Run 11. Outer Step 405. Inner Step 9.

config.json

@@ -1,9 +1,9 @@
 {
-  "_name_or_path": "mujerry/dd",
+  "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "SUCCESS",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "SUCCESS",
     "102": "NOT_ALIVE",
@@ -15,29 +15,29 @@
     "108": "NOT_ALIVE",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
-    "112": "NON_PARTICIPATING",
+    "110": "SUCCESS",
+    "111": "NOT_ALIVE",
+    "112": "SUCCESS",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "FAIL",
+    "115": "NON_PARTICIPATING",
     "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "NOT_ALIVE",
+    "127": "SUCCESS",
     "128": "SUCCESS",
     "129": "NOT_ALIVE",
-    "13": "NON_PARTICIPATING",
-    "130": "NOT_ALIVE",
+    "13": "SUCCESS",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
@@ -47,18 +47,18 @@
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
     "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "NOT_ALIVE",
+    "15": "SUCCESS",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
@@ -94,11 +94,11 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "SUCCESS",
     "183": "NOT_ALIVE",
-    "184": "NOT_ALIVE",
-    "185": "NON_PARTICIPATING",
-    "186": "NON_PARTICIPATING",
+    "184": "NON_PARTICIPATING",
+    "185": "SUCCESS",
+    "186": "SUCCESS",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -108,14 +108,14 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "SUCCESS",
+    "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
-    "2": "NON_PARTICIPATING",
+    "2": "SUCCESS",
     "20": "SUCCESS",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
@@ -128,55 +128,55 @@
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
-    "212": "NOT_ALIVE",
+    "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
-    "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
-    "217": "NOT_ALIVE",
+    "215": "SUCCESS",
+    "216": "NOT_ALIVE",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "NON_PARTICIPATING",
+    "221": "SUCCESS",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "NON_PARTICIPATING",
+    "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
     "230": "SUCCESS",
-    "231": "SUCCESS",
+    "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
+    "236": "NOT_ALIVE",
     "237": "SUCCESS",
-    "238": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "SUCCESS",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "NOT_ALIVE",
-    "247": "NOT_ALIVE",
+    "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "SUCCESS",
-    "252": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
-    "255": "NOT_ALIVE",
-    "26": "SUCCESS",
+    "255": "NON_PARTICIPATING",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -184,7 +184,7 @@
     "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
-    "33": "NOT_ALIVE",
+    "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -200,7 +200,7 @@
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NON_PARTICIPATING",
+    "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -216,18 +216,18 @@
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "62": "NOT_ALIVE",
+    "63": "NOT_ALIVE",
     "64": "SUCCESS",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
-    "69": "NON_PARTICIPATING",
-    "7": "NOT_ALIVE",
+    "69": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
     "71": "SUCCESS",
-    "72": "SUCCESS",
+    "72": "NON_PARTICIPATING",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "SUCCESS",
@@ -235,12 +235,12 @@
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
-    "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "8": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
-    "84": "NON_PARTICIPATING",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
@@ -251,10 +251,10 @@
     "91": "SUCCESS",
     "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "SUCCESS",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "SUCCESS",
+    "97": "NON_PARTICIPATING",
     "98": "SUCCESS",
     "99": "SUCCESS"
   },
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6749029,
-    6749058
+    6749161,
+    6749190
   ],
   "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": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6735770,
+  "last_allreduce_block": 6749137,
   "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 sha256:1cfabb2297324187f65f81dd689bb6c3ca35b47f034c6f48b01814360637d066
+oid sha256:a5d394d8303bfc3f7efdec292caeefad9bfe3aca07d747e412d39dcfc4ba699e
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd103c86dddb6fff0f116c4e10a831f1bebd1812940f25292c659fa31bc0b7c8
+oid sha256:9db9d64d7b85eb97d00b2280733ec941bd5e46e61f669d474608676f5245e1a0
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebaa0073d453093ff94213d99b1716efcec441b75e76d58a396417ce88534f34
+oid sha256:fc7230cfa689b9d6d006e6bc98a9643a793b117e6914bcb9037e6e25615caeeb
 size 4400216536