Run 9. Outer Step 3. Inner Step 5.

config.json

@@ -2,102 +2,102 @@
   "_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": "NOT_ALIVE",
-    "102": "NON_PARTICIPATING",
-    "103": "NOT_ALIVE",
+    "102": "FAIL",
+    "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
-    "107": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "NON_PARTICIPATING",
+    "109": "NOT_ALIVE",
     "11": "NON_PARTICIPATING",
     "110": "NOT_ALIVE",
     "111": "FAIL",
-    "112": "NOT_ALIVE",
-    "113": "NOT_ALIVE",
-    "114": "FAIL",
+    "112": "FAIL",
+    "113": "NON_PARTICIPATING",
+    "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
-    "117": "NON_PARTICIPATING",
+    "117": "FAIL",
     "118": "FAIL",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
-    "122": "FAIL",
-    "123": "FAIL",
+    "122": "NON_PARTICIPATING",
+    "123": "NOT_ALIVE",
     "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
-    "126": "NON_PARTICIPATING",
-    "127": "FAIL",
-    "128": "FAIL",
+    "126": "FAIL",
+    "127": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
     "129": "NOT_ALIVE",
     "13": "NON_PARTICIPATING",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
     "134": "NOT_ALIVE",
-    "135": "FAIL",
+    "135": "NON_PARTICIPATING",
     "136": "NOT_ALIVE",
     "137": "NON_PARTICIPATING",
     "138": "NOT_ALIVE",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
-    "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
-    "143": "NOT_ALIVE",
+    "141": "NOT_ALIVE",
+    "142": "FAIL",
+    "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
-    "147": "FAIL",
-    "148": "NON_PARTICIPATING",
+    "146": "FAIL",
+    "147": "NON_PARTICIPATING",
+    "148": "FAIL",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
-    "150": "FAIL",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
-    "153": "FAIL",
-    "154": "NON_PARTICIPATING",
+    "153": "NON_PARTICIPATING",
+    "154": "FAIL",
     "155": "NOT_ALIVE",
-    "156": "FAIL",
-    "157": "FAIL",
+    "156": "NON_PARTICIPATING",
+    "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
-    "16": "NOT_ALIVE",
+    "16": "FAIL",
     "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
-    "163": "FAIL",
+    "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
-    "165": "FAIL",
+    "165": "NOT_ALIVE",
     "166": "NON_PARTICIPATING",
-    "167": "FAIL",
-    "168": "FAIL",
-    "169": "FAIL",
+    "167": "NON_PARTICIPATING",
+    "168": "NON_PARTICIPATING",
+    "169": "NON_PARTICIPATING",
     "17": "NON_PARTICIPATING",
-    "170": "FAIL",
-    "171": "NOT_ALIVE",
+    "170": "NOT_ALIVE",
+    "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "FAIL",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "FAIL",
-    "178": "NON_PARTICIPATING",
+    "177": "NOT_ALIVE",
+    "178": "FAIL",
     "179": "NOT_ALIVE",
-    "18": "NOT_ALIVE",
-    "180": "NON_PARTICIPATING",
-    "181": "NON_PARTICIPATING",
-    "182": "NON_PARTICIPATING",
-    "183": "FAIL",
-    "184": "NON_PARTICIPATING",
-    "185": "NON_PARTICIPATING",
+    "18": "NON_PARTICIPATING",
+    "180": "FAIL",
+    "181": "FAIL",
+    "182": "NOT_ALIVE",
+    "183": "NON_PARTICIPATING",
+    "184": "FAIL",
+    "185": "NOT_ALIVE",
     "186": "NON_PARTICIPATING",
     "187": "NOT_ALIVE",
     "188": "FAIL",
@@ -105,158 +105,158 @@
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
-    "192": "FAIL",
-    "193": "NON_PARTICIPATING",
-    "194": "NON_PARTICIPATING",
-    "195": "NOT_ALIVE",
-    "196": "NON_PARTICIPATING",
+    "192": "NON_PARTICIPATING",
+    "193": "FAIL",
+    "194": "FAIL",
+    "195": "NON_PARTICIPATING",
+    "196": "FAIL",
     "197": "NON_PARTICIPATING",
-    "198": "NOT_ALIVE",
-    "199": "NON_PARTICIPATING",
-    "2": "FAIL",
-    "20": "NON_PARTICIPATING",
-    "200": "NON_PARTICIPATING",
+    "198": "FAIL",
+    "199": "FAIL",
+    "2": "NOT_ALIVE",
+    "20": "FAIL",
+    "200": "FAIL",
     "201": "FAIL",
     "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "203": "FAIL",
     "204": "NOT_ALIVE",
     "205": "FAIL",
-    "206": "FAIL",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
-    "209": "NON_PARTICIPATING",
-    "21": "FAIL",
-    "210": "FAIL",
-    "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
-    "213": "FAIL",
+    "208": "NON_PARTICIPATING",
+    "209": "FAIL",
+    "21": "NON_PARTICIPATING",
+    "210": "NON_PARTICIPATING",
+    "211": "FAIL",
+    "212": "FAIL",
+    "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "FAIL",
+    "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
-    "217": "FAIL",
-    "218": "FAIL",
-    "219": "NOT_ALIVE",
-    "22": "NON_PARTICIPATING",
-    "220": "FAIL",
+    "217": "NON_PARTICIPATING",
+    "218": "NOT_ALIVE",
+    "219": "FAIL",
+    "22": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "FAIL",
     "222": "FAIL",
-    "223": "NON_PARTICIPATING",
+    "223": "FAIL",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
-    "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "226": "FAIL",
+    "227": "FAIL",
     "228": "NON_PARTICIPATING",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
-    "235": "NOT_ALIVE",
-    "236": "NOT_ALIVE",
-    "237": "FAIL",
-    "238": "FAIL",
+    "235": "FAIL",
+    "236": "NON_PARTICIPATING",
+    "237": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
+    "240": "NON_PARTICIPATING",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "FAIL",
-    "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
-    "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
+    "246": "FAIL",
+    "247": "FAIL",
+    "248": "FAIL",
+    "249": "FAIL",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
-    "251": "FAIL",
-    "252": "FAIL",
-    "253": "FAIL",
-    "254": "FAIL",
-    "255": "FAIL",
-    "26": "NOT_ALIVE",
-    "27": "NON_PARTICIPATING",
-    "28": "NON_PARTICIPATING",
-    "29": "NON_PARTICIPATING",
-    "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
-    "31": "NON_PARTICIPATING",
+    "250": "FAIL",
+    "251": "NON_PARTICIPATING",
+    "252": "NON_PARTICIPATING",
+    "253": "NON_PARTICIPATING",
+    "254": "NON_PARTICIPATING",
+    "255": "NON_PARTICIPATING",
+    "26": "FAIL",
+    "27": "FAIL",
+    "28": "FAIL",
+    "29": "FAIL",
+    "3": "FAIL",
+    "30": "NOT_ALIVE",
+    "31": "FAIL",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
-    "35": "NON_PARTICIPATING",
+    "35": "FAIL",
     "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
-    "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
-    "4": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
+    "39": "FAIL",
+    "4": "FAIL",
     "40": "NON_PARTICIPATING",
     "41": "NOT_ALIVE",
     "42": "NON_PARTICIPATING",
-    "43": "FAIL",
-    "44": "NON_PARTICIPATING",
+    "43": "NON_PARTICIPATING",
+    "44": "FAIL",
     "45": "NON_PARTICIPATING",
     "46": "NON_PARTICIPATING",
-    "47": "NON_PARTICIPATING",
-    "48": "NOT_ALIVE",
+    "47": "FAIL",
+    "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
     "5": "NON_PARTICIPATING",
-    "50": "NOT_ALIVE",
-    "51": "NON_PARTICIPATING",
-    "52": "NOT_ALIVE",
+    "50": "NON_PARTICIPATING",
+    "51": "FAIL",
+    "52": "NON_PARTICIPATING",
     "53": "FAIL",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "FAIL",
+    "56": "NOT_ALIVE",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
-    "59": "NOT_ALIVE",
-    "6": "NOT_ALIVE",
+    "59": "NON_PARTICIPATING",
+    "6": "NON_PARTICIPATING",
     "60": "NON_PARTICIPATING",
-    "61": "NON_PARTICIPATING",
-    "62": "NOT_ALIVE",
-    "63": "NOT_ALIVE",
-    "64": "NON_PARTICIPATING",
+    "61": "FAIL",
+    "62": "NON_PARTICIPATING",
+    "63": "NON_PARTICIPATING",
+    "64": "FAIL",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
-    "68": "NOT_ALIVE",
-    "69": "FAIL",
+    "67": "NOT_ALIVE",
+    "68": "NON_PARTICIPATING",
+    "69": "NON_PARTICIPATING",
     "7": "NON_PARTICIPATING",
-    "70": "NOT_ALIVE",
+    "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
     "72": "NON_PARTICIPATING",
     "73": "NOT_ALIVE",
     "74": "NON_PARTICIPATING",
     "75": "NOT_ALIVE",
-    "76": "NOT_ALIVE",
+    "76": "NON_PARTICIPATING",
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
-    "79": "NON_PARTICIPATING",
+    "79": "FAIL",
     "8": "FAIL",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NON_PARTICIPATING",
-    "83": "FAIL",
+    "83": "NON_PARTICIPATING",
     "84": "FAIL",
     "85": "NON_PARTICIPATING",
-    "86": "NOT_ALIVE",
-    "87": "NON_PARTICIPATING",
+    "86": "NON_PARTICIPATING",
+    "87": "FAIL",
     "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NON_PARTICIPATING",
-    "90": "FAIL",
-    "91": "FAIL",
-    "92": "NON_PARTICIPATING",
-    "93": "NOT_ALIVE",
+    "90": "NOT_ALIVE",
+    "91": "NON_PARTICIPATING",
+    "92": "NOT_ALIVE",
+    "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "FAIL",
-    "97": "FAIL",
+    "96": "NON_PARTICIPATING",
+    "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
-    "99": "FAIL"
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,18 +264,18 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6450041,
-    6450052,
-    6450064
+    6450112,
+    6450120,
+    6450150
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 79,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6449487,
+  "last_allreduce_block": 6450080,
   "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 sha256:470174bc460e7e0a3f22d9c26de852ef88ffa0218ae89fbcb905f3de5dbc5b4c
+oid sha256:7881fd7291527414e9e4c24ed03e72504163c8a621fb36cfb0b2afd43458964c
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d0598805af84890aafc35ed8016db566492d80d1fe9db25239c6a6a30c7f7169
+oid sha256:321254d802d74f2191dbb8559e948f1d6253137389a8805f21ded4ec44521786
 size 4924926842

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a2a11d2214a6ef78377089901393551c5ef14fbeeb7e099db3efd2061b678b2
+oid sha256:ffb8d6494858e6863dc366bcda34a7d85c586ba0965416547d8d8a79bb06b94c
 size 4400216536