Run 9. Outer Step 14. Inner Step 9.

config.json

@@ -2,7 +2,7 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
     "10": "FAIL",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
@@ -10,44 +10,44 @@
     "103": "NON_PARTICIPATING",
     "104": "FAIL",
     "105": "FAIL",
-    "106": "FAIL",
-    "107": "NON_PARTICIPATING",
+    "106": "NOT_ALIVE",
+    "107": "NOT_ALIVE",
     "108": "NOT_ALIVE",
-    "109": "NOT_ALIVE",
+    "109": "FAIL",
     "11": "NON_PARTICIPATING",
-    "110": "FAIL",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "FAIL",
-    "113": "FAIL",
+    "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
-    "118": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
-    "120": "NOT_ALIVE",
+    "120": "FAIL",
     "121": "NOT_ALIVE",
-    "122": "NON_PARTICIPATING",
-    "123": "NOT_ALIVE",
+    "122": "FAIL",
+    "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
-    "125": "NOT_ALIVE",
-    "126": "NON_PARTICIPATING",
+    "125": "NON_PARTICIPATING",
+    "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
-    "128": "FAIL",
+    "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "FAIL",
-    "130": "NON_PARTICIPATING",
+    "130": "FAIL",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
-    "135": "NOT_ALIVE",
+    "135": "FAIL",
     "136": "NOT_ALIVE",
-    "137": "NOT_ALIVE",
-    "138": "NOT_ALIVE",
+    "137": "NON_PARTICIPATING",
+    "138": "FAIL",
     "139": "NOT_ALIVE",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
@@ -56,134 +56,134 @@
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "NOT_ALIVE",
-    "149": "NON_PARTICIPATING",
+    "148": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NOT_ALIVE",
-    "153": "FAIL",
-    "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
-    "156": "NON_PARTICIPATING",
-    "157": "NON_PARTICIPATING",
+    "152": "NON_PARTICIPATING",
+    "153": "NON_PARTICIPATING",
+    "154": "FAIL",
+    "155": "NON_PARTICIPATING",
+    "156": "FAIL",
+    "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "FAIL",
-    "164": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
-    "167": "NON_PARTICIPATING",
+    "167": "NOT_ALIVE",
     "168": "NON_PARTICIPATING",
     "169": "FAIL",
-    "17": "FAIL",
+    "17": "NON_PARTICIPATING",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
-    "174": "NOT_ALIVE",
+    "173": "FAIL",
+    "174": "NON_PARTICIPATING",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "FAIL",
     "178": "NOT_ALIVE",
     "179": "NON_PARTICIPATING",
-    "18": "FAIL",
-    "180": "NON_PARTICIPATING",
+    "18": "NON_PARTICIPATING",
+    "180": "FAIL",
     "181": "NON_PARTICIPATING",
     "182": "NOT_ALIVE",
-    "183": "NON_PARTICIPATING",
-    "184": "FAIL",
+    "183": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
     "185": "NOT_ALIVE",
     "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
-    "189": "NON_PARTICIPATING",
+    "189": "FAIL",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
     "191": "NON_PARTICIPATING",
     "192": "NON_PARTICIPATING",
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
-    "196": "FAIL",
-    "197": "NON_PARTICIPATING",
-    "198": "FAIL",
-    "199": "NON_PARTICIPATING",
-    "2": "NOT_ALIVE",
-    "20": "FAIL",
+    "195": "FAIL",
+    "196": "NON_PARTICIPATING",
+    "197": "NOT_ALIVE",
+    "198": "NOT_ALIVE",
+    "199": "FAIL",
+    "2": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
     "200": "NON_PARTICIPATING",
     "201": "FAIL",
-    "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "202": "FAIL",
+    "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "FAIL",
-    "206": "NON_PARTICIPATING",
+    "205": "NOT_ALIVE",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "FAIL",
-    "209": "FAIL",
+    "209": "NON_PARTICIPATING",
     "21": "FAIL",
-    "210": "NON_PARTICIPATING",
+    "210": "NOT_ALIVE",
     "211": "FAIL",
-    "212": "NON_PARTICIPATING",
+    "212": "FAIL",
     "213": "NOT_ALIVE",
-    "214": "FAIL",
-    "215": "NON_PARTICIPATING",
-    "216": "NOT_ALIVE",
+    "214": "NON_PARTICIPATING",
+    "215": "FAIL",
+    "216": "FAIL",
     "217": "NON_PARTICIPATING",
-    "218": "NOT_ALIVE",
-    "219": "FAIL",
+    "218": "NON_PARTICIPATING",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "FAIL",
-    "221": "FAIL",
-    "222": "FAIL",
+    "220": "NON_PARTICIPATING",
+    "221": "NON_PARTICIPATING",
+    "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "FAIL",
     "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "227": "FAIL",
     "228": "NON_PARTICIPATING",
-    "229": "NOT_ALIVE",
+    "229": "NON_PARTICIPATING",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
-    "233": "FAIL",
-    "234": "FAIL",
+    "233": "NON_PARTICIPATING",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
-    "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
+    "236": "NON_PARTICIPATING",
+    "237": "FAIL",
+    "238": "FAIL",
     "239": "NON_PARTICIPATING",
     "24": "FAIL",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
-    "242": "NOT_ALIVE",
+    "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
-    "246": "FAIL",
+    "245": "NON_PARTICIPATING",
+    "246": "NON_PARTICIPATING",
     "247": "FAIL",
-    "248": "NOT_ALIVE",
-    "249": "NOT_ALIVE",
+    "248": "NON_PARTICIPATING",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "FAIL",
+    "250": "NON_PARTICIPATING",
     "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
-    "254": "NON_PARTICIPATING",
-    "255": "NON_PARTICIPATING",
-    "26": "NOT_ALIVE",
-    "27": "FAIL",
+    "254": "NOT_ALIVE",
+    "255": "FAIL",
+    "26": "NON_PARTICIPATING",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "FAIL",
-    "3": "NON_PARTICIPATING",
-    "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
-    "32": "NOT_ALIVE",
+    "3": "NOT_ALIVE",
+    "30": "FAIL",
+    "31": "FAIL",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
     "35": "NON_PARTICIPATING",
@@ -194,68 +194,68 @@
     "4": "NON_PARTICIPATING",
     "40": "FAIL",
     "41": "NOT_ALIVE",
-    "42": "FAIL",
-    "43": "NON_PARTICIPATING",
+    "42": "NON_PARTICIPATING",
+    "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
-    "45": "FAIL",
-    "46": "FAIL",
-    "47": "FAIL",
+    "45": "NON_PARTICIPATING",
+    "46": "NOT_ALIVE",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
-    "5": "NON_PARTICIPATING",
-    "50": "FAIL",
-    "51": "FAIL",
+    "5": "FAIL",
+    "50": "NON_PARTICIPATING",
+    "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
     "53": "FAIL",
-    "54": "NOT_ALIVE",
+    "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
-    "56": "FAIL",
-    "57": "NON_PARTICIPATING",
+    "56": "NOT_ALIVE",
+    "57": "FAIL",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
-    "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "59": "NOT_ALIVE",
+    "6": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
+    "61": "FAIL",
     "62": "FAIL",
     "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
     "65": "FAIL",
-    "66": "NON_PARTICIPATING",
-    "67": "NOT_ALIVE",
+    "66": "FAIL",
+    "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "NON_PARTICIPATING",
+    "69": "FAIL",
     "7": "FAIL",
     "70": "FAIL",
     "71": "NOT_ALIVE",
-    "72": "NON_PARTICIPATING",
-    "73": "NON_PARTICIPATING",
-    "74": "NON_PARTICIPATING",
+    "72": "FAIL",
+    "73": "NOT_ALIVE",
+    "74": "FAIL",
     "75": "NON_PARTICIPATING",
     "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
+    "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
-    "8": "NON_PARTICIPATING",
-    "80": "FAIL",
+    "8": "NOT_ALIVE",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
-    "84": "FAIL",
+    "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "FAIL",
+    "86": "NON_PARTICIPATING",
     "87": "NON_PARTICIPATING",
     "88": "FAIL",
-    "89": "NON_PARTICIPATING",
-    "9": "NON_PARTICIPATING",
-    "90": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
+    "9": "FAIL",
+    "90": "FAIL",
     "91": "FAIL",
-    "92": "NOT_ALIVE",
-    "93": "NON_PARTICIPATING",
+    "92": "FAIL",
+    "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
-    "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "95": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
     "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6458100,
-    6458137
+    6458270,
+    6458303
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 67,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6456970,
+  "last_allreduce_block": 6458233,
   "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 sha256:5f699595eb6e21b208a72f8fa5ea9a5319a13504c4e34d2e2307fcd323caf97b
+oid sha256:02b1e79441894a42687601e72f7a490a41f219984fee02244c1aeaf54cfcd0a6
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc67560963dbaa96a130b3dbd9a8d6fba555712ff521e5ff4c0cfc9a6f2e2010
+oid sha256:0ac493a30b7bef530c4be2cd90ce4c768a9c78b207dcf04c8ec63a468d781622
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7241094692f529128e2bd44e5eebba9bc513f1c398fc95250f909786a8e754b5
+oid sha256:087e50a5942712af7de76d600b1a64d495498a3b0879660a0d42d8575c6744fc
 size 4400216536