Run 5. Outer Step 74. Inner Step 1.

config.json

@@ -1,85 +1,85 @@
 {
-  "_name_or_path": "ncaagcc/tensor13",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NON_PARTICIPATING",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
-    "102": "NON_PARTICIPATING",
-    "103": "SUCCESS",
-    "104": "NON_PARTICIPATING",
+    "102": "NOT_ALIVE",
+    "103": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
-    "106": "NON_PARTICIPATING",
+    "106": "NOT_ALIVE",
     "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
-    "109": "NOT_ALIVE",
-    "11": "NOT_ALIVE",
-    "110": "SUCCESS",
+    "109": "NON_PARTICIPATING",
+    "11": "NON_PARTICIPATING",
+    "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
-    "112": "NON_PARTICIPATING",
+    "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
-    "114": "NOT_ALIVE",
+    "114": "NON_PARTICIPATING",
     "115": "NOT_ALIVE",
-    "116": "NON_PARTICIPATING",
-    "117": "NON_PARTICIPATING",
-    "118": "SUCCESS",
+    "116": "NOT_ALIVE",
+    "117": "SUCCESS",
+    "118": "NON_PARTICIPATING",
     "119": "NOT_ALIVE",
     "12": "NON_PARTICIPATING",
-    "120": "NON_PARTICIPATING",
-    "121": "NOT_ALIVE",
-    "122": "SUCCESS",
-    "123": "SUCCESS",
+    "120": "NOT_ALIVE",
+    "121": "NON_PARTICIPATING",
+    "122": "NON_PARTICIPATING",
+    "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
-    "125": "NON_PARTICIPATING",
+    "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
     "127": "NON_PARTICIPATING",
     "128": "NOT_ALIVE",
     "129": "NON_PARTICIPATING",
     "13": "NON_PARTICIPATING",
-    "130": "SUCCESS",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
-    "132": "NOT_ALIVE",
-    "133": "SUCCESS",
-    "134": "NOT_ALIVE",
-    "135": "NOT_ALIVE",
-    "136": "SUCCESS",
-    "137": "NON_PARTICIPATING",
-    "138": "NON_PARTICIPATING",
-    "139": "NON_PARTICIPATING",
-    "14": "NON_PARTICIPATING",
-    "140": "NON_PARTICIPATING",
-    "141": "NOT_ALIVE",
-    "142": "NOT_ALIVE",
+    "132": "NON_PARTICIPATING",
+    "133": "NON_PARTICIPATING",
+    "134": "NON_PARTICIPATING",
+    "135": "NON_PARTICIPATING",
+    "136": "NOT_ALIVE",
+    "137": "NOT_ALIVE",
+    "138": "SUCCESS",
+    "139": "NOT_ALIVE",
+    "14": "NOT_ALIVE",
+    "140": "NOT_ALIVE",
+    "141": "NON_PARTICIPATING",
+    "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "SUCCESS",
+    "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
     "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
-    "153": "NOT_ALIVE",
-    "154": "NON_PARTICIPATING",
+    "153": "SUCCESS",
+    "154": "NOT_ALIVE",
     "155": "NON_PARTICIPATING",
-    "156": "SUCCESS",
-    "157": "NON_PARTICIPATING",
-    "158": "NON_PARTICIPATING",
-    "159": "NOT_ALIVE",
+    "156": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
+    "158": "NOT_ALIVE",
+    "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
-    "164": "SUCCESS",
+    "164": "NOT_ALIVE",
     "165": "SUCCESS",
     "166": "NON_PARTICIPATING",
-    "167": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
     "17": "NON_PARTICIPATING",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
@@ -88,31 +88,31 @@
     "174": "NON_PARTICIPATING",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
-    "177": "SUCCESS",
-    "178": "NOT_ALIVE",
-    "179": "NOT_ALIVE",
-    "18": "SUCCESS",
-    "180": "NOT_ALIVE",
+    "177": "NON_PARTICIPATING",
+    "178": "NON_PARTICIPATING",
+    "179": "NON_PARTICIPATING",
+    "18": "NON_PARTICIPATING",
+    "180": "NON_PARTICIPATING",
     "181": "SUCCESS",
-    "182": "NOT_ALIVE",
+    "182": "NON_PARTICIPATING",
     "183": "NON_PARTICIPATING",
-    "184": "NON_PARTICIPATING",
+    "184": "NOT_ALIVE",
     "185": "NON_PARTICIPATING",
-    "186": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
-    "190": "NOT_ALIVE",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
-    "194": "NON_PARTICIPATING",
+    "193": "NOT_ALIVE",
+    "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NON_PARTICIPATING",
-    "198": "NON_PARTICIPATING",
-    "199": "NON_PARTICIPATING",
+    "198": "NOT_ALIVE",
+    "199": "NOT_ALIVE",
     "2": "NON_PARTICIPATING",
     "20": "NOT_ALIVE",
     "200": "NON_PARTICIPATING",
@@ -120,135 +120,135 @@
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NON_PARTICIPATING",
-    "205": "NON_PARTICIPATING",
-    "206": "SUCCESS",
+    "205": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
+    "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
     "21": "SUCCESS",
-    "210": "SUCCESS",
-    "211": "NON_PARTICIPATING",
+    "210": "NON_PARTICIPATING",
+    "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
-    "213": "NOT_ALIVE",
+    "213": "SUCCESS",
     "214": "NON_PARTICIPATING",
-    "215": "SUCCESS",
-    "216": "NON_PARTICIPATING",
-    "217": "NOT_ALIVE",
-    "218": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
+    "216": "NOT_ALIVE",
+    "217": "NON_PARTICIPATING",
+    "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
     "22": "NON_PARTICIPATING",
     "220": "NON_PARTICIPATING",
-    "221": "SUCCESS",
+    "221": "NOT_ALIVE",
     "222": "NON_PARTICIPATING",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
-    "225": "SUCCESS",
+    "225": "NON_PARTICIPATING",
     "226": "NOT_ALIVE",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
-    "230": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
-    "232": "NON_PARTICIPATING",
-    "233": "NOT_ALIVE",
-    "234": "NOT_ALIVE",
+    "232": "NOT_ALIVE",
+    "233": "NON_PARTICIPATING",
+    "234": "NON_PARTICIPATING",
     "235": "NOT_ALIVE",
-    "236": "NON_PARTICIPATING",
-    "237": "NOT_ALIVE",
+    "236": "NOT_ALIVE",
+    "237": "NON_PARTICIPATING",
     "238": "SUCCESS",
     "239": "NON_PARTICIPATING",
-    "24": "SUCCESS",
+    "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
-    "241": "NOT_ALIVE",
-    "242": "SUCCESS",
+    "241": "NON_PARTICIPATING",
+    "242": "NOT_ALIVE",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "NOT_ALIVE",
+    "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
-    "254": "NON_PARTICIPATING",
+    "254": "NOT_ALIVE",
     "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
-    "27": "SUCCESS",
+    "26": "NOT_ALIVE",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
-    "29": "NOT_ALIVE",
+    "29": "NON_PARTICIPATING",
     "3": "NON_PARTICIPATING",
     "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
-    "32": "SUCCESS",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
+    "34": "NOT_ALIVE",
     "35": "SUCCESS",
     "36": "NON_PARTICIPATING",
     "37": "NON_PARTICIPATING",
     "38": "SUCCESS",
-    "39": "NOT_ALIVE",
-    "4": "NON_PARTICIPATING",
+    "39": "SUCCESS",
+    "4": "NOT_ALIVE",
     "40": "SUCCESS",
-    "41": "SUCCESS",
+    "41": "NOT_ALIVE",
     "42": "NON_PARTICIPATING",
-    "43": "SUCCESS",
+    "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
-    "45": "NON_PARTICIPATING",
+    "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
-    "5": "NON_PARTICIPATING",
-    "50": "NOT_ALIVE",
+    "5": "NOT_ALIVE",
+    "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NOT_ALIVE",
     "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
-    "56": "NOT_ALIVE",
+    "55": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
     "57": "NON_PARTICIPATING",
-    "58": "SUCCESS",
-    "59": "NON_PARTICIPATING",
-    "6": "NOT_ALIVE",
+    "58": "NON_PARTICIPATING",
+    "59": "SUCCESS",
+    "6": "NON_PARTICIPATING",
     "60": "NON_PARTICIPATING",
     "61": "NOT_ALIVE",
     "62": "NOT_ALIVE",
-    "63": "NON_PARTICIPATING",
+    "63": "NOT_ALIVE",
     "64": "NON_PARTICIPATING",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
-    "68": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
+    "68": "NOT_ALIVE",
     "69": "NON_PARTICIPATING",
     "7": "NON_PARTICIPATING",
-    "70": "NON_PARTICIPATING",
-    "71": "NOT_ALIVE",
+    "70": "NOT_ALIVE",
+    "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
     "74": "NON_PARTICIPATING",
-    "75": "SUCCESS",
+    "75": "NON_PARTICIPATING",
     "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
     "80": "NON_PARTICIPATING",
-    "81": "NON_PARTICIPATING",
-    "82": "NOT_ALIVE",
+    "81": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
     "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
-    "86": "NOT_ALIVE",
-    "87": "NON_PARTICIPATING",
-    "88": "NOT_ALIVE",
-    "89": "NON_PARTICIPATING",
+    "86": "NON_PARTICIPATING",
+    "87": "NOT_ALIVE",
+    "88": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
     "9": "NON_PARTICIPATING",
-    "90": "SUCCESS",
-    "91": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
+    "91": "NON_PARTICIPATING",
     "92": "NON_PARTICIPATING",
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    5911727,
-    5911727
+    5917760,
+    5917803
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 28,
+  "inner_step": 1,
   "intermediate_size": 5632,
-  "last_allreduce_block": 5910694,
+  "last_allreduce_block": 5917724,
   "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:aa8ca723e8d4e3170012316e7da95dd845532deac725b65c0a5ce20fd83b7594
+oid sha256:dcd5f17a92be98be90135a5a94b3982720d04b132f57dab8f0558bf4c646976a
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3c974c0119be874e548dfd98ce0c408555b2f5b32574163711878075c831b2b
+oid sha256:39ac8bc19cd377ce48ce1afe668f9de9a3b668cfc28bdbf8849dba0f8fb98fb3
 size 4400216536