Run 11. Outer Step 206. Inner Step 9.

config.json

@@ -2,10 +2,10 @@
   "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "SUCCESS",
+    "1": "NOT_ALIVE",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "SUCCESS",
@@ -13,14 +13,14 @@
     "106": "SUCCESS",
     "107": "SUCCESS",
     "108": "SUCCESS",
-    "109": "NOT_ALIVE",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
+    "111": "NOT_ALIVE",
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
@@ -33,10 +33,10 @@
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
-    "128": "NOT_ALIVE",
-    "129": "NOT_ALIVE",
-    "13": "SUCCESS",
+    "127": "SUCCESS",
+    "128": "NON_PARTICIPATING",
+    "129": "NON_PARTICIPATING",
+    "13": "NOT_ALIVE",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
@@ -48,13 +48,13 @@
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "NOT_ALIVE",
+    "140": "SUCCESS",
     "141": "SUCCESS",
     "142": "NON_PARTICIPATING",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
+    "146": "SUCCESS",
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
@@ -62,20 +62,20 @@
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
-    "154": "NON_PARTICIPATING",
+    "153": "NON_PARTICIPATING",
+    "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
-    "16": "NOT_ALIVE",
+    "16": "NON_PARTICIPATING",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
-    "165": "SUCCESS",
+    "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
@@ -86,7 +86,7 @@
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "NON_PARTICIPATING",
+    "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "SUCCESS",
@@ -98,15 +98,15 @@
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
-    "186": "NOT_ALIVE",
-    "187": "NON_PARTICIPATING",
+    "186": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
@@ -114,7 +114,7 @@
     "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "NOT_ALIVE",
+    "20": "SUCCESS",
     "200": "NON_PARTICIPATING",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
@@ -143,30 +143,30 @@
     "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "NOT_ALIVE",
-    "227": "NOT_ALIVE",
+    "226": "SUCCESS",
+    "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "SUCCESS",
+    "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
-    "234": "NOT_ALIVE",
+    "233": "FAIL",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "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",
-    "245": "SUCCESS",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
-    "247": "NOT_ALIVE",
+    "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
@@ -176,7 +176,7 @@
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "SUCCESS",
+    "26": "NOT_ALIVE",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -185,7 +185,7 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "SUCCESS",
+    "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
@@ -193,8 +193,8 @@
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "NON_PARTICIPATING",
-    "42": "NOT_ALIVE",
+    "41": "NOT_ALIVE",
+    "42": "NON_PARTICIPATING",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "SUCCESS",
@@ -204,7 +204,7 @@
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
@@ -214,9 +214,9 @@
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
+    "62": "NOT_ALIVE",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
@@ -231,12 +231,12 @@
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "NOT_ALIVE",
+    "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
-    "8": "NON_PARTICIPATING",
-    "80": "SUCCESS",
+    "8": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "SUCCESS",
     "83": "SUCCESS",
@@ -253,7 +253,7 @@
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
     "97": "SUCCESS",
     "98": "SUCCESS",
     "99": "NON_PARTICIPATING"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6661448,
-    6661465
+    6661561,
+    6661577
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 55,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6661023,
+  "last_allreduce_block": 6661467,
   "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:ec55585f10ab9e3b0c224d52acf7bc512c3283dc43eecb85e47297222ef8570c
+oid sha256:73774e2a30282fb76aec5582cee499de8d052e8a955f5386c1fba566e6c22d2c
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f7400edc9a1438da9051ae0d145b0ea76ad611bdb018e5cd15bdff6e0f56d82
+oid sha256:b808045774fb988a35639cfaa5e68cd8075224a3bf8a6b92a7e42e891eb37ab6
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7b30cbebd1618c6c3591142573fd4f6508b7d5e8407f7807c06cbb22ce2afcc6
+oid sha256:f9008b45f99ada07d9f473e7233ba425f6ab10f160674ed8a4ec4b375131d734
 size 4400216536