Run 8. Outer Step 107. Inner Step 5.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "mizzaay/mark",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
@@ -8,88 +8,88 @@
     "101": "NON_PARTICIPATING",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
-    "105": "NOT_ALIVE",
+    "104": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
-    "113": "SUCCESS",
+    "111": "SUCCESS",
+    "112": "NON_PARTICIPATING",
+    "113": "NOT_ALIVE",
     "114": "NON_PARTICIPATING",
-    "115": "NON_PARTICIPATING",
+    "115": "NOT_ALIVE",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
     "119": "NON_PARTICIPATING",
-    "12": "NOT_ALIVE",
+    "12": "SUCCESS",
     "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
-    "125": "NOT_ALIVE",
-    "126": "FAIL",
+    "125": "SUCCESS",
+    "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
     "128": "SUCCESS",
     "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NON_PARTICIPATING",
-    "132": "NON_PARTICIPATING",
+    "132": "NOT_ALIVE",
     "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "NOT_ALIVE",
-    "136": "NOT_ALIVE",
-    "137": "NON_PARTICIPATING",
+    "136": "NON_PARTICIPATING",
+    "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
     "139": "SUCCESS",
     "14": "SUCCESS",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "NON_PARTICIPATING",
-    "143": "SUCCESS",
-    "144": "NON_PARTICIPATING",
-    "145": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
+    "144": "SUCCESS",
+    "145": "NON_PARTICIPATING",
     "146": "NOT_ALIVE",
     "147": "SUCCESS",
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NON_PARTICIPATING",
     "152": "NON_PARTICIPATING",
     "153": "NON_PARTICIPATING",
     "154": "SUCCESS",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "NON_PARTICIPATING",
-    "158": "FAIL",
-    "159": "SUCCESS",
+    "157": "NOT_ALIVE",
+    "158": "SUCCESS",
+    "159": "NON_PARTICIPATING",
     "16": "SUCCESS",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "NOT_ALIVE",
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
     "165": "NOT_ALIVE",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
-    "170": "NOT_ALIVE",
-    "171": "NOT_ALIVE",
+    "170": "SUCCESS",
+    "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
     "173": "SUCCESS",
     "174": "NON_PARTICIPATING",
     "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "SUCCESS",
-    "178": "NON_PARTICIPATING",
+    "178": "NOT_ALIVE",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
@@ -97,72 +97,72 @@
     "182": "NOT_ALIVE",
     "183": "SUCCESS",
     "184": "SUCCESS",
-    "185": "SUCCESS",
+    "185": "NON_PARTICIPATING",
     "186": "NOT_ALIVE",
-    "187": "NON_PARTICIPATING",
+    "187": "SUCCESS",
     "188": "SUCCESS",
     "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
     "193": "NON_PARTICIPATING",
     "194": "SUCCESS",
-    "195": "NOT_ALIVE",
-    "196": "SUCCESS",
+    "195": "NON_PARTICIPATING",
+    "196": "NON_PARTICIPATING",
     "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
+    "198": "SUCCESS",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "SUCCESS",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "SUCCESS",
-    "204": "NOT_ALIVE",
+    "204": "NON_PARTICIPATING",
     "205": "NOT_ALIVE",
     "206": "SUCCESS",
-    "207": "NON_PARTICIPATING",
+    "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
-    "209": "NON_PARTICIPATING",
-    "21": "SUCCESS",
+    "209": "SUCCESS",
+    "21": "NOT_ALIVE",
     "210": "NON_PARTICIPATING",
     "211": "NOT_ALIVE",
-    "212": "NOT_ALIVE",
+    "212": "NON_PARTICIPATING",
     "213": "SUCCESS",
-    "214": "SUCCESS",
-    "215": "NON_PARTICIPATING",
+    "214": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
     "216": "SUCCESS",
     "217": "SUCCESS",
-    "218": "NON_PARTICIPATING",
+    "218": "SUCCESS",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "NON_PARTICIPATING",
+    "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
-    "223": "SUCCESS",
+    "223": "FAIL",
     "224": "NON_PARTICIPATING",
     "225": "SUCCESS",
     "226": "NOT_ALIVE",
     "227": "SUCCESS",
-    "228": "SUCCESS",
+    "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
     "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
     "234": "SUCCESS",
-    "235": "NON_PARTICIPATING",
+    "235": "SUCCESS",
     "236": "SUCCESS",
-    "237": "SUCCESS",
+    "237": "NOT_ALIVE",
     "238": "SUCCESS",
     "239": "SUCCESS",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
-    "243": "NOT_ALIVE",
+    "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
     "245": "SUCCESS",
     "246": "NON_PARTICIPATING",
@@ -176,16 +176,16 @@
     "253": "SUCCESS",
     "254": "NOT_ALIVE",
     "255": "SUCCESS",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "SUCCESS",
     "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
-    "3": "SUCCESS",
+    "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
     "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
-    "33": "NON_PARTICIPATING",
-    "34": "NON_PARTICIPATING",
+    "33": "NOT_ALIVE",
+    "34": "NOT_ALIVE",
     "35": "SUCCESS",
     "36": "NOT_ALIVE",
     "37": "NON_PARTICIPATING",
@@ -193,37 +193,37 @@
     "39": "NON_PARTICIPATING",
     "4": "NOT_ALIVE",
     "40": "NON_PARTICIPATING",
-    "41": "NON_PARTICIPATING",
+    "41": "SUCCESS",
     "42": "NOT_ALIVE",
     "43": "SUCCESS",
     "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
-    "46": "SUCCESS",
+    "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "SUCCESS",
     "52": "NON_PARTICIPATING",
-    "53": "NON_PARTICIPATING",
-    "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
+    "53": "NOT_ALIVE",
+    "54": "NOT_ALIVE",
+    "55": "NOT_ALIVE",
+    "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "SUCCESS",
+    "59": "NON_PARTICIPATING",
     "6": "NOT_ALIVE",
-    "60": "SUCCESS",
-    "61": "NOT_ALIVE",
-    "62": "NOT_ALIVE",
+    "60": "NON_PARTICIPATING",
+    "61": "NON_PARTICIPATING",
+    "62": "SUCCESS",
     "63": "NON_PARTICIPATING",
     "64": "NON_PARTICIPATING",
     "65": "NOT_ALIVE",
-    "66": "NOT_ALIVE",
+    "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
     "68": "NON_PARTICIPATING",
-    "69": "NON_PARTICIPATING",
+    "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
@@ -232,31 +232,31 @@
     "74": "NOT_ALIVE",
     "75": "NON_PARTICIPATING",
     "76": "NON_PARTICIPATING",
-    "77": "NON_PARTICIPATING",
+    "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
-    "81": "NON_PARTICIPATING",
+    "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
-    "86": "SUCCESS",
-    "87": "SUCCESS",
-    "88": "NON_PARTICIPATING",
+    "85": "NOT_ALIVE",
+    "86": "FAIL",
+    "87": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
+    "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
-    "93": "NON_PARTICIPATING",
+    "93": "NOT_ALIVE",
     "94": "SUCCESS",
-    "95": "NON_PARTICIPATING",
-    "96": "NON_PARTICIPATING",
+    "95": "NOT_ALIVE",
+    "96": "SUCCESS",
     "97": "NON_PARTICIPATING",
-    "98": "NOT_ALIVE",
-    "99": "NOT_ALIVE"
+    "98": "SUCCESS",
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6428232,
-    6428245
+    6428308,
+    6428324
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 76,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6427668,
+  "last_allreduce_block": 6428267,
   "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 sha256:1527bd51f672b52460fa8249e4a52afc94ef47a02111f14cc8860421aa1d84ae
+oid sha256:2e151342c4e663bb77c8b3396783bb5d403383eb923dc02f2bcc971f2b6394ea
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:271c6ae5d06d880cfdf0c28b006404cbfe9cdb49f9b68b909cbf11ab16a4a90f
+oid sha256:e4b4f8d86decf97a74e1adc2e20d9d91750dea0e18dd379474031a50c429dc87
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d92a943d7c8fc78673be91b9327b8bc881fbbf7c7e01abe06b13c1776e2b3349
+oid sha256:4e46e7855774dd8b1ae9400943702c276820328d69329ff66c23006622c8ee45
 size 4400216536