Run 11. Outer Step 326. Inner Step 9.

config.json

@@ -11,30 +11,30 @@
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NOT_ALIVE",
-    "108": "SUCCESS",
+    "107": "SUCCESS",
+    "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NON_PARTICIPATING",
+    "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
-    "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "112": "NON_PARTICIPATING",
+    "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
-    "116": "NOT_ALIVE",
+    "115": "NOT_ALIVE",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "SUCCESS",
+    "123": "NON_PARTICIPATING",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "SUCCESS",
+    "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
     "130": "NOT_ALIVE",
@@ -48,45 +48,45 @@
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "SUCCESS",
-    "141": "SUCCESS",
+    "140": "NON_PARTICIPATING",
+    "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "NOT_ALIVE",
+    "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "SUCCESS",
+    "15": "NON_PARTICIPATING",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NON_PARTICIPATING",
     "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",
+    "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "NON_PARTICIPATING",
+    "167": "SUCCESS",
     "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",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "SUCCESS",
@@ -94,11 +94,11 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "NON_PARTICIPATING",
-    "186": "NOT_ALIVE",
+    "186": "NON_PARTICIPATING",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -110,11 +110,11 @@
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "SUCCESS",
+    "197": "NOT_ALIVE",
     "198": "SUCCESS",
-    "199": "SUCCESS",
+    "199": "NON_PARTICIPATING",
     "2": "NOT_ALIVE",
-    "20": "NOT_ALIVE",
+    "20": "SUCCESS",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
@@ -124,14 +124,14 @@
     "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "SUCCESS",
+    "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "NOT_ALIVE",
+    "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
@@ -143,20 +143,20 @@
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "SUCCESS",
+    "226": "NON_PARTICIPATING",
     "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
+    "234": "NOT_ALIVE",
     "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
+    "236": "SUCCESS",
     "237": "SUCCESS",
-    "238": "NOT_ALIVE",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
@@ -172,7 +172,7 @@
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "SUCCESS",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
@@ -182,7 +182,7 @@
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NON_PARTICIPATING",
+    "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NON_PARTICIPATING",
     "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",
@@ -210,16 +210,16 @@
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "NOT_ALIVE",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "NOT_ALIVE",
+    "59": "SUCCESS",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
-    "62": "SUCCESS",
+    "61": "NON_PARTICIPATING",
+    "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "SUCCESS",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
@@ -234,29 +234,29 @@
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "SUCCESS",
-    "8": "NON_PARTICIPATING",
+    "79": "NOT_ALIVE",
+    "8": "SUCCESS",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
+    "83": "SUCCESS",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
-    "88": "NOT_ALIVE",
-    "89": "NON_PARTICIPATING",
+    "88": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
-    "92": "NOT_ALIVE",
+    "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "SUCCESS",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NON_PARTICIPATING",
+    "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "SUCCESS"
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6710306,
-    6710350
+    6713065,
+    6713092
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 36,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6710079,
+  "last_allreduce_block": 6712205,
   "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 sha256:455ee76a58b6b4f0b9387a13f7ad70d2bade8ef98baa99086cf14d2574a9a883
+oid sha256:1c73b71c1f62734cd0257889b43af84db05980978deba2590d2014df67360491
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd2699eeadc2602386252b3f6da7a8ccdc8413b036b39dca041dc2995d808401
+oid sha256:49b2f0f42cde46b597a0c3abfa126e2dbbefd49360f281bb846a2e03a14f2196
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a21f3f0bd42dd8395911b221ea11e476072e45c068f15ea3551d5dedc999d1d
+oid sha256:a616c45043c2ccf41951598e6899564f15085eff1f2ac86336d02b187577ddad
 size 4400216536