Run 11. Outer Step 217. Inner Step 9.

config.json

@@ -3,33 +3,33 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "SUCCESS",
-    "10": "NON_PARTICIPATING",
+    "10": "SUCCESS",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
-    "103": "FAIL",
+    "103": "SUCCESS",
     "104": "NON_PARTICIPATING",
-    "105": "SUCCESS",
+    "105": "NOT_ALIVE",
     "106": "SUCCESS",
     "107": "NON_PARTICIPATING",
-    "108": "SUCCESS",
+    "108": "NOT_ALIVE",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "SUCCESS",
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "SUCCESS",
+    "120": "NON_PARTICIPATING",
     "121": "SUCCESS",
     "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
@@ -37,11 +37,11 @@
     "128": "SUCCESS",
     "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
-    "130": "SUCCESS",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
+    "134": "NON_PARTICIPATING",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
@@ -49,17 +49,17 @@
     "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "SUCCESS",
-    "141": "NON_PARTICIPATING",
+    "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
+    "146": "SUCCESS",
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NON_PARTICIPATING",
+    "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "SUCCESS",
@@ -70,7 +70,7 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "SUCCESS",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
@@ -84,12 +84,12 @@
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "SUCCESS",
+    "173": "FAIL",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "SUCCESS",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -99,16 +99,16 @@
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
     "186": "SUCCESS",
-    "187": "NON_PARTICIPATING",
+    "187": "SUCCESS",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "SUCCESS",
+    "193": "NON_PARTICIPATING",
     "194": "NOT_ALIVE",
-    "195": "SUCCESS",
+    "195": "FAIL",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
@@ -144,7 +144,7 @@
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
-    "227": "NON_PARTICIPATING",
+    "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
@@ -155,12 +155,12 @@
     "234": "NON_PARTICIPATING",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
-    "237": "SUCCESS",
+    "237": "NON_PARTICIPATING",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -171,7 +171,7 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "SUCCESS",
+    "251": "NOT_ALIVE",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -188,26 +188,26 @@
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "SUCCESS",
-    "38": "NON_PARTICIPATING",
+    "37": "NOT_ALIVE",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
     "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
     "41": "SUCCESS",
-    "42": "SUCCESS",
+    "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NON_PARTICIPATING",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NON_PARTICIPATING",
+    "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NON_PARTICIPATING",
+    "51": "NOT_ALIVE",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "SUCCESS",
+    "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
@@ -217,7 +217,7 @@
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
     "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
@@ -231,32 +231,32 @@
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "NOT_ALIVE",
+    "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "79": "NOT_ALIVE",
     "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
+    "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
+    "82": "SUCCESS",
     "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
-    "88": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "NOT_ALIVE",
+    "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "SUCCESS",
-    "98": "SUCCESS",
-    "99": "NON_PARTICIPATING"
+    "98": "NON_PARTICIPATING",
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6665587,
-    6665612
+    6665712,
+    6665730
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 63,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6665282,
+  "last_allreduce_block": 6665673,
   "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 sha256:1cc42fd8a78779d00e888f8de56c7feaf6d8101abb025f90da64fccb649f91f5
+oid sha256:63bc3831de7ac8400bb568beadc6bda7a96ac5868dec16cc5933bac97e6b9a2b
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15fc40083c679cbe392e8beff299d595964a432cc8d99348464d7b858f963390
+oid sha256:a0ebb7a4a4ab62b32254cb354ce3289083c22a2ac2fd4fcdb7b536d57510d08c
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fec2d314ace1e94fef8b6ce114ff5b02b35523d428968daa4eac309c538ce111
+oid sha256:eba86dfcfe89d1d9fa7c8b8e749259bf93f014e4077b802735627787c9ae957c
 size 4400216536