Run 11. Outer Step 322. Inner Step 9.

config.json

@@ -11,33 +11,33 @@
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
+    "107": "NOT_ALIVE",
     "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
-    "116": "NON_PARTICIPATING",
+    "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
+    "128": "SUCCESS",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
-    "130": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
@@ -46,27 +46,27 @@
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NOT_ALIVE",
+    "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "NON_PARTICIPATING",
-    "141": "NON_PARTICIPATING",
+    "140": "SUCCESS",
+    "141": "SUCCESS",
     "142": "NOT_ALIVE",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "SUCCESS",
+    "147": "NOT_ALIVE",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "NON_PARTICIPATING",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
@@ -84,7 +84,7 @@
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
@@ -94,12 +94,12 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "SUCCESS",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "SUCCESS",
+    "184": "NON_PARTICIPATING",
     "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -111,10 +111,10 @@
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
-    "198": "NON_PARTICIPATING",
+    "198": "SUCCESS",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
+    "20": "NOT_ALIVE",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
@@ -130,10 +130,10 @@
     "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NOT_ALIVE",
-    "215": "NON_PARTICIPATING",
+    "214": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
@@ -149,22 +149,22 @@
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "NOT_ALIVE",
+    "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
+    "236": "NOT_ALIVE",
     "237": "SUCCESS",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
+    "245": "NOT_ALIVE",
     "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
@@ -172,7 +172,7 @@
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "NON_PARTICIPATING",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
@@ -184,7 +184,7 @@
     "30": "NOT_ALIVE",
     "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
-    "33": "NOT_ALIVE",
+    "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -210,7 +210,7 @@
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
@@ -222,36 +222,36 @@
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "NOT_ALIVE",
+    "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NOT_ALIVE",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NOT_ALIVE",
+    "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
-    "78": "NON_PARTICIPATING",
-    "79": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
+    "79": "SUCCESS",
     "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
-    "84": "NOT_ALIVE",
+    "83": "NON_PARTICIPATING",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
-    "87": "NOT_ALIVE",
-    "88": "NON_PARTICIPATING",
+    "87": "SUCCESS",
+    "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "SUCCESS",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NON_PARTICIPATING",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6710008,
-    6710037
+    6710110,
+    6710139
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 58,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6709280,
+  "last_allreduce_block": 6710079,
   "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:f7c9836f5d7addd3e0b32176571134797c193b3f4c24480c5e5b646f864729b1
+oid sha256:70dcec22505a8f0be3e432a86e4c6bcc68c2a513d67ca8e4a48c15d78370beba
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:430de06ad79d45744824a42b7bab6b46671b5538d85aea518cc5f6e696f9eb0f
+oid sha256:f33f5b6d75b6c8f879d3e70d0a6fe664cc2925905e8800c08edc7ab938c70c4d
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e00980dcb477ec8bc57424d2c99ec16b916feef1700bc4da3303cc5ded14607a
+oid sha256:2798e1c792c248f25555d6b7dbed7cd525ffbe9b7ac90866cf88e0379d5cf414
 size 4400216536