Run 11. Outer Step 336. Inner Step 9.

config.json

@@ -1,22 +1,22 @@
 {
-  "_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",
     "103": "NOT_ALIVE",
     "104": "SUCCESS",
-    "105": "NOT_ALIVE",
+    "105": "SUCCESS",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
+    "107": "NON_PARTICIPATING",
     "108": "SUCCESS",
-    "109": "NOT_ALIVE",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
-    "111": "NOT_ALIVE",
+    "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
@@ -24,25 +24,25 @@
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "SUCCESS",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
     "128": "SUCCESS",
-    "129": "NON_PARTICIPATING",
-    "13": "SUCCESS",
+    "129": "NOT_ALIVE",
+    "13": "NOT_ALIVE",
     "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "SUCCESS",
-    "135": "SUCCESS",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -51,18 +51,18 @@
     "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": "SUCCESS",
     "149": "NOT_ALIVE",
-    "15": "SUCCESS",
+    "15": "NOT_ALIVE",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -79,7 +79,7 @@
     "166": "NOT_ALIVE",
     "167": "SUCCESS",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
@@ -89,7 +89,7 @@
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
+    "178": "SUCCESS",
     "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -108,7 +108,7 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "SUCCESS",
@@ -121,29 +121,29 @@
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
+    "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "SUCCESS",
+    "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
+    "216": "NOT_ALIVE",
     "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "NOT_ALIVE",
+    "221": "SUCCESS",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "NON_PARTICIPATING",
+    "226": "SUCCESS",
     "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
@@ -154,37 +154,37 @@
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
+    "236": "NOT_ALIVE",
     "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NON_PARTICIPATING",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
+    "245": "NOT_ALIVE",
     "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
+    "251": "SUCCESS",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NON_PARTICIPATING",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NON_PARTICIPATING",
+    "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
-    "33": "NOT_ALIVE",
+    "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -210,13 +210,13 @@
     "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
-    "59": "SUCCESS",
+    "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "SUCCESS",
+    "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
@@ -226,26 +226,26 @@
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NON_PARTICIPATING",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NON_PARTICIPATING",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "SUCCESS",
+    "79": "NON_PARTICIPATING",
     "8": "SUCCESS",
-    "80": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
+    "83": "SUCCESS",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
     "91": "SUCCESS",
@@ -254,7 +254,7 @@
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NON_PARTICIPATING",
+    "97": "NOT_ALIVE",
     "98": "SUCCESS",
     "99": "SUCCESS"
   },
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6716580,
-    6716609
+    6716712,
+    6716756
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 31,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6713550,
+  "last_allreduce_block": 6716676,
   "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:c0748fb0dc29f42c1293c201acfe3c829edb2c5e08b27b8f81f0295eea1fd0c4
+oid sha256:17d25b8dd82bf84327902ae3cdc11c5c582d0957d61f39e72d6e58b12cbffcd5
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2106a7226a8bfa0f162515f69585998dbdffc6e55bbc9969b062c3598d049cd4
+oid sha256:1686f80ffd3bc76b7bcf9436ebda9702b4adbeb29f2e7a357abd02b23a1a4d39
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9dc51cd2380749b14ebe5404d41fce830bdf5bbc9d4305ee2d279869e30a973b
+oid sha256:d63dd885c72fa1d47e11e00ba94ef1f4587231c3863f6228f7a31db340a265b8
 size 4400216536