Run 11. Outer Step 51. Inner Step 8.

config.json

@@ -8,7 +8,7 @@
     "101": "NOT_ALIVE",
     "102": "SUCCESS",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
+    "104": "SUCCESS",
     "105": "SUCCESS",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
@@ -17,32 +17,32 @@
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
     "111": "SUCCESS",
-    "112": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
-    "116": "NON_PARTICIPATING",
+    "116": "NOT_ALIVE",
     "117": "SUCCESS",
-    "118": "SUCCESS",
+    "118": "NOT_ALIVE",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
     "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
-    "124": "SUCCESS",
+    "124": "NOT_ALIVE",
     "125": "SUCCESS",
     "126": "SUCCESS",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
-    "13": "NON_PARTICIPATING",
-    "130": "NON_PARTICIPATING",
+    "13": "NOT_ALIVE",
+    "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
-    "135": "SUCCESS",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -51,7 +51,7 @@
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "SUCCESS",
-    "143": "NOT_ALIVE",
+    "143": "SUCCESS",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
@@ -62,12 +62,12 @@
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NOT_ALIVE",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "SUCCESS",
-    "158": "SUCCESS",
+    "158": "NOT_ALIVE",
     "159": "SUCCESS",
     "16": "SUCCESS",
     "160": "NOT_ALIVE",
@@ -76,7 +76,7 @@
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
+    "166": "SUCCESS",
     "167": "NOT_ALIVE",
     "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
@@ -84,15 +84,15 @@
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "NOT_ALIVE",
+    "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
     "18": "NON_PARTICIPATING",
-    "180": "NOT_ALIVE",
+    "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
@@ -103,7 +103,7 @@
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
-    "190": "SUCCESS",
+    "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
@@ -133,7 +133,7 @@
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
@@ -147,18 +147,18 @@
     "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
-    "232": "NON_PARTICIPATING",
+    "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "SUCCESS",
     "235": "NOT_ALIVE",
-    "236": "NOT_ALIVE",
+    "236": "NON_PARTICIPATING",
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
     "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
@@ -170,7 +170,7 @@
     "248": "SUCCESS",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "SUCCESS",
     "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
@@ -181,12 +181,12 @@
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
-    "30": "NOT_ALIVE",
+    "30": "NON_PARTICIPATING",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "SUCCESS",
-    "35": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
     "38": "NON_PARTICIPATING",
@@ -198,38 +198,38 @@
     "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
-    "46": "SUCCESS",
+    "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "SUCCESS",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NON_PARTICIPATING",
+    "51": "NOT_ALIVE",
     "52": "SUCCESS",
-    "53": "NOT_ALIVE",
+    "53": "SUCCESS",
     "54": "NOT_ALIVE",
-    "55": "NOT_ALIVE",
+    "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "SUCCESS",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "SUCCESS",
     "62": "SUCCESS",
     "63": "NON_PARTICIPATING",
-    "64": "SUCCESS",
+    "64": "NOT_ALIVE",
     "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "NON_PARTICIPATING",
+    "69": "SUCCESS",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
-    "74": "NON_PARTICIPATING",
+    "74": "NOT_ALIVE",
     "75": "SUCCESS",
     "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
@@ -239,13 +239,13 @@
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
-    "85": "NOT_ALIVE",
-    "86": "SUCCESS",
+    "85": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
     "87": "SUCCESS",
-    "88": "NOT_ALIVE",
-    "89": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
+    "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
@@ -253,10 +253,10 @@
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
-    "98": "SUCCESS",
-    "99": "NOT_ALIVE"
+    "98": "NOT_ALIVE",
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6570507,
-    6570536
+    6570611,
+    6570639
   ],
   "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": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6570193,
+  "last_allreduce_block": 6570588,
   "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 sha256:23877eb0d2c5b1aa30cfb75a28f145da261f425a1750cef3d46c3067f5080618
+oid sha256:f032fd9f9c274f424caaafaf382e868a78f425a92a273a58fbfda4d3684e6815
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8eeb86530898852f044b5fb69ea29db3e9e73fcd44e7deae304c98852ee8c05c
+oid sha256:a3f46620fef3465d4be16f893fe123cc54634ac0338a908bbc30a6c5a2340786
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:880db5864fd3456d3ef954c1801f2f6cf291e9c111f4839763125e2cb9d58e46
+oid sha256:e88a1f9330db9d14a6f5c4c179ae6e079fa01c60339ca5bb718a0acf1f100f4e
 size 4400216536