Run 11. Outer Step 320. Inner Step 9.

config.json

@@ -11,17 +11,17 @@
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NOT_ALIVE",
-    "108": "NOT_ALIVE",
-    "109": "NOT_ALIVE",
+    "107": "SUCCESS",
+    "108": "SUCCESS",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
-    "112": "NON_PARTICIPATING",
+    "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
+    "115": "SUCCESS",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
@@ -37,18 +37,18 @@
     "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
+    "139": "NOT_ALIVE",
     "14": "NON_PARTICIPATING",
-    "140": "SUCCESS",
+    "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
@@ -58,11 +58,11 @@
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "NON_PARTICIPATING",
+    "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NOT_ALIVE",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -77,14 +77,14 @@
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "SUCCESS",
+    "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
@@ -94,12 +94,12 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "SUCCESS",
     "183": "NOT_ALIVE",
-    "184": "NON_PARTICIPATING",
+    "184": "SUCCESS",
     "185": "NON_PARTICIPATING",
-    "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -111,7 +111,7 @@
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
-    "198": "SUCCESS",
+    "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
@@ -121,7 +121,7 @@
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
@@ -129,16 +129,16 @@
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
     "212": "NOT_ALIVE",
-    "213": "NOT_ALIVE",
-    "214": "NON_PARTICIPATING",
+    "213": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
-    "217": "SUCCESS",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
-    "219": "FAIL",
+    "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "FAIL",
+    "221": "SUCCESS",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
@@ -149,16 +149,16 @@
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "SUCCESS",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
-    "235": "NOT_ALIVE",
-    "236": "NON_PARTICIPATING",
-    "237": "NOT_ALIVE",
-    "238": "NON_PARTICIPATING",
+    "235": "NON_PARTICIPATING",
+    "236": "SUCCESS",
+    "237": "SUCCESS",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "NON_PARTICIPATING",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
@@ -188,7 +188,7 @@
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "NON_PARTICIPATING",
+    "37": "SUCCESS",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
@@ -200,29 +200,29 @@
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "NON_PARTICIPATING",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
+    "62": "SUCCESS",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
@@ -230,31 +230,31 @@
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "SUCCESS",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
-    "79": "SUCCESS",
+    "78": "NON_PARTICIPATING",
+    "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
-    "87": "SUCCESS",
+    "87": "NOT_ALIVE",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
-    "92": "SUCCESS",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "SUCCESS",
+    "97": "NON_PARTICIPATING",
     "98": "SUCCESS",
     "99": "SUCCESS"
   },
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6709015,
-    6709053
+    6709300,
+    6709345
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 31,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6708514,
+  "last_allreduce_block": 6709280,
   "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:a6a4be5fcf82d50797fb52f1e75d3c090421b0bd2f7dbba843f7b6bd42e4ae04
+oid sha256:2dc555c07e72b3d681b26639c1299a33d031dc854546a09f7c4a9cae6e67b51a
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8ce4da7d9e7330a783176b06cefd29b4606178cb2d2b1cfba1313b510a74cb4
+oid sha256:56baccd9067692f568256d92a9106d5c34519e66ebe6cd149c2237de4cd65e59
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50c209f0be159bd5808a2c7c90ecd95711dfadec432abc492d4c284f917b80bb
+oid sha256:e2a93ecd68a4edb41c1aaadb7507094a89d492ca6d6025054b2eccf598f54ad6
 size 4400216536