Run 10. Outer Step 115. Inner Step 5.

config.json

@@ -1,42 +1,42 @@
 {
-  "_name_or_path": "ncaagcc/mja1",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
-    "10": "NON_PARTICIPATING",
+    "1": "SUCCESS",
+    "10": "FAIL",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "NOT_ALIVE",
     "102": "SUCCESS",
-    "103": "SUCCESS",
+    "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NON_PARTICIPATING",
-    "106": "SUCCESS",
+    "105": "SUCCESS",
+    "106": "NON_PARTICIPATING",
     "107": "SUCCESS",
     "108": "NON_PARTICIPATING",
     "109": "NON_PARTICIPATING",
-    "11": "NON_PARTICIPATING",
+    "11": "SUCCESS",
     "110": "NOT_ALIVE",
-    "111": "SUCCESS",
-    "112": "SUCCESS",
-    "113": "SUCCESS",
-    "114": "SUCCESS",
-    "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
-    "117": "SUCCESS",
-    "118": "SUCCESS",
+    "111": "NOT_ALIVE",
+    "112": "NOT_ALIVE",
+    "113": "NOT_ALIVE",
+    "114": "NON_PARTICIPATING",
+    "115": "SUCCESS",
+    "116": "SUCCESS",
+    "117": "NOT_ALIVE",
+    "118": "NOT_ALIVE",
     "119": "SUCCESS",
-    "12": "NON_PARTICIPATING",
+    "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "121": "SUCCESS",
     "122": "NON_PARTICIPATING",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
-    "125": "SUCCESS",
-    "126": "SUCCESS",
+    "125": "NON_PARTICIPATING",
+    "126": "NOT_ALIVE",
     "127": "NON_PARTICIPATING",
-    "128": "SUCCESS",
-    "129": "NON_PARTICIPATING",
-    "13": "SUCCESS",
+    "128": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
+    "13": "NON_PARTICIPATING",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
@@ -46,41 +46,41 @@
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
-    "139": "NON_PARTICIPATING",
+    "139": "SUCCESS",
     "14": "NOT_ALIVE",
     "140": "SUCCESS",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
     "143": "NOT_ALIVE",
-    "144": "NON_PARTICIPATING",
+    "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "NON_PARTICIPATING",
+    "148": "NOT_ALIVE",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "SUCCESS",
     "154": "NON_PARTICIPATING",
     "155": "NOT_ALIVE",
     "156": "NON_PARTICIPATING",
-    "157": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
     "159": "SUCCESS",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
-    "161": "SUCCESS",
+    "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
-    "163": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "SUCCESS",
-    "167": "NON_PARTICIPATING",
-    "168": "SUCCESS",
-    "169": "NON_PARTICIPATING",
-    "17": "NON_PARTICIPATING",
+    "166": "NOT_ALIVE",
+    "167": "NOT_ALIVE",
+    "168": "NON_PARTICIPATING",
+    "169": "NOT_ALIVE",
+    "17": "NOT_ALIVE",
     "170": "SUCCESS",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
@@ -89,75 +89,75 @@
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
-    "184": "FAIL",
+    "184": "NOT_ALIVE",
     "185": "SUCCESS",
     "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
-    "188": "SUCCESS",
+    "187": "SUCCESS",
+    "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
-    "190": "NOT_ALIVE",
+    "190": "SUCCESS",
     "191": "NOT_ALIVE",
-    "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
+    "192": "NON_PARTICIPATING",
+    "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
-    "196": "SUCCESS",
+    "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "SUCCESS",
-    "199": "NON_PARTICIPATING",
+    "198": "NON_PARTICIPATING",
+    "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
+    "20": "NON_PARTICIPATING",
     "200": "SUCCESS",
     "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "SUCCESS",
-    "206": "NOT_ALIVE",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
-    "21": "NOT_ALIVE",
+    "209": "NOT_ALIVE",
+    "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
-    "211": "NOT_ALIVE",
-    "212": "NON_PARTICIPATING",
+    "211": "SUCCESS",
+    "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
-    "214": "NOT_ALIVE",
+    "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
-    "221": "SUCCESS",
+    "220": "SUCCESS",
+    "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
-    "223": "NOT_ALIVE",
+    "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
     "227": "NOT_ALIVE",
-    "228": "SUCCESS",
+    "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
-    "230": "NON_PARTICIPATING",
+    "23": "SUCCESS",
+    "230": "SUCCESS",
     "231": "SUCCESS",
-    "232": "NON_PARTICIPATING",
-    "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
+    "232": "SUCCESS",
+    "233": "NOT_ALIVE",
+    "234": "NOT_ALIVE",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
+    "238": "SUCCESS",
+    "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
     "241": "NOT_ALIVE",
@@ -165,7 +165,7 @@
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
-    "246": "SUCCESS",
+    "246": "NOT_ALIVE",
     "247": "NOT_ALIVE",
     "248": "SUCCESS",
     "249": "NOT_ALIVE",
@@ -174,60 +174,60 @@
     "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
     "253": "NOT_ALIVE",
-    "254": "NON_PARTICIPATING",
-    "255": "NOT_ALIVE",
+    "254": "NOT_ALIVE",
+    "255": "SUCCESS",
     "26": "SUCCESS",
-    "27": "NOT_ALIVE",
-    "28": "SUCCESS",
+    "27": "SUCCESS",
+    "28": "NOT_ALIVE",
     "29": "NON_PARTICIPATING",
-    "3": "NOT_ALIVE",
-    "30": "NON_PARTICIPATING",
-    "31": "NON_PARTICIPATING",
+    "3": "SUCCESS",
+    "30": "NOT_ALIVE",
+    "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "SUCCESS",
-    "36": "NON_PARTICIPATING",
-    "37": "NOT_ALIVE",
+    "36": "NOT_ALIVE",
+    "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
+    "39": "NOT_ALIVE",
+    "4": "NOT_ALIVE",
     "40": "SUCCESS",
     "41": "NOT_ALIVE",
-    "42": "SUCCESS",
+    "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "SUCCESS",
-    "45": "NON_PARTICIPATING",
+    "45": "SUCCESS",
     "46": "SUCCESS",
     "47": "NOT_ALIVE",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
-    "52": "SUCCESS",
+    "51": "NOT_ALIVE",
+    "52": "NON_PARTICIPATING",
     "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NON_PARTICIPATING",
-    "57": "NOT_ALIVE",
+    "56": "NOT_ALIVE",
+    "57": "SUCCESS",
     "58": "NOT_ALIVE",
-    "59": "SUCCESS",
+    "59": "NOT_ALIVE",
     "6": "NON_PARTICIPATING",
     "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
+    "61": "NON_PARTICIPATING",
     "62": "NON_PARTICIPATING",
     "63": "SUCCESS",
-    "64": "NOT_ALIVE",
-    "65": "SUCCESS",
+    "64": "SUCCESS",
+    "65": "NOT_ALIVE",
     "66": "SUCCESS",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NON_PARTICIPATING",
-    "7": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
-    "72": "NON_PARTICIPATING",
+    "71": "NOT_ALIVE",
+    "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "SUCCESS",
     "75": "NOT_ALIVE",
@@ -236,27 +236,27 @@
     "78": "NOT_ALIVE",
     "79": "NON_PARTICIPATING",
     "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
+    "80": "SUCCESS",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
-    "83": "NOT_ALIVE",
-    "84": "SUCCESS",
-    "85": "NOT_ALIVE",
+    "82": "FAIL",
+    "83": "NON_PARTICIPATING",
+    "84": "NOT_ALIVE",
+    "85": "SUCCESS",
     "86": "SUCCESS",
-    "87": "NON_PARTICIPATING",
+    "87": "SUCCESS",
     "88": "NOT_ALIVE",
-    "89": "SUCCESS",
-    "9": "NON_PARTICIPATING",
+    "89": "NOT_ALIVE",
+    "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "SUCCESS",
-    "92": "SUCCESS",
+    "91": "NOT_ALIVE",
+    "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
     "94": "SUCCESS",
     "95": "NOT_ALIVE",
     "96": "SUCCESS",
-    "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
-    "99": "NON_PARTICIPATING"
+    "97": "NON_PARTICIPATING",
+    "98": "SUCCESS",
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6516715,
-    6516735
+    6527816,
+    6527823
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
@@ -274,7 +274,7 @@
   "initializer_range": 0.02,
   "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6516415,
+  "last_allreduce_block": 6527593,
   "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 sha256:450d3a2aca22fe2ba957f228cb6d20e71759d7f01106238e34ffe383fa6dd4cf
+oid sha256:da3d65c211fe8e14ba38b01d212667b9d7a06f7ba3d8c2b86ccb59535ff6ce09
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f97fdb837f95718ddede0c467951cc398cad2ea7768f2b0018f433b10f769910
+oid sha256:934e2a25d711b9d247927b3eb61fcfb233bc04435e38674a4580f6466abc3604
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:036c99a2131f6a0cab360a5761369589e532d25e7b50ad425af5b96cfeb19950
+oid sha256:a6316ab5ab7c20c93032c203da039c7ccbc2a3631a4b50886c24c58b9c02096c
 size 4400216536