Run 11. Outer Step 200. Inner Step 9.

config.json

@@ -5,10 +5,10 @@
     "1": "SUCCESS",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
+    "104": "SUCCESS",
     "105": "NOT_ALIVE",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
@@ -20,28 +20,28 @@
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
-    "116": "NON_PARTICIPATING",
+    "115": "NON_PARTICIPATING",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
+    "120": "NOT_ALIVE",
     "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
-    "123": "NOT_ALIVE",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
-    "13": "NON_PARTICIPATING",
+    "128": "SUCCESS",
+    "129": "NON_PARTICIPATING",
+    "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NON_PARTICIPATING",
+    "134": "SUCCESS",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
@@ -50,20 +50,20 @@
     "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "SUCCESS",
-    "142": "NON_PARTICIPATING",
+    "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
-    "147": "SUCCESS",
-    "148": "NOT_ALIVE",
+    "147": "NOT_ALIVE",
+    "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "SUCCESS",
-    "154": "NOT_ALIVE",
+    "154": "NON_PARTICIPATING",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "SUCCESS",
@@ -72,24 +72,24 @@
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "NOT_ALIVE",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "SUCCESS",
     "166": "NOT_ALIVE",
-    "167": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NON_PARTICIPATING",
+    "178": "SUCCESS",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -99,7 +99,7 @@
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
     "186": "SUCCESS",
-    "187": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -108,11 +108,11 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
+    "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "NOT_ALIVE",
+    "199": "NON_PARTICIPATING",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
     "200": "NON_PARTICIPATING",
@@ -126,14 +126,14 @@
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
-    "210": "SUCCESS",
+    "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
     "215": "NOT_ALIVE",
-    "216": "NOT_ALIVE",
-    "217": "NOT_ALIVE",
+    "216": "NON_PARTICIPATING",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
@@ -144,14 +144,14 @@
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "NON_PARTICIPATING",
+    "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "SUCCESS",
-    "230": "SUCCESS",
+    "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
+    "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
@@ -160,7 +160,7 @@
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -176,7 +176,7 @@
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "SUCCESS",
+    "26": "NOT_ALIVE",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -191,13 +191,13 @@
     "37": "SUCCESS",
     "38": "SUCCESS",
     "39": "NOT_ALIVE",
-    "4": "NON_PARTICIPATING",
+    "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
     "42": "SUCCESS",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
-    "45": "SUCCESS",
+    "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
@@ -216,7 +216,7 @@
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
+    "62": "NOT_ALIVE",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
@@ -231,16 +231,16 @@
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "NOT_ALIVE",
+    "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
     "8": "SUCCESS",
-    "80": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
     "83": "SUCCESS",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
@@ -251,7 +251,7 @@
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
     "97": "SUCCESS",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6658966,
-    6658994
+    6659110,
+    6659138
   ],
   "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": 6658685,
+  "last_allreduce_block": 6659044,
   "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 sha256:260d32bdf9a17c7e40e5b45a23dfe736a1ba88f5fe6f229544a9912c2620ea28
+oid sha256:2edd53ea14f90342c3bff2d4b540cc7a8e78d07c38cf2507467ec77cba6a0291
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08530056b89d9b8c4d3024bf88dcc1168e6fe36637f98c667d33971d232485a9
+oid sha256:4a69b037e75d3c06a3329ddf450aeb0d7e61d47b1c752a34ddb17aad81d137a0
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1135cab8fef0a439487f2ca6cf46f9690396a2161ed559a66ea836bfc8ba8c98
+oid sha256:722a2e986d9dbd0b8bba21d26b9ab266513654331bf7d9fd90e14a184052d09d
 size 4400216536