Run 11. Outer Step 197. Inner Step 8.

config.json

@@ -2,13 +2,13 @@
   "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
-    "10": "NON_PARTICIPATING",
+    "1": "SUCCESS",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
-    "103": "NON_PARTICIPATING",
-    "104": "SUCCESS",
+    "103": "NOT_ALIVE",
+    "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
@@ -20,10 +20,10 @@
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "FAIL",
+    "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
-    "118": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
@@ -41,20 +41,20 @@
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
+    "134": "NOT_ALIVE",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
+    "139": "NOT_ALIVE",
     "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "SUCCESS",
     "142": "NOT_ALIVE",
-    "143": "SUCCESS",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "SUCCESS",
+    "146": "NON_PARTICIPATING",
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
@@ -62,11 +62,11 @@
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NOT_ALIVE",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "FAIL",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
@@ -75,21 +75,21 @@
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
-    "165": "NON_PARTICIPATING",
+    "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NON_PARTICIPATING",
+    "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -98,7 +98,7 @@
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
-    "186": "SUCCESS",
+    "186": "NON_PARTICIPATING",
     "187": "SUCCESS",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -108,14 +108,14 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "SUCCESS",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "NON_PARTICIPATING",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
-    "200": "SUCCESS",
+    "200": "NOT_ALIVE",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
@@ -126,21 +126,21 @@
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
+    "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
     "215": "NOT_ALIVE",
-    "216": "NOT_ALIVE",
+    "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
-    "221": "NON_PARTICIPATING",
+    "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
-    "223": "NOT_ALIVE",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
@@ -153,30 +153,30 @@
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
-    "235": "NON_PARTICIPATING",
+    "235": "NOT_ALIVE",
     "236": "SUCCESS",
     "237": "SUCCESS",
-    "238": "FAIL",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NON_PARTICIPATING",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "NOT_ALIVE",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
-    "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
+    "245": "NON_PARTICIPATING",
+    "246": "NOT_ALIVE",
+    "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "SUCCESS",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -188,12 +188,12 @@
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "SUCCESS",
+    "37": "NON_PARTICIPATING",
     "38": "SUCCESS",
     "39": "NOT_ALIVE",
-    "4": "SUCCESS",
+    "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "NOT_ALIVE",
+    "41": "SUCCESS",
     "42": "SUCCESS",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
@@ -209,14 +209,14 @@
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
-    "56": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
+    "62": "SUCCESS",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
@@ -226,7 +226,7 @@
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -235,27 +235,27 @@
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
-    "8": "NOT_ALIVE",
+    "8": "NON_PARTICIPATING",
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
-    "83": "SUCCESS",
-    "84": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
+    "83": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
-    "88": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
+    "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
-    "97": "SUCCESS",
-    "98": "NON_PARTICIPATING",
+    "97": "NON_PARTICIPATING",
+    "98": "SUCCESS",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6657462,
-    6657490
+    6657559,
+    6657587
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 50,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6657129,
+  "last_allreduce_block": 6657522,
   "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 sha256:3cf9d0d26b73dfe12c19f21074a324ee100d0c921a24e7bd92a6e3ac118f0df9
+oid sha256:f76703cec5cb21e1b0a048c1905e5d306c8f0abe807d7c691ff0d66003df080d
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7f92bc91b879db558b285fd681299876db6a1aa4642cef42795996dc90ac2ae
+oid sha256:555c6391042158b03ae3eea92e3d99b0a97ccb4f81b2bbf7b34422d498e48974
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d0d008ee7b456dab494db2471ec22ae37d38d8062538f5dfc82a32eea83d5a9c
+oid sha256:c3438f5e50da9e3a5dcbabf35c4d7b9cff9b61464203e827197b52aa5dac0c4e
 size 4400216536