Run 5. Outer Step 136. Inner Step 1.

config.json

@@ -2,143 +2,143 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
+    "101": "NOT_ALIVE",
     "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
-    "106": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
+    "106": "NON_PARTICIPATING",
     "107": "NOT_ALIVE",
-    "108": "NON_PARTICIPATING",
-    "109": "NOT_ALIVE",
+    "108": "NOT_ALIVE",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
-    "111": "NOT_ALIVE",
-    "112": "NON_PARTICIPATING",
+    "111": "NON_PARTICIPATING",
+    "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NON_PARTICIPATING",
     "115": "NOT_ALIVE",
     "116": "NON_PARTICIPATING",
     "117": "SUCCESS",
-    "118": "NOT_ALIVE",
+    "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
     "121": "NOT_ALIVE",
-    "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
+    "123": "NON_PARTICIPATING",
     "124": "SUCCESS",
     "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
     "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
-    "132": "NOT_ALIVE",
+    "132": "NON_PARTICIPATING",
     "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
     "135": "NON_PARTICIPATING",
-    "136": "NON_PARTICIPATING",
+    "136": "SUCCESS",
     "137": "NON_PARTICIPATING",
     "138": "SUCCESS",
     "139": "NON_PARTICIPATING",
-    "14": "SUCCESS",
-    "140": "NOT_ALIVE",
-    "141": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
+    "140": "NON_PARTICIPATING",
+    "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
-    "143": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "NOT_ALIVE",
+    "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
-    "149": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
     "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
-    "154": "SUCCESS",
+    "153": "SUCCESS",
+    "154": "NON_PARTICIPATING",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NON_PARTICIPATING",
-    "160": "NOT_ALIVE",
-    "161": "NON_PARTICIPATING",
+    "160": "NON_PARTICIPATING",
+    "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
-    "164": "SUCCESS",
-    "165": "SUCCESS",
-    "166": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
+    "165": "NOT_ALIVE",
+    "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
     "168": "SUCCESS",
     "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
-    "170": "NOT_ALIVE",
+    "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
-    "172": "NON_PARTICIPATING",
-    "173": "NON_PARTICIPATING",
-    "174": "NOT_ALIVE",
+    "172": "NOT_ALIVE",
+    "173": "NOT_ALIVE",
+    "174": "NON_PARTICIPATING",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "NON_PARTICIPATING",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
+    "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "NOT_ALIVE",
     "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
+    "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
     "191": "NON_PARTICIPATING",
-    "192": "NOT_ALIVE",
+    "192": "NON_PARTICIPATING",
     "193": "NOT_ALIVE",
-    "194": "NON_PARTICIPATING",
+    "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
-    "196": "NON_PARTICIPATING",
-    "197": "NOT_ALIVE",
+    "196": "NOT_ALIVE",
+    "197": "NON_PARTICIPATING",
     "198": "NON_PARTICIPATING",
     "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
-    "20": "NOT_ALIVE",
+    "20": "NON_PARTICIPATING",
     "200": "NOT_ALIVE",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
-    "204": "NON_PARTICIPATING",
-    "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "204": "SUCCESS",
+    "205": "NON_PARTICIPATING",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
-    "209": "NOT_ALIVE",
+    "208": "NON_PARTICIPATING",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "NOT_ALIVE",
-    "211": "NOT_ALIVE",
-    "212": "SUCCESS",
-    "213": "SUCCESS",
+    "210": "NON_PARTICIPATING",
+    "211": "NON_PARTICIPATING",
+    "212": "NOT_ALIVE",
+    "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
-    "215": "SUCCESS",
+    "215": "NON_PARTICIPATING",
     "216": "NOT_ALIVE",
-    "217": "NON_PARTICIPATING",
+    "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
     "219": "NOT_ALIVE",
     "22": "NON_PARTICIPATING",
-    "220": "NOT_ALIVE",
-    "221": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
+    "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
@@ -146,76 +146,76 @@
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
-    "230": "NOT_ALIVE",
+    "230": "NON_PARTICIPATING",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "NOT_ALIVE",
     "235": "NON_PARTICIPATING",
     "236": "NOT_ALIVE",
-    "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
+    "237": "NOT_ALIVE",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
-    "241": "NON_PARTICIPATING",
+    "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
-    "243": "NOT_ALIVE",
+    "243": "NON_PARTICIPATING",
     "244": "NOT_ALIVE",
-    "245": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
-    "249": "NOT_ALIVE",
-    "25": "SUCCESS",
+    "249": "NON_PARTICIPATING",
+    "25": "NON_PARTICIPATING",
     "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
-    "254": "NON_PARTICIPATING",
+    "254": "SUCCESS",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NON_PARTICIPATING",
     "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
+    "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
     "32": "SUCCESS",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NON_PARTICIPATING",
-    "36": "NOT_ALIVE",
+    "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
     "38": "NON_PARTICIPATING",
     "39": "SUCCESS",
     "4": "NON_PARTICIPATING",
     "40": "NON_PARTICIPATING",
-    "41": "NOT_ALIVE",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
-    "44": "NOT_ALIVE",
-    "45": "NOT_ALIVE",
+    "44": "NON_PARTICIPATING",
+    "45": "NON_PARTICIPATING",
     "46": "NOT_ALIVE",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "SUCCESS",
-    "50": "NOT_ALIVE",
+    "50": "NON_PARTICIPATING",
     "51": "NOT_ALIVE",
     "52": "SUCCESS",
-    "53": "NOT_ALIVE",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
     "57": "NON_PARTICIPATING",
     "58": "NON_PARTICIPATING",
     "59": "NON_PARTICIPATING",
     "6": "NON_PARTICIPATING",
     "60": "NOT_ALIVE",
-    "61": "NOT_ALIVE",
+    "61": "NON_PARTICIPATING",
     "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
@@ -227,34 +227,34 @@
     "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
-    "72": "NOT_ALIVE",
-    "73": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
+    "73": "NON_PARTICIPATING",
     "74": "NOT_ALIVE",
     "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
-    "77": "NON_PARTICIPATING",
+    "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NON_PARTICIPATING",
     "8": "NOT_ALIVE",
-    "80": "NOT_ALIVE",
-    "81": "NOT_ALIVE",
+    "80": "NON_PARTICIPATING",
+    "81": "NON_PARTICIPATING",
     "82": "NON_PARTICIPATING",
     "83": "NON_PARTICIPATING",
-    "84": "NOT_ALIVE",
-    "85": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
+    "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
-    "87": "NOT_ALIVE",
+    "87": "NON_PARTICIPATING",
     "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
-    "9": "NOT_ALIVE",
+    "9": "NON_PARTICIPATING",
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
-    "92": "NON_PARTICIPATING",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
-    "97": "NOT_ALIVE",
+    "97": "NON_PARTICIPATING",
     "98": "NOT_ALIVE",
     "99": "NOT_ALIVE"
   },
@@ -264,20 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6008739,
-    6008743,
-    6008748,
-    6008752,
-    6008756
+    6008831,
+    6008835
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 361,
+  "inner_step": 1,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6005880,
+  "last_allreduce_block": 6008781,
   "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 sha256:707f922f55f7dbafac2c23ed0c60602451fef3806e7f4d5055fd375c1796abcf
+oid sha256:f857b2f88b15edfc5de8fddf44bd5db752c9712baaecd9db7cc38ce253ab892f
 size 44586379

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fcb61230092fca02e2cd12ee2d252831936f940662a47f861f2c7913989f3bfa
+oid sha256:d3b3733f8451331a8d14ebec1321fe8ae1b8cdae7b8f3bef4ddb41b287fa989d
 size 8800564561

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e08870e9327c555239af34b20868c392d45208e0f2f0fe7791932b0056f5673
+oid sha256:6085257f9e9042ea72b9d59b3306fa723f8f250f80ddad611f0bf45863973230
 size 4400216536