Run 11. Outer Step 184. Inner Step 9.

config.json

@@ -7,13 +7,13 @@
     "100": "NOT_ALIVE",
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
-    "103": "NON_PARTICIPATING",
+    "103": "NOT_ALIVE",
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
-    "108": "SUCCESS",
-    "109": "SUCCESS",
+    "108": "NOT_ALIVE",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
@@ -21,36 +21,36 @@
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
-    "116": "NOT_ALIVE",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NON_PARTICIPATING",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NON_PARTICIPATING",
-    "123": "SUCCESS",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
     "128": "NOT_ALIVE",
-    "129": "NOT_ALIVE",
-    "13": "NON_PARTICIPATING",
+    "129": "SUCCESS",
+    "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
-    "135": "SUCCESS",
+    "134": "NOT_ALIVE",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
-    "14": "NON_PARTICIPATING",
-    "140": "NON_PARTICIPATING",
-    "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
+    "139": "NOT_ALIVE",
+    "14": "NOT_ALIVE",
+    "140": "SUCCESS",
+    "141": "NOT_ALIVE",
+    "142": "SUCCESS",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
@@ -62,13 +62,13 @@
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
-    "154": "SUCCESS",
+    "153": "SUCCESS",
+    "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "FAIL",
     "158": "NOT_ALIVE",
-    "159": "NON_PARTICIPATING",
+    "159": "SUCCESS",
     "16": "NON_PARTICIPATING",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
@@ -81,14 +81,14 @@
     "168": "NOT_ALIVE",
     "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
-    "170": "NON_PARTICIPATING",
+    "170": "SUCCESS",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
-    "177": "SUCCESS",
+    "177": "NOT_ALIVE",
     "178": "SUCCESS",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
@@ -96,10 +96,10 @@
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
-    "184": "SUCCESS",
+    "184": "NOT_ALIVE",
     "185": "SUCCESS",
     "186": "SUCCESS",
-    "187": "SUCCESS",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -108,13 +108,13 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "SUCCESS",
+    "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
+    "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
@@ -128,11 +128,11 @@
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
-    "212": "SUCCESS",
+    "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
-    "216": "NOT_ALIVE",
+    "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
@@ -140,20 +140,20 @@
     "220": "NOT_ALIVE",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "NOT_ALIVE",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "NON_PARTICIPATING",
+    "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "SUCCESS",
-    "230": "NON_PARTICIPATING",
-    "231": "NOT_ALIVE",
+    "23": "NOT_ALIVE",
+    "230": "SUCCESS",
+    "231": "SUCCESS",
     "232": "NON_PARTICIPATING",
     "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
-    "235": "NON_PARTICIPATING",
+    "235": "NOT_ALIVE",
     "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "SUCCESS",
@@ -171,29 +171,29 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
+    "251": "SUCCESS",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "NON_PARTICIPATING",
+    "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "SUCCESS",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "SUCCESS",
-    "38": "NOT_ALIVE",
+    "37": "NOT_ALIVE",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "NOT_ALIVE",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
@@ -201,7 +201,7 @@
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
-    "49": "SUCCESS",
+    "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
@@ -209,12 +209,12 @@
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
-    "56": "NOT_ALIVE",
+    "56": "SUCCESS",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NON_PARTICIPATING",
-    "60": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
@@ -234,17 +234,17 @@
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
-    "8": "NOT_ALIVE",
+    "79": "SUCCESS",
+    "8": "SUCCESS",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
-    "88": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
@@ -253,7 +253,7 @@
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
     "98": "SUCCESS",
     "99": "NON_PARTICIPATING"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6651358,
-    6651387
+    6651931,
+    6651960
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 49,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6651029,
+  "last_allreduce_block": 6651834,
   "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 sha256:8c40071d6ea44fd70c6572cbe39b969fccad9cc1449d0768096c3bc02648f9a9
+oid sha256:349017dc9118880e9a68a92619977098ce3da844c8db330bd7075cae7aa7de03
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78d15d297faf27f5e1c467a98742bfe74a78008443656704905dfbb4bb9492e6
+oid sha256:8d4800cd7daefb76ab7b144d7e013e918e56372a4e4dce0441e4e379da912e96
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c752b2cc64904c40544534b037282aed182f51def0f38ef0a8c357074c4b965
+oid sha256:ff95426233f8d9c6746f82ef3bd2f093309390a12d00828553dd1fd3f118e909
 size 4400216536