Run 11. Outer Step 406. Inner Step 9.

config.json

@@ -3,31 +3,31 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NON_PARTICIPATING",
+    "10": "SUCCESS",
     "100": "NOT_ALIVE",
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "SUCCESS",
+    "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
-    "108": "NOT_ALIVE",
+    "107": "SUCCESS",
+    "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
-    "111": "NOT_ALIVE",
+    "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
-    "116": "SUCCESS",
+    "116": "NOT_ALIVE",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
@@ -51,7 +51,7 @@
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
+    "143": "SUCCESS",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
@@ -79,7 +79,7 @@
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
@@ -90,16 +90,16 @@
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "SUCCESS",
+    "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
+    "185": "NOT_ALIVE",
     "186": "SUCCESS",
-    "187": "NON_PARTICIPATING",
+    "187": "SUCCESS",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -127,15 +127,15 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "SUCCESS",
-    "212": "NON_PARTICIPATING",
+    "211": "NOT_ALIVE",
+    "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
     "215": "SUCCESS",
-    "216": "NOT_ALIVE",
+    "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
-    "219": "NOT_ALIVE",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
     "221": "SUCCESS",
@@ -148,17 +148,17 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "SUCCESS",
+    "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NON_PARTICIPATING",
+    "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
+    "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "NON_PARTICIPATING",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "SUCCESS",
     "242": "NOT_ALIVE",
@@ -176,7 +176,7 @@
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -186,21 +186,21 @@
     "32": "NOT_ALIVE",
     "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
-    "35": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "NOT_ALIVE",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -212,22 +212,22 @@
     "56": "NOT_ALIVE",
     "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "SUCCESS",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NOT_ALIVE",
+    "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
     "64": "SUCCESS",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "SUCCESS",
-    "72": "NON_PARTICIPATING",
+    "72": "SUCCESS",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "SUCCESS",
@@ -236,7 +236,7 @@
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
     "8": "SUCCESS",
-    "80": "NOT_ALIVE",
+    "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -254,9 +254,9 @@
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NON_PARTICIPATING",
-    "98": "SUCCESS",
-    "99": "SUCCESS"
+    "97": "NOT_ALIVE",
+    "98": "NOT_ALIVE",
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6749424,
-    6749455
+    6749510,
+    6749540
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 40,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6749137,
+  "last_allreduce_block": 6749479,
   "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 sha256:8692313f7f6d9eef8fd3acb2be7cf9c9dec30de89de9580701135c394b008eca
+oid sha256:5c6787abc13dfc2471c1eb7af46cfcdc5b56d6b96fb0f244d3a6c953dbb45c70
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04cd4f3c1fff83e28b53d77c27e779f74fc57c2fe043f543ff65d07e4d7b1c72
+oid sha256:28f5694d545f818c11e517172164fe84c2097a6eca7509b5d0f97762ce43c59c
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e061ffd24ea39b03068a3f625f94c415090d0907a5a29df8b738c9aaacdbf4d1
+oid sha256:aba6e3a298578555f65ecca67203cffabb07c6375cff01289f1db787c218c3aa
 size 4400216536