Run 11. Outer Step 248. Inner Step 9.

config.json

@@ -5,17 +5,17 @@
     "1": "NOT_ALIVE",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "SUCCESS",
-    "105": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "SUCCESS",
     "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NOT_ALIVE",
@@ -27,27 +27,27 @@
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NOT_ALIVE",
+    "128": "SUCCESS",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
     "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "SUCCESS",
+    "134": "SUCCESS",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "SUCCESS",
     "142": "NOT_ALIVE",
@@ -55,8 +55,8 @@
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "SUCCESS",
-    "148": "SUCCESS",
+    "147": "NOT_ALIVE",
+    "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
     "150": "SUCCESS",
@@ -66,37 +66,37 @@
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
+    "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "SUCCESS",
-    "163": "NOT_ALIVE",
+    "162": "NOT_ALIVE",
+    "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "SUCCESS",
+    "167": "NOT_ALIVE",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NON_PARTICIPATING",
-    "179": "SUCCESS",
+    "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
-    "184": "SUCCESS",
+    "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
     "186": "NON_PARTICIPATING",
     "187": "SUCCESS",
@@ -106,16 +106,16 @@
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "SUCCESS",
+    "193": "NON_PARTICIPATING",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "NOT_ALIVE",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
@@ -127,11 +127,11 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "SUCCESS",
+    "211": "NON_PARTICIPATING",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
-    "215": "NON_PARTICIPATING",
+    "215": "SUCCESS",
     "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
@@ -144,23 +144,23 @@
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "NON_PARTICIPATING",
+    "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
+    "23": "NOT_ALIVE",
     "230": "SUCCESS",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
-    "235": "NOT_ALIVE",
-    "236": "NOT_ALIVE",
+    "235": "NON_PARTICIPATING",
+    "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -185,11 +185,11 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "SUCCESS",
+    "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "NON_PARTICIPATING",
-    "38": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
     "39": "NOT_ALIVE",
     "4": "SUCCESS",
     "40": "NOT_ALIVE",
@@ -207,7 +207,7 @@
     "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "NOT_ALIVE",
+    "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
     "57": "NON_PARTICIPATING",
@@ -215,18 +215,18 @@
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -234,24 +234,24 @@
     "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NOT_ALIVE",
-    "8": "SUCCESS",
+    "79": "SUCCESS",
+    "8": "NON_PARTICIPATING",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
+    "83": "SUCCESS",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
     "87": "NON_PARTICIPATING",
     "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "SUCCESS",
+    "91": "FAIL",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6677992,
-    6678022
+    6678111,
+    6678142
   ],
   "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": 6677295,
+  "last_allreduce_block": 6678056,
   "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 sha256:552e1cdbf2ccebc995e0a9b40cf3d67f9fa9e4cf62a3dfed6698f3ebce4f0089
+oid sha256:c06ebfbe95dae4483de3e9112ac1c97ac5cb1aa58e0e96904796b012abbeaad9
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f3877e2217fc7026b22b15a793e883de71ad9ddc894612d1889a30ddd06a9aa
+oid sha256:24803c86158bfe4019a4fdba8307e0f422a9e8790641edebf74508bce5e38601
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c62e657b71cf3aa5157e198a468db60e44f602af2dc51c3ed3be9344082d7a9
+oid sha256:0f6cf61c20fa1ea7f4d8e865e2d7bb3789da4a54c670c51fbfa9d05803caae2d
 size 4400216536