Run 8. Outer Step 124. Inner Step 5.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "zeras141a/a6",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
@@ -13,16 +13,16 @@
     "106": "NOT_ALIVE",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "NON_PARTICIPATING",
+    "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
+    "111": "SUCCESS",
+    "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "SUCCESS",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "NON_PARTICIPATING",
+    "117": "SUCCESS",
     "118": "SUCCESS",
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
@@ -30,27 +30,27 @@
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "NON_PARTICIPATING",
-    "124": "NOT_ALIVE",
-    "125": "FAIL",
+    "124": "NON_PARTICIPATING",
+    "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
+    "128": "NOT_ALIVE",
     "129": "NON_PARTICIPATING",
-    "13": "SUCCESS",
-    "130": "FAIL",
-    "131": "NON_PARTICIPATING",
-    "132": "NOT_ALIVE",
+    "13": "NOT_ALIVE",
+    "130": "SUCCESS",
+    "131": "NOT_ALIVE",
+    "132": "NON_PARTICIPATING",
     "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
-    "137": "NON_PARTICIPATING",
+    "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
     "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
-    "141": "NOT_ALIVE",
-    "142": "NOT_ALIVE",
+    "141": "SUCCESS",
+    "142": "SUCCESS",
     "143": "SUCCESS",
     "144": "NON_PARTICIPATING",
     "145": "NOT_ALIVE",
@@ -59,57 +59,57 @@
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
     "15": "NOT_ALIVE",
-    "150": "NON_PARTICIPATING",
-    "151": "NON_PARTICIPATING",
-    "152": "NON_PARTICIPATING",
+    "150": "NOT_ALIVE",
+    "151": "NOT_ALIVE",
+    "152": "SUCCESS",
     "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "SUCCESS",
+    "158": "NON_PARTICIPATING",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
     "165": "NOT_ALIVE",
-    "166": "NON_PARTICIPATING",
+    "166": "SUCCESS",
     "167": "NON_PARTICIPATING",
-    "168": "SUCCESS",
+    "168": "NON_PARTICIPATING",
     "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "SUCCESS",
     "172": "NON_PARTICIPATING",
-    "173": "FAIL",
-    "174": "NOT_ALIVE",
+    "173": "SUCCESS",
+    "174": "NON_PARTICIPATING",
     "175": "SUCCESS",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "NON_PARTICIPATING",
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
-    "181": "NON_PARTICIPATING",
+    "181": "NOT_ALIVE",
     "182": "SUCCESS",
-    "183": "SUCCESS",
+    "183": "NOT_ALIVE",
     "184": "SUCCESS",
-    "185": "NON_PARTICIPATING",
+    "185": "SUCCESS",
     "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
-    "188": "NON_PARTICIPATING",
-    "189": "SUCCESS",
+    "188": "NOT_ALIVE",
+    "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NON_PARTICIPATING",
     "192": "SUCCESS",
-    "193": "NON_PARTICIPATING",
-    "194": "NON_PARTICIPATING",
-    "195": "SUCCESS",
-    "196": "FAIL",
+    "193": "NOT_ALIVE",
+    "194": "SUCCESS",
+    "195": "NOT_ALIVE",
+    "196": "SUCCESS",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
     "199": "SUCCESS",
@@ -117,119 +117,119 @@
     "20": "NOT_ALIVE",
     "200": "SUCCESS",
     "201": "SUCCESS",
-    "202": "NOT_ALIVE",
+    "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NOT_ALIVE",
     "205": "NON_PARTICIPATING",
     "206": "NOT_ALIVE",
     "207": "NON_PARTICIPATING",
-    "208": "NON_PARTICIPATING",
-    "209": "SUCCESS",
+    "208": "NOT_ALIVE",
+    "209": "NON_PARTICIPATING",
     "21": "SUCCESS",
-    "210": "SUCCESS",
-    "211": "FAIL",
+    "210": "NON_PARTICIPATING",
+    "211": "SUCCESS",
     "212": "NON_PARTICIPATING",
-    "213": "SUCCESS",
-    "214": "SUCCESS",
-    "215": "NOT_ALIVE",
-    "216": "SUCCESS",
+    "213": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
+    "215": "NON_PARTICIPATING",
+    "216": "NOT_ALIVE",
     "217": "SUCCESS",
     "218": "SUCCESS",
-    "219": "NOT_ALIVE",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "SUCCESS",
     "221": "NON_PARTICIPATING",
-    "222": "NOT_ALIVE",
-    "223": "SUCCESS",
+    "222": "NON_PARTICIPATING",
+    "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "SUCCESS",
-    "226": "NON_PARTICIPATING",
-    "227": "NON_PARTICIPATING",
+    "226": "SUCCESS",
+    "227": "NOT_ALIVE",
     "228": "SUCCESS",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NOT_ALIVE",
+    "230": "NON_PARTICIPATING",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
-    "234": "SUCCESS",
-    "235": "NON_PARTICIPATING",
+    "234": "NON_PARTICIPATING",
+    "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
-    "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
+    "237": "SUCCESS",
+    "238": "SUCCESS",
+    "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
-    "241": "NON_PARTICIPATING",
+    "241": "NOT_ALIVE",
     "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "245": "SUCCESS",
+    "246": "SUCCESS",
     "247": "SUCCESS",
     "248": "SUCCESS",
     "249": "SUCCESS",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "FAIL",
+    "251": "SUCCESS",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
     "254": "NOT_ALIVE",
-    "255": "NON_PARTICIPATING",
+    "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "SUCCESS",
+    "27": "NON_PARTICIPATING",
     "28": "NOT_ALIVE",
     "29": "NON_PARTICIPATING",
     "3": "NOT_ALIVE",
-    "30": "NON_PARTICIPATING",
-    "31": "NOT_ALIVE",
-    "32": "NOT_ALIVE",
+    "30": "SUCCESS",
+    "31": "NON_PARTICIPATING",
+    "32": "NON_PARTICIPATING",
     "33": "NON_PARTICIPATING",
-    "34": "NOT_ALIVE",
+    "34": "NON_PARTICIPATING",
     "35": "NON_PARTICIPATING",
-    "36": "SUCCESS",
+    "36": "FAIL",
     "37": "NON_PARTICIPATING",
-    "38": "NON_PARTICIPATING",
+    "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
-    "4": "SUCCESS",
-    "40": "SUCCESS",
+    "4": "NOT_ALIVE",
+    "40": "NOT_ALIVE",
     "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
     "43": "SUCCESS",
     "44": "NOT_ALIVE",
-    "45": "SUCCESS",
+    "45": "NOT_ALIVE",
     "46": "NON_PARTICIPATING",
-    "47": "NOT_ALIVE",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
-    "50": "NOT_ALIVE",
+    "50": "SUCCESS",
     "51": "NON_PARTICIPATING",
-    "52": "NON_PARTICIPATING",
-    "53": "NON_PARTICIPATING",
+    "52": "SUCCESS",
+    "53": "NOT_ALIVE",
     "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
+    "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
-    "6": "NOT_ALIVE",
+    "6": "SUCCESS",
     "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
+    "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
-    "63": "NOT_ALIVE",
+    "63": "NON_PARTICIPATING",
     "64": "NON_PARTICIPATING",
-    "65": "SUCCESS",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
-    "67": "NOT_ALIVE",
-    "68": "NON_PARTICIPATING",
-    "69": "SUCCESS",
+    "67": "NON_PARTICIPATING",
+    "68": "SUCCESS",
+    "69": "NON_PARTICIPATING",
     "7": "NOT_ALIVE",
     "70": "NON_PARTICIPATING",
-    "71": "FAIL",
-    "72": "SUCCESS",
-    "73": "NOT_ALIVE",
-    "74": "FAIL",
+    "71": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
+    "73": "NON_PARTICIPATING",
+    "74": "SUCCESS",
     "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
@@ -238,25 +238,25 @@
     "8": "NOT_ALIVE",
     "80": "NOT_ALIVE",
     "81": "NON_PARTICIPATING",
-    "82": "NON_PARTICIPATING",
+    "82": "SUCCESS",
     "83": "NOT_ALIVE",
     "84": "NON_PARTICIPATING",
-    "85": "NOT_ALIVE",
+    "85": "NON_PARTICIPATING",
     "86": "SUCCESS",
-    "87": "FAIL",
+    "87": "SUCCESS",
     "88": "SUCCESS",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
     "91": "SUCCESS",
-    "92": "NOT_ALIVE",
-    "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "92": "NON_PARTICIPATING",
+    "93": "NON_PARTICIPATING",
+    "94": "SUCCESS",
     "95": "NON_PARTICIPATING",
-    "96": "FAIL",
+    "96": "SUCCESS",
     "97": "NON_PARTICIPATING",
     "98": "NOT_ALIVE",
-    "99": "NON_PARTICIPATING"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6441939,
-    6441949
+    6442008,
+    6442022
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 113,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6441307,
+  "last_allreduce_block": 6441956,
   "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:b028875137d181c9389c360eea86d8fbe4c91e9edd9e9a948f14de20dd2159ba
+oid sha256:672f85df74f424a61523684edb8aeb48ae5c0d4883516aacc2ab7d930f917463
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27fa30570c86640f97c833f0d3b6b6da41eb73d613ce9791a578fe387f917b2b
+oid sha256:5feb00f82fb774d31620f7d551b5e09a40dfb36684e7252408ce3d5accb630cf
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64c7dadf0ac5d21794ef4fd518fc05b12219c3e1896a1b172389a3fb38cb139f
+oid sha256:2938d3ea76f5dd9c8b382d185cd2a447f99b56684285a735a00d45292087205f
 size 4400216536