Run 11. Outer Step 213. Inner Step 9.

config.json

@@ -2,60 +2,60 @@
   "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
+    "1": "SUCCESS",
     "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
-    "103": "NON_PARTICIPATING",
-    "104": "SUCCESS",
-    "105": "NON_PARTICIPATING",
-    "106": "NOT_ALIVE",
-    "107": "SUCCESS",
+    "103": "FAIL",
+    "104": "NON_PARTICIPATING",
+    "105": "SUCCESS",
+    "106": "SUCCESS",
+    "107": "NON_PARTICIPATING",
     "108": "SUCCESS",
-    "109": "NOT_ALIVE",
+    "109": "FAIL",
     "11": "NOT_ALIVE",
-    "110": "SUCCESS",
-    "111": "NON_PARTICIPATING",
+    "110": "NON_PARTICIPATING",
+    "111": "SUCCESS",
     "112": "NOT_ALIVE",
-    "113": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
-    "116": "SUCCESS",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "SUCCESS",
+    "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "NOT_ALIVE",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
+    "128": "NOT_ALIVE",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
+    "134": "NON_PARTICIPATING",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
+    "139": "FAIL",
     "14": "NON_PARTICIPATING",
-    "140": "SUCCESS",
+    "140": "NOT_ALIVE",
     "141": "SUCCESS",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
+    "143": "FAIL",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "SUCCESS",
-    "147": "NON_PARTICIPATING",
+    "146": "NOT_ALIVE",
+    "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
@@ -66,7 +66,7 @@
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NOT_ALIVE",
+    "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
@@ -77,14 +77,14 @@
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
-    "167": "SUCCESS",
+    "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
@@ -98,7 +98,7 @@
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
     "185": "NOT_ALIVE",
-    "186": "SUCCESS",
+    "186": "NOT_ALIVE",
     "187": "SUCCESS",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -106,9 +106,9 @@
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "NOT_ALIVE",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
+    "195": "SUCCESS",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
@@ -124,16 +124,16 @@
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "SUCCESS",
+    "210": "NON_PARTICIPATING",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "SUCCESS",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
@@ -143,24 +143,24 @@
     "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "NOT_ALIVE",
+    "226": "SUCCESS",
     "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
-    "230": "SUCCESS",
-    "231": "SUCCESS",
+    "230": "NON_PARTICIPATING",
+    "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
-    "234": "NON_PARTICIPATING",
-    "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
-    "237": "SUCCESS",
-    "238": "NOT_ALIVE",
+    "234": "SUCCESS",
+    "235": "NOT_ALIVE",
+    "236": "NON_PARTICIPATING",
+    "237": "NON_PARTICIPATING",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
@@ -171,7 +171,7 @@
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
+    "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
@@ -189,18 +189,18 @@
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
-    "38": "SUCCESS",
+    "38": "FAIL",
     "39": "NOT_ALIVE",
     "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
-    "41": "NON_PARTICIPATING",
-    "42": "SUCCESS",
+    "41": "SUCCESS",
+    "42": "NON_PARTICIPATING",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
     "45": "NON_PARTICIPATING",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -208,16 +208,16 @@
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
-    "55": "SUCCESS",
+    "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "SUCCESS",
-    "62": "NOT_ALIVE",
-    "63": "SUCCESS",
+    "62": "NON_PARTICIPATING",
+    "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
@@ -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",
@@ -238,8 +238,8 @@
     "8": "SUCCESS",
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
-    "83": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
+    "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -248,7 +248,7 @@
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "NOT_ALIVE",
+    "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6664130,
-    6664156
+    6664219,
+    6664243
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 72,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6663738,
+  "last_allreduce_block": 6664151,
   "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 sha256:622e2dab95113e834dbc4b9f99a5e36805db968fb2039afb9b4f0918db683e4d
+oid sha256:b0e57b8fd61e1665b0c2a4812e4fe77603e59c02b79294c4c35d26cee21c9d9f
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7fa646861abca5eaa32c99a71f0f7b7937ea8c85bd8107d7a16b2e31f5cacee
+oid sha256:b0cc9b5f27af39e78bad053d647db3b057bc1d79fd1e2f47b33b477d8254c156
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3096c1c85eda8c2569805531e4960532de22a9654948920b830185c276ffee5
+oid sha256:1266a52b2ec7f7e65cd6d67de02647ba6a675dba73b04a95db7467d83961ae17
 size 4400216536