Run 11. Outer Step 230. Inner Step 9.

config.json

@@ -2,84 +2,84 @@
   "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
-    "10": "FAIL",
+    "1": "NOT_ALIVE",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
-    "103": "FAIL",
-    "104": "NON_PARTICIPATING",
+    "103": "NOT_ALIVE",
+    "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
-    "106": "FAIL",
+    "106": "SUCCESS",
     "107": "NON_PARTICIPATING",
-    "108": "FAIL",
-    "109": "NOT_ALIVE",
+    "108": "NOT_ALIVE",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "SUCCESS",
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NON_PARTICIPATING",
+    "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "FAIL",
+    "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "FAIL",
-    "128": "NON_PARTICIPATING",
+    "127": "SUCCESS",
+    "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
-    "13": "FAIL",
-    "130": "FAIL",
+    "13": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NOT_ALIVE",
-    "135": "NOT_ALIVE",
+    "134": "SUCCESS",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "FAIL",
-    "141": "NON_PARTICIPATING",
+    "140": "SUCCESS",
+    "141": "SUCCESS",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
-    "147": "FAIL",
-    "148": "NON_PARTICIPATING",
+    "146": "SUCCESS",
+    "147": "NON_PARTICIPATING",
+    "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "FAIL",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "FAIL",
+    "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "FAIL",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
-    "162": "FAIL",
+    "162": "SUCCESS",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "FAIL",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
@@ -89,94 +89,94 @@
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NON_PARTICIPATING",
-    "179": "NON_PARTICIPATING",
+    "178": "SUCCESS",
+    "179": "SUCCESS",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "NON_PARTICIPATING",
-    "185": "FAIL",
-    "186": "FAIL",
-    "187": "FAIL",
+    "184": "SUCCESS",
+    "185": "NON_PARTICIPATING",
+    "186": "SUCCESS",
+    "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "FAIL",
+    "193": "NON_PARTICIPATING",
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "FAIL",
+    "197": "SUCCESS",
     "198": "NOT_ALIVE",
-    "199": "NOT_ALIVE",
+    "199": "NON_PARTICIPATING",
     "2": "NOT_ALIVE",
-    "20": "FAIL",
-    "200": "NOT_ALIVE",
-    "201": "NOT_ALIVE",
+    "20": "NOT_ALIVE",
+    "200": "NON_PARTICIPATING",
+    "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "FAIL",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "NON_PARTICIPATING",
+    "210": "FAIL",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "FAIL",
+    "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "SUCCESS",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "FAIL",
-    "227": "NON_PARTICIPATING",
+    "226": "SUCCESS",
+    "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
-    "230": "FAIL",
-    "231": "FAIL",
+    "23": "SUCCESS",
+    "230": "NOT_ALIVE",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "FAIL",
-    "234": "NOT_ALIVE",
+    "233": "SUCCESS",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "FAIL",
-    "237": "NON_PARTICIPATING",
-    "238": "FAIL",
+    "236": "NON_PARTICIPATING",
+    "237": "SUCCESS",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "FAIL",
+    "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "FAIL",
+    "241": "SUCCESS",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "FAIL",
+    "245": "SUCCESS",
     "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "FAIL",
-    "252": "FAIL",
+    "251": "NON_PARTICIPATING",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "FAIL",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -185,78 +185,78 @@
     "31": "NOT_ALIVE",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "FAIL",
+    "34": "NON_PARTICIPATING",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "NOT_ALIVE",
-    "38": "FAIL",
+    "37": "SUCCESS",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
-    "4": "FAIL",
+    "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
-    "41": "NON_PARTICIPATING",
+    "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
-    "45": "NON_PARTICIPATING",
+    "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
-    "51": "NOT_ALIVE",
+    "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "FAIL",
+    "54": "SUCCESS",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "NON_PARTICIPATING",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "FAIL",
-    "62": "NOT_ALIVE",
-    "63": "NOT_ALIVE",
+    "61": "SUCCESS",
+    "62": "NON_PARTICIPATING",
+    "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
-    "65": "FAIL",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "FAIL",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "FAIL",
+    "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "FAIL",
+    "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
-    "8": "FAIL",
-    "80": "FAIL",
+    "79": "SUCCESS",
+    "8": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "FAIL",
-    "83": "FAIL",
-    "84": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
+    "83": "SUCCESS",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
-    "88": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "FAIL",
+    "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
-    "97": "NOT_ALIVE",
-    "98": "FAIL",
-    "99": "FAIL"
+    "97": "SUCCESS",
+    "98": "NON_PARTICIPATING",
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6670816,
-    6670848
+    6670914,
+    6670947
   ],
   "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": 6670454,
+  "last_allreduce_block": 6670848,
   "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:f960571a2de876e50668e5d5520c394e586138632cb087d9b6c941443f2ec79d
+oid sha256:e33397cfdc8a679843f6cc8daa04009559ddad86beb9e6e7ba61da3633469854
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7406c2991c754115129b5fd7f393d142c53393f11f2c3dfd93ae057ca9a8b9e3
+oid sha256:a915c5d6008fb5af2dbca0264395a7e563389e96044e99671842097943888c7b
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:34a423b09d52a5a3e9351f806c01636807f709e82e6b4d19c5edd04a08b0c7e7
+oid sha256:6010194ad242c9ebec6e7e0f17d11ea0d10751b35a2a51bcdb0b3993bd1e3a00
 size 4400216536