Run 11. Outer Step 234. Inner Step 9.

config.json

@@ -5,31 +5,31 @@
     "1": "NON_PARTICIPATING",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
-    "107": "NON_PARTICIPATING",
+    "107": "SUCCESS",
     "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
-    "113": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NON_PARTICIPATING",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "SUCCESS",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
@@ -41,8 +41,8 @@
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "134": "NOT_ALIVE",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -51,12 +51,12 @@
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "NON_PARTICIPATING",
-    "148": "NOT_ALIVE",
+    "147": "SUCCESS",
+    "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
@@ -84,19 +84,19 @@
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
+    "178": "NON_PARTICIPATING",
+    "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
     "186": "SUCCESS",
     "187": "SUCCESS",
@@ -108,39 +108,39 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NON_PARTICIPATING",
     "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
-    "200": "NOT_ALIVE",
+    "200": "NON_PARTICIPATING",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NOT_ALIVE",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
+    "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
     "214": "SUCCESS",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "NOT_ALIVE",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "NON_PARTICIPATING",
+    "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
@@ -153,30 +153,30 @@
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
-    "235": "NOT_ALIVE",
-    "236": "NON_PARTICIPATING",
+    "235": "NON_PARTICIPATING",
+    "236": "SUCCESS",
     "237": "SUCCESS",
-    "238": "SUCCESS",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "FAIL",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "SUCCESS",
     "246": "NOT_ALIVE",
-    "247": "NON_PARTICIPATING",
+    "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
+    "251": "SUCCESS",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "SUCCESS",
+    "26": "NOT_ALIVE",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -191,14 +191,14 @@
     "37": "SUCCESS",
     "38": "SUCCESS",
     "39": "NOT_ALIVE",
-    "4": "FAIL",
+    "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
-    "45": "NON_PARTICIPATING",
-    "46": "NOT_ALIVE",
+    "45": "NOT_ALIVE",
+    "46": "NON_PARTICIPATING",
     "47": "NOT_ALIVE",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
@@ -208,13 +208,13 @@
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
-    "55": "NON_PARTICIPATING",
+    "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
@@ -222,7 +222,7 @@
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
@@ -234,13 +234,13 @@
     "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "79": "SUCCESS",
     "8": "SUCCESS",
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
+    "82": "NOT_ALIVE",
     "83": "SUCCESS",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
@@ -248,14 +248,14 @@
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "FAIL",
+    "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
-    "97": "NOT_ALIVE",
-    "98": "SUCCESS",
+    "97": "SUCCESS",
+    "98": "NON_PARTICIPATING",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6672322,
-    6672352
+    6672412,
+    6672455
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6672001,
+  "last_allreduce_block": 6672349,
   "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:bc69c578ec15e45c962b7597df8953be06fed541b583a2f48d1d499875a082da
+oid sha256:eeac14b9bb174e72f8f8224aadc5a9b19c27c80e40dfa91836d2cddef863e61a
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eea41ec5d160d6022de085aced788835a7f81320c32ff4479226a108fd4ad97b
+oid sha256:b3e97f57118cbd5c6a3223ae65b5c5a43ea52fddbad04a29fbe11ad88507debf
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35fb1c1b5e6431dae1d6ca5c34cb16b2c6c09c6297f37b04113b2b83d6869918
+oid sha256:d4a7d64cd9bf9d9698926f846cb64e1ba10f10536bff9ec89674388430d02dd5
 size 4400216536