Run 6. Outer Step 35. Inner Step 5.

config.json

@@ -1,259 +1,259 @@
 {
-  "_name_or_path": "zeras141a/cabec1f5",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
-    "10": "NOT_ALIVE",
-    "100": "NON_PARTICIPATING",
+    "1": "SUCCESS",
+    "10": "FAIL",
+    "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NON_PARTICIPATING",
-    "103": "NOT_ALIVE",
+    "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
     "108": "NON_PARTICIPATING",
     "109": "NOT_ALIVE",
-    "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "11": "FAIL",
+    "110": "FAIL",
     "111": "SUCCESS",
-    "112": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
     "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "SUCCESS",
     "116": "NOT_ALIVE",
     "117": "NON_PARTICIPATING",
-    "118": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "SUCCESS",
-    "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
-    "124": "NON_PARTICIPATING",
+    "122": "SUCCESS",
+    "123": "SUCCESS",
+    "124": "FAIL",
     "125": "NOT_ALIVE",
-    "126": "SUCCESS",
-    "127": "NOT_ALIVE",
+    "126": "FAIL",
+    "127": "FAIL",
     "128": "NOT_ALIVE",
-    "129": "NOT_ALIVE",
+    "129": "SUCCESS",
     "13": "NOT_ALIVE",
-    "130": "SUCCESS",
+    "130": "FAIL",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "NON_PARTICIPATING",
-    "134": "NOT_ALIVE",
-    "135": "NON_PARTICIPATING",
-    "136": "NOT_ALIVE",
-    "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
+    "133": "SUCCESS",
+    "134": "SUCCESS",
+    "135": "SUCCESS",
+    "136": "NON_PARTICIPATING",
+    "137": "FAIL",
+    "138": "FAIL",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
     "140": "SUCCESS",
     "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
-    "144": "NOT_ALIVE",
-    "145": "NOT_ALIVE",
+    "143": "SUCCESS",
+    "144": "FAIL",
+    "145": "FAIL",
     "146": "NON_PARTICIPATING",
     "147": "SUCCESS",
-    "148": "SUCCESS",
-    "149": "NON_PARTICIPATING",
+    "148": "NON_PARTICIPATING",
+    "149": "FAIL",
     "15": "SUCCESS",
-    "150": "NOT_ALIVE",
+    "150": "SUCCESS",
     "151": "NON_PARTICIPATING",
-    "152": "NON_PARTICIPATING",
-    "153": "NOT_ALIVE",
-    "154": "NON_PARTICIPATING",
-    "155": "NON_PARTICIPATING",
+    "152": "SUCCESS",
+    "153": "FAIL",
+    "154": "SUCCESS",
+    "155": "FAIL",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
-    "16": "NOT_ALIVE",
-    "160": "NON_PARTICIPATING",
-    "161": "NOT_ALIVE",
+    "16": "FAIL",
+    "160": "SUCCESS",
+    "161": "SUCCESS",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
-    "166": "NON_PARTICIPATING",
+    "166": "SUCCESS",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "FAIL",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
-    "170": "NON_PARTICIPATING",
+    "170": "SUCCESS",
     "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NON_PARTICIPATING",
-    "175": "NON_PARTICIPATING",
+    "175": "FAIL",
     "176": "NOT_ALIVE",
     "177": "NON_PARTICIPATING",
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
-    "180": "NOT_ALIVE",
+    "180": "FAIL",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "NOT_ALIVE",
-    "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
-    "188": "SUCCESS",
-    "189": "NOT_ALIVE",
-    "19": "NOT_ALIVE",
-    "190": "NOT_ALIVE",
-    "191": "NON_PARTICIPATING",
+    "185": "NON_PARTICIPATING",
+    "186": "FAIL",
+    "187": "FAIL",
+    "188": "FAIL",
+    "189": "SUCCESS",
+    "19": "SUCCESS",
+    "190": "SUCCESS",
+    "191": "SUCCESS",
     "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
-    "199": "NOT_ALIVE",
+    "198": "NOT_ALIVE",
+    "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "NOT_ALIVE",
-    "200": "NON_PARTICIPATING",
-    "201": "NOT_ALIVE",
-    "202": "SUCCESS",
+    "20": "FAIL",
+    "200": "SUCCESS",
+    "201": "SUCCESS",
+    "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
-    "204": "NOT_ALIVE",
-    "205": "NON_PARTICIPATING",
-    "206": "NON_PARTICIPATING",
+    "204": "NON_PARTICIPATING",
+    "205": "SUCCESS",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
-    "21": "SUCCESS",
-    "210": "NOT_ALIVE",
+    "209": "SUCCESS",
+    "21": "FAIL",
+    "210": "SUCCESS",
     "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
+    "212": "SUCCESS",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "SUCCESS",
-    "217": "NON_PARTICIPATING",
+    "216": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NON_PARTICIPATING",
-    "219": "NOT_ALIVE",
+    "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "NOT_ALIVE",
+    "221": "FAIL",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "NON_PARTICIPATING",
     "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
-    "228": "NOT_ALIVE",
+    "227": "NOT_ALIVE",
+    "228": "SUCCESS",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "SUCCESS",
     "230": "SUCCESS",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
+    "233": "FAIL",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "NOT_ALIVE",
     "237": "NON_PARTICIPATING",
-    "238": "NOT_ALIVE",
-    "239": "NOT_ALIVE",
+    "238": "FAIL",
+    "239": "FAIL",
     "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
-    "241": "NON_PARTICIPATING",
-    "242": "FAIL",
+    "240": "FAIL",
+    "241": "SUCCESS",
+    "242": "SUCCESS",
     "243": "NOT_ALIVE",
-    "244": "NON_PARTICIPATING",
+    "244": "SUCCESS",
     "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "246": "SUCCESS",
     "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
-    "249": "NOT_ALIVE",
-    "25": "SUCCESS",
+    "249": "SUCCESS",
+    "25": "NOT_ALIVE",
     "250": "NOT_ALIVE",
-    "251": "NON_PARTICIPATING",
-    "252": "NOT_ALIVE",
+    "251": "FAIL",
+    "252": "FAIL",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
     "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
+    "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "NOT_ALIVE",
-    "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
+    "3": "FAIL",
+    "30": "SUCCESS",
+    "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
-    "35": "NON_PARTICIPATING",
-    "36": "NOT_ALIVE",
+    "35": "SUCCESS",
+    "36": "SUCCESS",
     "37": "NON_PARTICIPATING",
     "38": "NON_PARTICIPATING",
-    "39": "NOT_ALIVE",
+    "39": "SUCCESS",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
+    "40": "FAIL",
     "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
-    "43": "NON_PARTICIPATING",
+    "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
-    "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
-    "47": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
+    "46": "SUCCESS",
+    "47": "FAIL",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "SUCCESS",
     "51": "NOT_ALIVE",
-    "52": "NON_PARTICIPATING",
-    "53": "NOT_ALIVE",
-    "54": "NOT_ALIVE",
+    "52": "SUCCESS",
+    "53": "SUCCESS",
+    "54": "FAIL",
     "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
-    "57": "NON_PARTICIPATING",
-    "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "56": "FAIL",
+    "57": "NOT_ALIVE",
+    "58": "FAIL",
+    "59": "FAIL",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
+    "61": "SUCCESS",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
-    "66": "NOT_ALIVE",
+    "66": "SUCCESS",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
-    "69": "NOT_ALIVE",
+    "69": "SUCCESS",
     "7": "NOT_ALIVE",
-    "70": "NOT_ALIVE",
+    "70": "SUCCESS",
     "71": "SUCCESS",
-    "72": "NOT_ALIVE",
-    "73": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
+    "73": "SUCCESS",
     "74": "NOT_ALIVE",
-    "75": "NOT_ALIVE",
+    "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
-    "77": "NON_PARTICIPATING",
-    "78": "NON_PARTICIPATING",
+    "77": "FAIL",
+    "78": "FAIL",
     "79": "NON_PARTICIPATING",
     "8": "NOT_ALIVE",
-    "80": "SUCCESS",
+    "80": "FAIL",
     "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "NON_PARTICIPATING",
+    "83": "SUCCESS",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
-    "87": "NOT_ALIVE",
+    "87": "FAIL",
     "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
-    "9": "NOT_ALIVE",
+    "9": "SUCCESS",
     "90": "NOT_ALIVE",
-    "91": "NOT_ALIVE",
+    "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
-    "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
+    "95": "NON_PARTICIPATING",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "NON_PARTICIPATING",
     "99": "SUCCESS"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6151826,
-    6151859
+    6152075,
+    6152113
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 58,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6150699,
+  "last_allreduce_block": 6152019,
   "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:f61a2ef2901e18d982a18ac3e31f3344f62d0417f0a80c936d224bd1ab23738f
+oid sha256:6d593c80af178d0059d5c2962ad9f11fcd23c3e08e454afbcb8048a6d217a82d
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7c0a830a66accbd54a7cf0975e64f473376d6b09e3d7421b21083591011e706
+oid sha256:d75876443b814234d352bb753e419ac3986399e0f305407bd623c31c3ea5e51e
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5182d238af5553653d241940792bbe09adc7df2ac1c00de1a280b12ed5624444
+oid sha256:35cb6ac17e93f88cc6a86db055f4049910c60366d82322a9f4466eda28749db6
 size 4400216536