIVE",
     "155": "SUCCESS",
-    "156": "NOT_ALIVE",
+    "156": "SUCCESS",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
     "159": "SUCCESS",
-    "16": "NOT_ALIVE",
+    "16": "SUCCESS",
     "160": "SUCCESS",
     "161": "SUCCESS",
-    "162": "FAIL",
+    "162": "NOT_ALIVE",
     "163": "NON_PARTICIPATING",
-    "164": "SUCCESS",
-    "165": "SUCCESS",
+    "164": "NOT_ALIVE",
+    "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
-    "167": "FAIL",
-    "168": "FAIL",
-    "169": "SUCCESS",
+    "167": "NOT_ALIVE",
+    "168": "SUCCESS",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
-    "170": "FAIL",
+    "170": "SUCCESS",
     "171": "NOT_ALIVE",
-    "172": "FAIL",
-    "173": "NON_PARTICIPATING",
-    "174": "SUCCESS",
-    "175": "SUCCESS",
+    "172": "NOT_ALIVE",
+    "173": "SUCCESS",
+    "174": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
-    "178": "NON_PARTICIPATING",
-    "179": "FAIL",
-    "18": "SUCCESS",
-    "180": "SUCCESS",
-    "181": "FAIL",
+    "178": "NOT_ALIVE",
+    "179": "NOT_ALIVE",
+    "18": "NOT_ALIVE",
+    "180": "NON_PARTICIPATING",
+    "181": "NON_PARTICIPATING",
     "182": "NOT_ALIVE",
     "183": "SUCCESS",
     "184": "SUCCESS",
-    "185": "FAIL",
+    "185": "NON_PARTICIPATING",
     "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
-    "188": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
+    "188": "SUCCESS",
     "189": "SUCCESS",
-    "19": "SUCCESS",
-    "190": "SUCCESS",
+    "19": "NOT_ALIVE",
+    "190": "NOT_ALIVE",
     "191": "SUCCESS",
-    "192": "NON_PARTICIPATING",
+    "192": "NOT_ALIVE",
     "193": "NON_PARTICIPATING",
     "194": "SUCCESS",
     "195": "NON_PARTICIPATING",
-    "196": "NON_PARTICIPATING",
-    "197": "NON_PARTICIPATING",
-    "198": "FAIL",
+    "196": "NOT_ALIVE",
+    "197": "NOT_ALIVE",
+    "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
     "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
-    "203": "SUCCESS",
-    "204": "FAIL",
-    "205": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
+    "204": "NOT_ALIVE",
+    "205": "NOT_ALIVE",
     "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "SUCCESS",
     "209": "NON_PARTICIPATING",
-    "21": "NOT_ALIVE",
+    "21": "SUCCESS",
     "210": "SUCCESS",
-    "211": "NOT_ALIVE",
+    "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "SUCCESS",
+    "215": "NOT_ALIVE",
     "216": "SUCCESS",
-    "217": "FAIL",
-    "218": "SUCCESS",
+    "217": "NOT_ALIVE",
+    "218": "FAIL",
     "219": "SUCCESS",
     "22": "SUCCESS",
     "220": "SUCCESS",
-    "221": "NON_PARTICIPATING",
+    "221": "NOT_ALIVE",
     "222": "SUCCESS",
-    "223": "NON_PARTICIPATING",
+    "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "SUCCESS",
-    "226": "SUCCESS",
-    "227": "NON_PARTICIPATING",
-    "228": "NON_PARTICIPATING",
-    "229": "SUCCESS",
+    "226": "NOT_ALIVE",
+    "227": "NOT_ALIVE",
+    "228": "SUCCESS",
+    "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
@@ -154,108 +154,108 @@
     "233": "NON_PARTICIPATING",
     "234": "SUCCESS",
     "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
-    "237": "NOT_ALIVE",
-    "238": "NOT_ALIVE",
+    "236": "SUCCESS",
+    "237": "SUCCESS",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
-    "240": "NON_PARTICIPATING",
+    "24": "SUCCESS",
+    "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
-    "242": "SUCCESS",
-    "243": "SUCCESS",
-    "244": "SUCCESS",
+    "242": "NOT_ALIVE",
+    "243": "NON_PARTICIPATING",
+    "244": "NON_PARTICIPATING",
     "245": "SUCCESS",
-    "246": "FAIL",
-    "247": "NON_PARTICIPATING",
-    "248": "NOT_ALIVE",
+    "246": "NOT_ALIVE",
+    "247": "SUCCESS",
+    "248": "SUCCESS",
     "249": "SUCCESS",
     "25": "SUCCESS",
     "250": "SUCCESS",
     "251": "SUCCESS",
-    "252": "SUCCESS",
-    "253": "NOT_ALIVE",
-    "254": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
+    "253": "SUCCESS",
+    "254": "NOT_ALIVE",
     "255": "SUCCESS",
-    "26": "SUCCESS",
+    "26": "NOT_ALIVE",
     "27": "SUCCESS",
-    "28": "NON_PARTICIPATING",
-    "29": "NON_PARTICIPATING",
+    "28": "NOT_ALIVE",
+    "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "SUCCESS",
-    "33": "NON_PARTICIPATING",
+    "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
-    "35": "SUCCESS",
-    "36": "SUCCESS",
+    "35": "NOT_ALIVE",
+    "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
-    "38": "NON_PARTICIPATING",
-    "39": "NOT_ALIVE",
-    "4": "SUCCESS",
-    "40": "NON_PARTICIPATING",
-    "41": "FAIL",
-    "42": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
+    "39": "NON_PARTICIPATING",
+    "4": "NOT_ALIVE",
+    "40": "NOT_ALIVE",
+    "41": "SUCCESS",
+    "42": "NOT_ALIVE",
     "43": "SUCCESS",
     "44": "SUCCESS",
     "45": "NOT_ALIVE",
-    "46": "NOT_ALIVE",
-    "47": "SUCCESS",
+    "46": "NON_PARTICIPATING",
+    "47": "NON_PARTICIPATING",
     "48": "SUCCESS",
-    "49": "FAIL",
+    "49": "NON_PARTICIPATING",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "SUCCESS",
+    "51": "NOT_ALIVE",
     "52": "SUCCESS",
-    "53": "FAIL",
+    "53": "NOT_ALIVE",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "FAIL",
-    "57": "SUCCESS",
-    "58": "NOT_ALIVE",
+    "56": "SUCCESS",
+    "57": "NON_PARTICIPATING",
+    "58": "SUCCESS",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "SUCCESS",
-    "61": "SUCCESS",
-    "62": "SUCCESS",
-    "63": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
+    "61": "NOT_ALIVE",
+    "62": "NOT_ALIVE",
+    "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "FAIL",
+    "65": "SUCCESS",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
     "68": "NON_PARTICIPATING",
-    "69": "NON_PARTICIPATING",
+    "69": "SUCCESS",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "SUCCESS",
-    "72": "NON_PARTICIPATING",
+    "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
-    "74": "SUCCESS",
+    "74": "FAIL",
     "75": "NOT_ALIVE",
     "76": "SUCCESS",
-    "77": "SUCCESS",
-    "78": "NON_PARTICIPATING",
+    "77": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
     "79": "NOT_ALIVE",
-    "8": "SUCCESS",
+    "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
-    "82": "NON_PARTICIPATING",
-    "83": "NOT_ALIVE",
-    "84": "NON_PARTICIPATING",
-    "85": "SUCCESS",
-    "86": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
+    "83": "NON_PARTICIPATING",
+    "84": "SUCCESS",
+    "85": "NOT_ALIVE",
+    "86": "NOT_ALIVE",
     "87": "SUCCESS",
-    "88": "NOT_ALIVE",
+    "88": "SUCCESS",
     "89": "NOT_ALIVE",
-    "9": "FAIL",
+    "9": "SUCCESS",
     "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
-    "92": "NON_PARTICIPATING",
+    "91": "NOT_ALIVE",
+    "92": "NOT_ALIVE",
     "93": "SUCCESS",
-    "94": "SUCCESS",
-    "95": "FAIL",
+    "94": "NON_PARTICIPATING",
+    "95": "NOT_ALIVE",
     "96": "SUCCESS",
     "97": "NOT_ALIVE",
-    "98": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6409397,
-    6409412
+    6411616,
+    6411632
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 63,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6407463,
+  "last_allreduce_block": 6411158,
   "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 sha256:2e1d07a637fe566f19d0274450865ce8eabc01f851ab21bf9131995a7fbf1293
+oid sha256:efdc4f4c9becab3ef45f3d0314ac5a0bc74b55e73c94e8242ea98b0ceefa2a4f
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa59e6a8e64dbe0802d76e0bcebcba5e67dd208c7d64e35489201a247b7f66b0
+oid sha256:222efb0441b92c6fa4a2730cea72c5cde7136f657a5b5c0b8bcf08ac1025c95d
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6dfb87e3466f2bce691f867967b2b605db778ad09ae81dca02a4601ece9a89dd
+oid sha256:8dbd56db586bd60d3bf46d2907c842dbb0e6b57a36102c7c98650f56bdc5142e
 size 4400216536