Run 5. Outer Step 86. Inner Step 1.

config.json

@@ -2,7 +2,7 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
+    "1": "NOT_ALIVE",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
@@ -12,28 +12,28 @@
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
     "107": "NOT_ALIVE",
-    "108": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
     "109": "NON_PARTICIPATING",
     "11": "SUCCESS",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
-    "114": "NON_PARTICIPATING",
+    "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
-    "118": "NOT_ALIVE",
+    "118": "NON_PARTICIPATING",
     "119": "NOT_ALIVE",
-    "12": "NOT_ALIVE",
+    "12": "NON_PARTICIPATING",
     "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
+    "123": "NOT_ALIVE",
     "124": "NON_PARTICIPATING",
-    "125": "NOT_ALIVE",
+    "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
-    "127": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "NON_PARTICIPATING",
@@ -41,49 +41,49 @@
     "131": "NOT_ALIVE",
     "132": "NON_PARTICIPATING",
     "133": "NON_PARTICIPATING",
-    "134": "NOT_ALIVE",
+    "134": "NON_PARTICIPATING",
     "135": "NON_PARTICIPATING",
     "136": "NON_PARTICIPATING",
-    "137": "SUCCESS",
-    "138": "NON_PARTICIPATING",
+    "137": "NON_PARTICIPATING",
+    "138": "SUCCESS",
     "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
+    "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "NOT_ALIVE",
+    "149": "NON_PARTICIPATING",
     "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
     "153": "SUCCESS",
-    "154": "NON_PARTICIPATING",
+    "154": "NOT_ALIVE",
     "155": "NON_PARTICIPATING",
     "156": "SUCCESS",
-    "157": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
-    "164": "NON_PARTICIPATING",
+    "164": "SUCCESS",
     "165": "SUCCESS",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "SUCCESS",
     "169": "NOT_ALIVE",
     "17": "NON_PARTICIPATING",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
-    "172": "SUCCESS",
+    "172": "NON_PARTICIPATING",
     "173": "SUCCESS",
     "174": "NOT_ALIVE",
     "175": "NOT_ALIVE",
@@ -93,7 +93,7 @@
     "179": "NON_PARTICIPATING",
     "18": "NON_PARTICIPATING",
     "180": "NOT_ALIVE",
-    "181": "NON_PARTICIPATING",
+    "181": "SUCCESS",
     "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
     "184": "NON_PARTICIPATING",
@@ -106,32 +106,32 @@
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
+    "193": "NOT_ALIVE",
     "194": "NON_PARTICIPATING",
     "195": "NON_PARTICIPATING",
     "196": "NON_PARTICIPATING",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
-    "199": "NOT_ALIVE",
+    "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
     "20": "NOT_ALIVE",
-    "200": "NON_PARTICIPATING",
+    "200": "NOT_ALIVE",
     "201": "NON_PARTICIPATING",
     "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
     "204": "NON_PARTICIPATING",
-    "205": "NOT_ALIVE",
+    "205": "NON_PARTICIPATING",
     "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
-    "21": "SUCCESS",
-    "210": "NOT_ALIVE",
+    "21": "NON_PARTICIPATING",
+    "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
-    "212": "NOT_ALIVE",
+    "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
@@ -142,100 +142,100 @@
     "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
-    "225": "NON_PARTICIPATING",
+    "225": "SUCCESS",
     "226": "NOT_ALIVE",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
     "23": "NON_PARTICIPATING",
-    "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
-    "232": "NOT_ALIVE",
-    "233": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
+    "231": "NOT_ALIVE",
+    "232": "NON_PARTICIPATING",
+    "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "NOT_ALIVE",
+    "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
     "24": "NON_PARTICIPATING",
     "240": "NON_PARTICIPATING",
     "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
-    "243": "NOT_ALIVE",
+    "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
-    "247": "NOT_ALIVE",
-    "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "247": "NON_PARTICIPATING",
+    "248": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
     "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
-    "253": "NON_PARTICIPATING",
-    "254": "NON_PARTICIPATING",
+    "253": "NOT_ALIVE",
+    "254": "NOT_ALIVE",
     "255": "NON_PARTICIPATING",
     "26": "NON_PARTICIPATING",
     "27": "NON_PARTICIPATING",
-    "28": "NON_PARTICIPATING",
+    "28": "NOT_ALIVE",
     "29": "NON_PARTICIPATING",
-    "3": "NON_PARTICIPATING",
-    "30": "SUCCESS",
+    "3": "NOT_ALIVE",
+    "30": "NON_PARTICIPATING",
     "31": "NON_PARTICIPATING",
-    "32": "SUCCESS",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
-    "35": "SUCCESS",
+    "34": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
     "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
-    "38": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
     "39": "SUCCESS",
     "4": "NON_PARTICIPATING",
     "40": "SUCCESS",
-    "41": "SUCCESS",
+    "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
-    "43": "NOT_ALIVE",
+    "43": "NON_PARTICIPATING",
     "44": "NON_PARTICIPATING",
     "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
-    "49": "NON_PARTICIPATING",
-    "5": "NON_PARTICIPATING",
+    "49": "NOT_ALIVE",
+    "5": "SUCCESS",
     "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
+    "51": "NOT_ALIVE",
     "52": "NON_PARTICIPATING",
     "53": "NON_PARTICIPATING",
     "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
+    "55": "NOT_ALIVE",
+    "56": "SUCCESS",
     "57": "NON_PARTICIPATING",
     "58": "NON_PARTICIPATING",
-    "59": "SUCCESS",
+    "59": "NON_PARTICIPATING",
     "6": "NON_PARTICIPATING",
     "60": "NON_PARTICIPATING",
     "61": "NON_PARTICIPATING",
     "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
     "64": "NON_PARTICIPATING",
-    "65": "NON_PARTICIPATING",
+    "65": "NOT_ALIVE",
     "66": "NON_PARTICIPATING",
     "67": "NOT_ALIVE",
     "68": "NON_PARTICIPATING",
-    "69": "NOT_ALIVE",
-    "7": "NON_PARTICIPATING",
+    "69": "NON_PARTICIPATING",
+    "7": "SUCCESS",
     "70": "NON_PARTICIPATING",
-    "71": "NOT_ALIVE",
+    "71": "NON_PARTICIPATING",
     "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
     "74": "NON_PARTICIPATING",
     "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
-    "77": "NON_PARTICIPATING",
-    "78": "NOT_ALIVE",
+    "77": "NOT_ALIVE",
+    "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
-    "8": "NON_PARTICIPATING",
+    "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
     "81": "NON_PARTICIPATING",
     "82": "NON_PARTICIPATING",
@@ -244,18 +244,18 @@
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NON_PARTICIPATING",
-    "88": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
     "9": "NON_PARTICIPATING",
     "90": "NON_PARTICIPATING",
-    "91": "NOT_ALIVE",
+    "91": "NON_PARTICIPATING",
     "92": "NON_PARTICIPATING",
     "93": "NON_PARTICIPATING",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NOT_ALIVE"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    5933693,
-    5933712
+    5933786,
+    5933802
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 41,
+  "inner_step": 1,
   "intermediate_size": 5632,
-  "last_allreduce_block": 5932508,
+  "last_allreduce_block": 5933681,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8c2ff437c45e3abe3e0b650098f980604950cac42f6e134ce19f4bcc394741d
+oid sha256:a1f3ea93b1484a780c834db3ff9cb75ac709c62ca55c4159057eb9b47fd4a6ed
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec1ad0483252ebe58631dba58437365171983cdf51229ab53733aef187b9cc36
+oid sha256:0d3192a7c2b080ffb7154dc383cf12de7100682cc5fa163f725018c221fda175
 size 4400216536