Run 11. Outer Step 342. Inner Step 9.

config.json

@@ -3,12 +3,12 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NOT_ALIVE",
+    "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "SUCCESS",
+    "104": "FAIL",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "SUCCESS",
@@ -17,11 +17,11 @@
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
+    "112": "NOT_ALIVE",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
-    "116": "NON_PARTICIPATING",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
@@ -33,7 +33,7 @@
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "SUCCESS",
+    "127": "NOT_ALIVE",
     "128": "SUCCESS",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
@@ -41,7 +41,7 @@
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
+    "134": "SUCCESS",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
@@ -49,7 +49,7 @@
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
     "140": "NON_PARTICIPATING",
-    "141": "NON_PARTICIPATING",
+    "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
@@ -58,26 +58,26 @@
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "NON_PARTICIPATING",
+    "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "NON_PARTICIPATING",
     "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": "SUCCESS",
+    "167": "NOT_ALIVE",
     "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
@@ -89,7 +89,7 @@
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "SUCCESS",
+    "178": "NOT_ALIVE",
     "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -97,8 +97,8 @@
     "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
-    "186": "NOT_ALIVE",
+    "185": "NON_PARTICIPATING",
+    "186": "NON_PARTICIPATING",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -110,12 +110,12 @@
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "SUCCESS",
-    "198": "NON_PARTICIPATING",
-    "199": "SUCCESS",
+    "197": "NOT_ALIVE",
+    "198": "NOT_ALIVE",
+    "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
-    "200": "NON_PARTICIPATING",
+    "200": "NOT_ALIVE",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
@@ -127,12 +127,12 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "NON_PARTICIPATING",
+    "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
-    "216": "NOT_ALIVE",
+    "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
@@ -148,14 +148,14 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
+    "230": "SUCCESS",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
-    "233": "NON_PARTICIPATING",
+    "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "NOT_ALIVE",
+    "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
@@ -170,13 +170,13 @@
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NOT_ALIVE",
+    "250": "NON_PARTICIPATING",
     "251": "SUCCESS",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -184,7 +184,7 @@
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
-    "33": "NON_PARTICIPATING",
+    "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -210,7 +210,7 @@
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "NOT_ALIVE",
+    "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
@@ -219,7 +219,7 @@
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "SUCCESS",
+    "65": "FAIL",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
@@ -235,27 +235,27 @@
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
-    "8": "NON_PARTICIPATING",
+    "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
-    "84": "NON_PARTICIPATING",
+    "84": "SUCCESS",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
-    "87": "NON_PARTICIPATING",
+    "87": "NOT_ALIVE",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "NOT_ALIVE",
-    "92": "NON_PARTICIPATING",
+    "91": "SUCCESS",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "SUCCESS",
-    "98": "SUCCESS",
+    "97": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6718411,
-    6718443
+    6719367,
+    6719396
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 22,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6718281,
+  "last_allreduce_block": 6719096,
   "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 sha256:212fa2f402d0948fab71dea1db16b816fbbf15ac821c555aabf80ea153058ded
+oid sha256:e3cdf1add8c45bbede24ded51c9e6ce05fc6f2898732da8cb2f35c9548eb0fbe
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2132e44f44a8adc3aa6287e155686de2c0cd56f5439203e6e4a96b52f1156137
+oid sha256:2a10a45b6eb1cd1efc6220b4c7d017e12aaf3994603d30068d0eb869b365119a
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a466c6cec41d1f256acaf6770e328569b7770e8638d721caa983929efc3d3259
+oid sha256:efac2f071eac4eef9f922aa07c2834045163ad9256f75aca21bc0a7015ad23df
 size 4400216536