Run 9. Outer Step 11. Inner Step 9.

config.json

@@ -1,48 +1,48 @@
 {
-  "_name_or_path": "MBA4215/m6",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "FAIL",
+    "1": "NON_PARTICIPATING",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
-    "104": "FAIL",
+    "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
-    "106": "NON_PARTICIPATING",
+    "106": "FAIL",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "FAIL",
-    "11": "NON_PARTICIPATING",
+    "109": "NON_PARTICIPATING",
+    "11": "FAIL",
     "110": "NOT_ALIVE",
-    "111": "FAIL",
-    "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
+    "111": "NON_PARTICIPATING",
+    "112": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
-    "116": "NOT_ALIVE",
-    "117": "NON_PARTICIPATING",
+    "116": "FAIL",
+    "117": "NOT_ALIVE",
     "118": "NON_PARTICIPATING",
     "119": "NOT_ALIVE",
     "12": "NON_PARTICIPATING",
-    "120": "NON_PARTICIPATING",
+    "120": "NOT_ALIVE",
     "121": "NOT_ALIVE",
-    "122": "FAIL",
-    "123": "NON_PARTICIPATING",
-    "124": "FAIL",
-    "125": "NON_PARTICIPATING",
+    "122": "NON_PARTICIPATING",
+    "123": "FAIL",
+    "124": "NON_PARTICIPATING",
+    "125": "FAIL",
     "126": "NON_PARTICIPATING",
-    "127": "FAIL",
-    "128": "NOT_ALIVE",
-    "129": "NON_PARTICIPATING",
+    "127": "NON_PARTICIPATING",
+    "128": "FAIL",
+    "129": "FAIL",
     "13": "FAIL",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
@@ -54,7 +54,7 @@
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NON_PARTICIPATING",
-    "146": "FAIL",
+    "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
     "148": "NOT_ALIVE",
     "149": "NON_PARTICIPATING",
@@ -64,83 +64,83 @@
     "152": "NOT_ALIVE",
     "153": "FAIL",
     "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
-    "156": "FAIL",
+    "155": "NON_PARTICIPATING",
+    "156": "NON_PARTICIPATING",
     "157": "FAIL",
     "158": "NON_PARTICIPATING",
     "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
-    "162": "FAIL",
-    "163": "NOT_ALIVE",
+    "162": "NON_PARTICIPATING",
+    "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
-    "165": "NON_PARTICIPATING",
+    "165": "FAIL",
     "166": "NON_PARTICIPATING",
-    "167": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "FAIL",
-    "17": "FAIL",
+    "169": "NON_PARTICIPATING",
+    "17": "NON_PARTICIPATING",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "FAIL",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NON_PARTICIPATING",
-    "178": "NON_PARTICIPATING",
-    "179": "NON_PARTICIPATING",
-    "18": "FAIL",
-    "180": "FAIL",
-    "181": "FAIL",
-    "182": "FAIL",
-    "183": "NON_PARTICIPATING",
-    "184": "NON_PARTICIPATING",
+    "178": "NOT_ALIVE",
+    "179": "FAIL",
+    "18": "NON_PARTICIPATING",
+    "180": "NON_PARTICIPATING",
+    "181": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
+    "183": "NOT_ALIVE",
+    "184": "FAIL",
     "185": "NOT_ALIVE",
-    "186": "FAIL",
-    "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
-    "189": "NOT_ALIVE",
-    "19": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
+    "188": "NON_PARTICIPATING",
+    "189": "NON_PARTICIPATING",
+    "19": "FAIL",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
     "193": "NON_PARTICIPATING",
-    "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
-    "196": "NON_PARTICIPATING",
+    "194": "FAIL",
+    "195": "NOT_ALIVE",
+    "196": "FAIL",
     "197": "NON_PARTICIPATING",
-    "198": "FAIL",
+    "198": "NOT_ALIVE",
     "199": "FAIL",
     "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
-    "200": "NON_PARTICIPATING",
-    "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "20": "NOT_ALIVE",
+    "200": "NOT_ALIVE",
+    "201": "NON_PARTICIPATING",
+    "202": "FAIL",
+    "203": "FAIL",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "FAIL",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
-    "209": "FAIL",
-    "21": "FAIL",
-    "210": "FAIL",
-    "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
+    "208": "FAIL",
+    "209": "NON_PARTICIPATING",
+    "21": "NOT_ALIVE",
+    "210": "NON_PARTICIPATING",
+    "211": "FAIL",
+    "212": "FAIL",
     "213": "NOT_ALIVE",
-    "214": "FAIL",
-    "215": "NON_PARTICIPATING",
-    "216": "FAIL",
-    "217": "NON_PARTICIPATING",
-    "218": "NOT_ALIVE",
-    "219": "NON_PARTICIPATING",
-    "22": "NON_PARTICIPATING",
-    "220": "NOT_ALIVE",
+    "214": "NOT_ALIVE",
+    "215": "FAIL",
+    "216": "NOT_ALIVE",
+    "217": "FAIL",
+    "218": "FAIL",
+    "219": "FAIL",
+    "22": "NOT_ALIVE",
+    "220": "FAIL",
     "221": "NON_PARTICIPATING",
-    "222": "NOT_ALIVE",
-    "223": "NON_PARTICIPATING",
+    "222": "NON_PARTICIPATING",
+    "223": "FAIL",
     "224": "NOT_ALIVE",
     "225": "FAIL",
     "226": "NON_PARTICIPATING",
@@ -149,114 +149,114 @@
     "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
-    "233": "FAIL",
-    "234": "FAIL",
+    "233": "NON_PARTICIPATING",
+    "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
     "236": "FAIL",
-    "237": "NOT_ALIVE",
-    "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
-    "24": "NON_PARTICIPATING",
+    "237": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
+    "239": "NOT_ALIVE",
+    "24": "FAIL",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
-    "243": "NON_PARTICIPATING",
+    "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "NOT_ALIVE",
-    "246": "FAIL",
-    "247": "FAIL",
-    "248": "NOT_ALIVE",
-    "249": "NON_PARTICIPATING",
+    "245": "NON_PARTICIPATING",
+    "246": "NON_PARTICIPATING",
+    "247": "NOT_ALIVE",
+    "248": "NON_PARTICIPATING",
+    "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
-    "251": "NON_PARTICIPATING",
-    "252": "NON_PARTICIPATING",
+    "250": "FAIL",
+    "251": "NOT_ALIVE",
+    "252": "FAIL",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "FAIL",
-    "27": "FAIL",
+    "26": "NOT_ALIVE",
+    "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
-    "29": "NON_PARTICIPATING",
+    "29": "NOT_ALIVE",
     "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
-    "31": "FAIL",
+    "30": "FAIL",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
-    "35": "NON_PARTICIPATING",
+    "34": "FAIL",
+    "35": "NOT_ALIVE",
     "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
     "38": "FAIL",
     "39": "NON_PARTICIPATING",
-    "4": "FAIL",
-    "40": "NOT_ALIVE",
+    "4": "NON_PARTICIPATING",
+    "40": "FAIL",
     "41": "NOT_ALIVE",
-    "42": "NON_PARTICIPATING",
-    "43": "NOT_ALIVE",
-    "44": "FAIL",
-    "45": "FAIL",
-    "46": "NON_PARTICIPATING",
-    "47": "FAIL",
+    "42": "NOT_ALIVE",
+    "43": "NON_PARTICIPATING",
+    "44": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
+    "46": "FAIL",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
-    "49": "NON_PARTICIPATING",
-    "5": "NON_PARTICIPATING",
-    "50": "NON_PARTICIPATING",
-    "51": "NOT_ALIVE",
-    "52": "FAIL",
-    "53": "NON_PARTICIPATING",
+    "49": "NOT_ALIVE",
+    "5": "FAIL",
+    "50": "FAIL",
+    "51": "NON_PARTICIPATING",
+    "52": "NON_PARTICIPATING",
+    "53": "FAIL",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
-    "56": "FAIL",
-    "57": "NON_PARTICIPATING",
+    "56": "NON_PARTICIPATING",
+    "57": "FAIL",
     "58": "NOT_ALIVE",
     "59": "NON_PARTICIPATING",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
-    "61": "FAIL",
+    "60": "FAIL",
+    "61": "NON_PARTICIPATING",
     "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
-    "66": "NON_PARTICIPATING",
+    "65": "NOT_ALIVE",
+    "66": "FAIL",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "FAIL",
-    "7": "NOT_ALIVE",
-    "70": "NON_PARTICIPATING",
+    "7": "FAIL",
+    "70": "FAIL",
     "71": "NOT_ALIVE",
     "72": "NON_PARTICIPATING",
-    "73": "FAIL",
-    "74": "FAIL",
-    "75": "FAIL",
+    "73": "NON_PARTICIPATING",
+    "74": "NOT_ALIVE",
+    "75": "NON_PARTICIPATING",
     "76": "FAIL",
     "77": "NOT_ALIVE",
-    "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "78": "NON_PARTICIPATING",
+    "79": "FAIL",
     "8": "NON_PARTICIPATING",
-    "80": "FAIL",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
     "84": "FAIL",
     "85": "NON_PARTICIPATING",
-    "86": "NOT_ALIVE",
-    "87": "FAIL",
-    "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
-    "9": "NOT_ALIVE",
+    "86": "FAIL",
+    "87": "NON_PARTICIPATING",
+    "88": "FAIL",
+    "89": "FAIL",
+    "9": "NON_PARTICIPATING",
     "90": "NON_PARTICIPATING",
-    "91": "FAIL",
-    "92": "NON_PARTICIPATING",
-    "93": "NOT_ALIVE",
-    "94": "FAIL",
+    "91": "NON_PARTICIPATING",
+    "92": "FAIL",
+    "93": "FAIL",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
-    "99": "NON_PARTICIPATING"
+    "99": "FAIL"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6456250,
-    6456282
+    6456383,
+    6456416
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 257,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6454064,
+  "last_allreduce_block": 6456333,
   "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 sha256:64dbd0adad299eb5d885b91c449ae853e502bdf619004228f00ec83cb8e1bd7d
+oid sha256:ff65667f0c28aa2a2b330ceba2e9b98fea8c1a943d857886eaefda173b4b7271
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2852ac09dd4c194d67344a9b57023e23fe9ba67ab796cb4f74a568584250329f
+oid sha256:97494b5232c63f35d643b6329ff4c3ac100c4adda28ad595d3ae87a0e6efdc45
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:69476a3ab54433b9264071e6e40480be8187335bd5e36d4e0d62ee04a34bbe1b
+oid sha256:b5b16e2734712c6f45ffec2050deab48970653a93b9596c7be70e564c755b5d6
 size 4400216536