Run 11. Outer Step 407. Inner Step 4.

config.json

@@ -3,41 +3,41 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "SUCCESS",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NOT_ALIVE",
+    "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "SUCCESS",
     "108": "SUCCESS",
-    "109": "SUCCESS",
+    "109": "NON_PARTICIPATING",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
     "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "SUCCESS",
+    "128": "NON_PARTICIPATING",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
-    "130": "NON_PARTICIPATING",
+    "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
@@ -51,7 +51,7 @@
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "SUCCESS",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
@@ -79,27 +79,27 @@
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "NOT_ALIVE",
+    "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "NOT_ALIVE",
+    "185": "SUCCESS",
     "186": "SUCCESS",
-    "187": "SUCCESS",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -111,7 +111,7 @@
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
+    "198": "FAIL",
     "199": "SUCCESS",
     "2": "SUCCESS",
     "20": "SUCCESS",
@@ -127,18 +127,18 @@
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
-    "213": "NON_PARTICIPATING",
+    "211": "SUCCESS",
+    "212": "NON_PARTICIPATING",
+    "213": "NOT_ALIVE",
     "214": "NOT_ALIVE",
     "215": "SUCCESS",
-    "216": "NON_PARTICIPATING",
+    "216": "NOT_ALIVE",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
-    "219": "NON_PARTICIPATING",
+    "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "SUCCESS",
+    "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
@@ -148,17 +148,17 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
+    "230": "SUCCESS",
     "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
     "241": "SUCCESS",
     "242": "NOT_ALIVE",
@@ -171,7 +171,7 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "SUCCESS",
+    "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -179,16 +179,16 @@
     "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
-    "29": "NOT_ALIVE",
+    "29": "NON_PARTICIPATING",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
-    "33": "NON_PARTICIPATING",
+    "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
-    "37": "SUCCESS",
+    "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
@@ -207,10 +207,10 @@
     "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "SUCCESS",
+    "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
@@ -222,21 +222,21 @@
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "NOT_ALIVE",
+    "68": "SUCCESS",
     "69": "NOT_ALIVE",
-    "7": "NON_PARTICIPATING",
+    "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
+    "71": "FAIL",
     "72": "SUCCESS",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "SUCCESS",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
     "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
+    "80": "SUCCESS",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -254,9 +254,9 @@
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
-    "99": "NOT_ALIVE"
+    "97": "NON_PARTICIPATING",
+    "98": "SUCCESS",
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,6 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6749580,
-    6749613,
     6749645,
     6749679,
     6749711,
@@ -274,7 +272,9 @@
     6749818,
     6749850,
     6749882,
-    6749915
+    6749915,
+    6749949,
+    6750007
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -283,9 +283,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 54,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6749479,
+  "last_allreduce_block": 6749857,
   "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:a8c5f1f88753b9de4e2012b0e6f57b07e4b7837f24503fb1c494be6088854894
+oid sha256:74314ae525b8a0656779ba574fcc8608a8363aad9eab9b1077f3dc56d6e657cd
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d61f411f1fd6dbfbeef3035a4a21bf4790b33a0b5b99597b4b67130e806eb5b9
+oid sha256:d703073f9c3560265c5e540ba3e3e4d03da72d1399ec5ce9da71def2f0e47eab
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fe914204fd3273d93dfdfaf810c9783cfb3ae33b12045b9dd4fc098dc4cf58f
+oid sha256:e50eca78414a5e636e691c825dd29ab59cb09abbdbae3bbba9504b098f416bcd
 size 4400216536