Run 11. Outer Step 238. Inner Step 8.

config.json

@@ -1,24 +1,24 @@
 {
-  "_name_or_path": "mujerry/dd",
+  "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
-    "10": "SUCCESS",
+    "1": "NOT_ALIVE",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
-    "106": "NON_PARTICIPATING",
+    "106": "NOT_ALIVE",
     "107": "SUCCESS",
-    "108": "SUCCESS",
+    "108": "NOT_ALIVE",
     "109": "SUCCESS",
-    "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "11": "NON_PARTICIPATING",
+    "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
@@ -26,37 +26,37 @@
     "118": "NOT_ALIVE",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
-    "121": "NOT_ALIVE",
+    "120": "SUCCESS",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "SUCCESS",
+    "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
-    "132": "NOT_ALIVE",
+    "132": "NON_PARTICIPATING",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "SUCCESS",
+    "134": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
-    "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
-    "14": "NOT_ALIVE",
+    "138": "NON_PARTICIPATING",
+    "139": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "SUCCESS",
-    "148": "NON_PARTICIPATING",
+    "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
     "150": "NON_PARTICIPATING",
@@ -65,7 +65,7 @@
     "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
-    "156": "NOT_ALIVE",
+    "156": "NON_PARTICIPATING",
     "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
@@ -74,26 +74,26 @@
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
     "163": "NOT_ALIVE",
-    "164": "NOT_ALIVE",
+    "164": "NON_PARTICIPATING",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
-    "171": "NON_PARTICIPATING",
-    "172": "NOT_ALIVE",
+    "171": "NOT_ALIVE",
+    "172": "NON_PARTICIPATING",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
-    "179": "SUCCESS",
+    "178": "SUCCESS",
+    "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
-    "181": "NOT_ALIVE",
+    "181": "NON_PARTICIPATING",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
@@ -104,7 +104,7 @@
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
-    "191": "NOT_ALIVE",
+    "191": "NON_PARTICIPATING",
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
@@ -114,20 +114,20 @@
     "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
+    "20": "NOT_ALIVE",
     "200": "NON_PARTICIPATING",
-    "201": "NOT_ALIVE",
+    "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
-    "204": "NOT_ALIVE",
+    "204": "NON_PARTICIPATING",
     "205": "NOT_ALIVE",
     "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
+    "209": "SUCCESS",
     "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
-    "211": "NOT_ALIVE",
+    "210": "NOT_ALIVE",
+    "211": "NON_PARTICIPATING",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
@@ -137,32 +137,32 @@
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "NOT_ALIVE",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "NON_PARTICIPATING",
+    "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
     "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
+    "23": "SUCCESS",
     "230": "SUCCESS",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "SUCCESS",
+    "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "SUCCESS",
-    "237": "NON_PARTICIPATING",
+    "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
     "241": "SUCCESS",
-    "242": "NOT_ALIVE",
-    "243": "NOT_ALIVE",
+    "242": "SUCCESS",
+    "243": "NON_PARTICIPATING",
     "244": "NOT_ALIVE",
     "245": "SUCCESS",
     "246": "NOT_ALIVE",
@@ -185,22 +185,22 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "SUCCESS",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "NOT_ALIVE",
-    "38": "SUCCESS",
+    "37": "SUCCESS",
+    "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
-    "4": "NON_PARTICIPATING",
+    "4": "SUCCESS",
     "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
-    "45": "NON_PARTICIPATING",
+    "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -208,7 +208,7 @@
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
-    "55": "NON_PARTICIPATING",
+    "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
@@ -220,13 +220,13 @@
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
-    "66": "NOT_ALIVE",
+    "66": "SUCCESS",
     "67": "NOT_ALIVE",
-    "68": "NON_PARTICIPATING",
+    "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
+    "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -234,28 +234,28 @@
     "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "SUCCESS",
-    "8": "SUCCESS",
+    "79": "NON_PARTICIPATING",
+    "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
-    "83": "NOT_ALIVE",
+    "82": "NOT_ALIVE",
+    "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
-    "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
+    "9": "NON_PARTICIPATING",
+    "90": "NOT_ALIVE",
     "91": "SUCCESS",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
-    "95": "NOT_ALIVE",
+    "95": "NON_PARTICIPATING",
     "96": "NOT_ALIVE",
-    "97": "SUCCESS",
-    "98": "NOT_ALIVE",
+    "97": "NOT_ALIVE",
+    "98": "SUCCESS",
     "99": "SUCCESS"
   },
   "architectures": [
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6673751,
-    6673794
+    6673925,
+    6673954
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6672695,
+  "last_allreduce_block": 6673846,
   "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 sha256:115053c148c379164b9ec7976a070c35942901912456c0b8d6564b26d0669d3b
+oid sha256:7a83ca5bda97e1e4690c01920a571951c7de2178a60445faa892ac8264fc6864
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b604c10d248d3565df8d6b58c7cd5a05e0a3224b61b7ec0f9bd1a4f4f3b924c
+oid sha256:c95fee00a3050d825207e95a2c29bc74ccbcf913968f5d8fa51175c17ac69882
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4968ada840c6a36020eeb98d93e9babdfdb9b1de1dff4190ed79efddb281c46
+oid sha256:163e9f2731ed9aabf881c3667c01c0394b487f8ecfb60650c561fcd59d08c15d
 size 4400216536