Run 11. Outer Step 318. Inner Step 9.

config.json

@@ -3,7 +3,7 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NON_PARTICIPATING",
+    "10": "SUCCESS",
     "100": "NOT_ALIVE",
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
@@ -11,37 +11,37 @@
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
-    "108": "SUCCESS",
+    "107": "NOT_ALIVE",
+    "108": "NOT_ALIVE",
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "SUCCESS",
-    "116": "SUCCESS",
+    "115": "NON_PARTICIPATING",
+    "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
     "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
-    "130": "SUCCESS",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
+    "134": "SUCCESS",
     "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
@@ -58,11 +58,11 @@
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "NOT_ALIVE",
+    "15": "NON_PARTICIPATING",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -72,7 +72,7 @@
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
@@ -94,12 +94,12 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "SUCCESS",
-    "185": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
+    "185": "NON_PARTICIPATING",
     "186": "NON_PARTICIPATING",
-    "187": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -121,7 +121,7 @@
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
@@ -129,16 +129,16 @@
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
     "212": "NOT_ALIVE",
-    "213": "NON_PARTICIPATING",
+    "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "NOT_ALIVE",
-    "217": "NON_PARTICIPATING",
+    "216": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
-    "219": "SUCCESS",
+    "219": "FAIL",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "SUCCESS",
+    "221": "FAIL",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
@@ -151,16 +151,16 @@
     "230": "SUCCESS",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "SUCCESS",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
-    "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
-    "237": "SUCCESS",
+    "235": "NOT_ALIVE",
+    "236": "NON_PARTICIPATING",
+    "237": "NOT_ALIVE",
     "238": "NON_PARTICIPATING",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "NOT_ALIVE",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -172,7 +172,7 @@
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "FAIL",
+    "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
@@ -182,7 +182,7 @@
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
@@ -204,39 +204,39 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NON_PARTICIPATING",
+    "51": "NOT_ALIVE",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
-    "59": "NOT_ALIVE",
+    "59": "NON_PARTICIPATING",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
-    "78": "NON_PARTICIPATING",
+    "78": "NOT_ALIVE",
     "79": "SUCCESS",
     "8": "NON_PARTICIPATING",
-    "80": "NOT_ALIVE",
+    "80": "NON_PARTICIPATING",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -245,18 +245,18 @@
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
-    "89": "NOT_ALIVE",
+    "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
-    "92": "NON_PARTICIPATING",
+    "92": "SUCCESS",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "NON_PARTICIPATING"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6708455,
-    6708487
+    6708575,
+    6708604
   ],
   "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": 6708109,
+  "last_allreduce_block": 6708514,
   "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 sha256:9dd032f1817d972ac4f2de946517b42e9a2fec51f6921d359355caf8eb02473d
+oid sha256:4166acb7cadeb88c398b4788cfda30500d76c09ffd75233cb9941cb97fa7b94e
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95f0ffaab994fb0fc304196d74338e022fec9cbbdced848c8839979c0a1d9e6a
+oid sha256:4ce65c447a48c3aeb5be8dfc60e0716a2436b84d91a2438448653e7ea44ef154
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2862ec64f8c3ae04bc092ffded0e40bdbbb7b740fb59de599cadb1f15bf40a6b
+oid sha256:08d104378cd7609824a76e8cbb3c90add3fd0e3dfcead9aa190af79e17dee56a
 size 4400216536