Run 11. Outer Step 316. Inner Step 9.

config.json

@@ -5,29 +5,29 @@
     "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
+    "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
-    "108": "SUCCESS",
-    "109": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
-    "113": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
-    "116": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NON_PARTICIPATING",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
@@ -35,14 +35,14 @@
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
     "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "NON_PARTICIPATING",
+    "134": "SUCCESS",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -62,24 +62,24 @@
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NOT_ALIVE",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "NON_PARTICIPATING",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NON_PARTICIPATING",
+    "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "SUCCESS",
+    "162": "NOT_ALIVE",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "SUCCESS",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
@@ -90,15 +90,15 @@
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "SUCCESS",
-    "179": "NOT_ALIVE",
+    "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "SUCCESS",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
     "185": "NOT_ALIVE",
-    "186": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -106,13 +106,13 @@
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
-    "198": "NON_PARTICIPATING",
-    "199": "NOT_ALIVE",
+    "198": "SUCCESS",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
     "200": "SUCCESS",
@@ -121,17 +121,17 @@
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NOT_ALIVE",
+    "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "NOT_ALIVE",
+    "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NON_PARTICIPATING",
-    "215": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
+    "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
@@ -149,7 +149,7 @@
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "NOT_ALIVE",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
@@ -158,7 +158,7 @@
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "SUCCESS",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
@@ -171,18 +171,18 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
+    "251": "SUCCESS",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "SUCCESS",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NON_PARTICIPATING",
+    "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
@@ -200,7 +200,7 @@
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NON_PARTICIPATING",
+    "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -219,27 +219,27 @@
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "FAIL",
     "72": "NOT_ALIVE",
-    "73": "NON_PARTICIPATING",
+    "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
-    "79": "NOT_ALIVE",
-    "8": "NON_PARTICIPATING",
+    "79": "SUCCESS",
+    "8": "SUCCESS",
     "80": "NON_PARTICIPATING",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
+    "83": "SUCCESS",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -249,14 +249,14 @@
     "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": "NOT_ALIVE",
+    "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "SUCCESS"
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6707230,
-    6707259
+    6708053,
+    6708080
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 0,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6706278,
+  "last_allreduce_block": 6707707,
   "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:cb998718c792155dd783c9c22b27b8401cc980a5de0666ba5b56ff2765eb18b1
+oid sha256:a2bb62bfafaec2c257a2414920c04e23edf7feb29995f40bb6a92a579d2c6f53
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:639a67b40b6e62967c42bb46228204c978e155461a09783d0c0339b2d0870956
+oid sha256:f9f921e15daa2a60f7ab2717e61ecfa0987ae1fce4ecc20cb841f43c9ec5fcd8
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c750973236bdeb3eaae834fe0532dfa93c2db224653e03996909d35e2199229
+oid sha256:d66250858545a4da830dee4bbf1a868abceb6334928b437529302e4a864c1ac9
 size 4400216536