Run 8. Outer Step 111. Inner Step 5.

config.json

@@ -7,11 +7,11 @@
     "100": "NON_PARTICIPATING",
     "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
-    "103": "SUCCESS",
+    "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
+    "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
@@ -20,177 +20,177 @@
     "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NON_PARTICIPATING",
-    "115": "NON_PARTICIPATING",
+    "115": "NOT_ALIVE",
     "116": "NON_PARTICIPATING",
-    "117": "NON_PARTICIPATING",
-    "118": "NON_PARTICIPATING",
+    "117": "FAIL",
+    "118": "SUCCESS",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
-    "123": "NOT_ALIVE",
-    "124": "SUCCESS",
-    "125": "FAIL",
+    "123": "NON_PARTICIPATING",
+    "124": "NOT_ALIVE",
+    "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
-    "128": "SUCCESS",
+    "128": "NOT_ALIVE",
     "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NON_PARTICIPATING",
-    "131": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
+    "131": "NOT_ALIVE",
     "132": "NON_PARTICIPATING",
-    "133": "SUCCESS",
+    "133": "NOT_ALIVE",
     "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
     "139": "SUCCESS",
-    "14": "SUCCESS",
+    "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "NOT_ALIVE",
+    "146": "SUCCESS",
     "147": "SUCCESS",
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
-    "152": "NON_PARTICIPATING",
-    "153": "NON_PARTICIPATING",
+    "152": "NOT_ALIVE",
+    "153": "NOT_ALIVE",
     "154": "SUCCESS",
     "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
-    "158": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
+    "158": "SUCCESS",
     "159": "NON_PARTICIPATING",
     "16": "SUCCESS",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
-    "167": "FAIL",
+    "166": "NOT_ALIVE",
+    "167": "NOT_ALIVE",
     "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "SUCCESS",
-    "171": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
     "173": "SUCCESS",
     "174": "NON_PARTICIPATING",
     "175": "SUCCESS",
     "176": "NOT_ALIVE",
-    "177": "SUCCESS",
-    "178": "NON_PARTICIPATING",
+    "177": "FAIL",
+    "178": "SUCCESS",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NON_PARTICIPATING",
-    "182": "SUCCESS",
+    "182": "NON_PARTICIPATING",
     "183": "SUCCESS",
     "184": "SUCCESS",
     "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
-    "187": "FAIL",
+    "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
-    "189": "SUCCESS",
+    "189": "FAIL",
     "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
-    "192": "NON_PARTICIPATING",
+    "192": "NOT_ALIVE",
     "193": "NON_PARTICIPATING",
-    "194": "NON_PARTICIPATING",
+    "194": "SUCCESS",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "SUCCESS",
-    "199": "SUCCESS",
+    "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "NON_PARTICIPATING",
+    "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "SUCCESS",
     "204": "NON_PARTICIPATING",
     "205": "NOT_ALIVE",
     "206": "SUCCESS",
-    "207": "NON_PARTICIPATING",
+    "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
-    "209": "FAIL",
-    "21": "SUCCESS",
-    "210": "NON_PARTICIPATING",
+    "209": "NON_PARTICIPATING",
+    "21": "NOT_ALIVE",
+    "210": "FAIL",
     "211": "NOT_ALIVE",
-    "212": "NON_PARTICIPATING",
+    "212": "NOT_ALIVE",
     "213": "SUCCESS",
     "214": "SUCCESS",
     "215": "NON_PARTICIPATING",
-    "216": "FAIL",
+    "216": "NOT_ALIVE",
     "217": "SUCCESS",
-    "218": "SUCCESS",
-    "219": "NOT_ALIVE",
+    "218": "NON_PARTICIPATING",
+    "219": "NON_PARTICIPATING",
     "22": "SUCCESS",
     "220": "SUCCESS",
     "221": "NON_PARTICIPATING",
     "222": "NON_PARTICIPATING",
-    "223": "FAIL",
-    "224": "NON_PARTICIPATING",
+    "223": "NON_PARTICIPATING",
+    "224": "NOT_ALIVE",
     "225": "SUCCESS",
-    "226": "NON_PARTICIPATING",
-    "227": "SUCCESS",
+    "226": "NOT_ALIVE",
+    "227": "NOT_ALIVE",
     "228": "SUCCESS",
     "229": "NON_PARTICIPATING",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
-    "231": "SUCCESS",
+    "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
+    "233": "SUCCESS",
+    "234": "FAIL",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "NOT_ALIVE",
+    "237": "SUCCESS",
     "238": "SUCCESS",
-    "239": "NON_PARTICIPATING",
+    "239": "SUCCESS",
     "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
+    "240": "SUCCESS",
     "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
-    "243": "NON_PARTICIPATING",
-    "244": "NON_PARTICIPATING",
-    "245": "NON_PARTICIPATING",
-    "246": "SUCCESS",
+    "243": "NOT_ALIVE",
+    "244": "NOT_ALIVE",
+    "245": "SUCCESS",
+    "246": "NON_PARTICIPATING",
     "247": "SUCCESS",
     "248": "SUCCESS",
-    "249": "SUCCESS",
+    "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "SUCCESS",
+    "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
     "253": "SUCCESS",
     "254": "NOT_ALIVE",
-    "255": "NON_PARTICIPATING",
+    "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
-    "28": "NON_PARTICIPATING",
+    "27": "FAIL",
+    "28": "NOT_ALIVE",
     "29": "NON_PARTICIPATING",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NON_PARTICIPATING",
+    "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NON_PARTICIPATING",
     "34": "NON_PARTICIPATING",
     "35": "NON_PARTICIPATING",
     "36": "FAIL",
     "37": "NON_PARTICIPATING",
-    "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
+    "38": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
     "41": "NON_PARTICIPATING",
@@ -198,65 +198,65 @@
     "43": "SUCCESS",
     "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
-    "46": "SUCCESS",
+    "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
-    "49": "NOT_ALIVE",
+    "49": "NON_PARTICIPATING",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "SUCCESS",
-    "52": "NON_PARTICIPATING",
-    "53": "NON_PARTICIPATING",
-    "54": "NON_PARTICIPATING",
+    "51": "NON_PARTICIPATING",
+    "52": "SUCCESS",
+    "53": "SUCCESS",
+    "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NOT_ALIVE",
     "61": "SUCCESS",
-    "62": "NOT_ALIVE",
+    "62": "FAIL",
     "63": "NON_PARTICIPATING",
-    "64": "NOT_ALIVE",
+    "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
-    "68": "NON_PARTICIPATING",
+    "68": "SUCCESS",
     "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
-    "72": "NOT_ALIVE",
+    "71": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
     "74": "NOT_ALIVE",
-    "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
+    "75": "NON_PARTICIPATING",
+    "76": "FAIL",
     "77": "NON_PARTICIPATING",
-    "78": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "SUCCESS",
+    "80": "NON_PARTICIPATING",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NON_PARTICIPATING",
-    "86": "NON_PARTICIPATING",
-    "87": "FAIL",
+    "86": "SUCCESS",
+    "87": "NON_PARTICIPATING",
     "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "NOT_ALIVE",
+    "91": "SUCCESS",
     "92": "NOT_ALIVE",
     "93": "NON_PARTICIPATING",
     "94": "SUCCESS",
     "95": "NON_PARTICIPATING",
     "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
-    "99": "SUCCESS"
+    "98": "SUCCESS",
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6430712,
-    6430727
+    6430800,
+    6430826
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 75,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6430136,
+  "last_allreduce_block": 6430762,
   "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:cab51dec7417cd6b0a518c19fcf45f81b264e8164908f24e7c828751c93c5907
+oid sha256:efd5f6e9e73c307940a94f70fafb22ea353b40bf66ad1d7d1472283a972aea3e
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a34a21c18cebbcd14851f7554c7fc15f37a7a6c17f83c775d1a99807de7d6a87
+oid sha256:f48707647b7378abf88836e4f26c89539d5132bb6a1db86a314d51914d9338fe
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3e788195ce912e4c659c26b1b4a80cd8a9459571051909ad569ff091bef1e4ff
+oid sha256:02e037c99fe486a9c3ff893945f44c76705ce46d968782bb1e7623893d90a9d4
 size 4400216536