Run 11. Outer Step 410. Inner Step 9.

config.json

@@ -5,49 +5,49 @@
     "1": "NOT_ALIVE",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "SUCCESS",
-    "105": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NON_PARTICIPATING",
+    "107": "SUCCESS",
     "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
-    "113": "NON_PARTICIPATING",
+    "112": "NON_PARTICIPATING",
+    "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
-    "116": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "SUCCESS",
+    "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "SUCCESS",
     "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "123": "NON_PARTICIPATING",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
     "129": "NOT_ALIVE",
-    "13": "NON_PARTICIPATING",
+    "13": "SUCCESS",
     "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "NON_PARTICIPATING",
+    "134": "SUCCESS",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
@@ -79,22 +79,22 @@
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "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": "SUCCESS",
+    "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "SUCCESS",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
@@ -112,8 +112,8 @@
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NOT_ALIVE",
-    "199": "SUCCESS",
-    "2": "SUCCESS",
+    "199": "NOT_ALIVE",
+    "2": "NOT_ALIVE",
     "20": "SUCCESS",
     "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
@@ -124,11 +124,11 @@
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
-    "212": "NON_PARTICIPATING",
+    "212": "SUCCESS",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
@@ -148,19 +148,19 @@
     "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": "NOT_ALIVE",
     "236": "SUCCESS",
     "237": "SUCCESS",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "NON_PARTICIPATING",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -175,11 +175,11 @@
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
-    "255": "NON_PARTICIPATING",
+    "255": "NOT_ALIVE",
     "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
-    "29": "NOT_ALIVE",
+    "29": "NON_PARTICIPATING",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
@@ -193,7 +193,7 @@
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
-    "41": "NON_PARTICIPATING",
+    "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
@@ -210,7 +210,7 @@
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
@@ -227,16 +227,16 @@
     "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "SUCCESS",
-    "72": "NOT_ALIVE",
+    "72": "SUCCESS",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "SUCCESS",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NOT_ALIVE",
+    "79": "SUCCESS",
     "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -248,8 +248,8 @@
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "SUCCESS",
-    "92": "SUCCESS",
+    "91": "NOT_ALIVE",
+    "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6750903,
-    6750942
+    6751011,
+    6751040
   ],
   "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": 6750582,
+  "last_allreduce_block": 6750947,
   "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 sha256:83460f322000a5f2bf2a39a61014b3262e5552abdca2f0341f4fec503815cbcc
+oid sha256:3408f060b22502ac97bb5518e270d8532ed5ff747160cb8b29340470402c20c8
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4adcc3664cf3d77ae5ea5ef34742a42f5cfe79c5952f98f0d79aab05969642f
+oid sha256:8e4bead8f14e07fd934429bca557acedbe3c1fe51baddfdc7c3283bceab16b65
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:716a681cb8ae4405c79d19cd784fe32c2f70573ab49e4bf6dda9d2168185cb11
+oid sha256:fbef60bceb2d4736df9d212c2b50ee6e8b090af85c84265f32d9debc99666254
 size 4400216536