Run 5. Outer Step 79. Inner Step 1.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "penguin102/c67-h21",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NON_PARTICIPATING",
@@ -9,87 +9,87 @@
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "NOT_ALIVE",
+    "107": "NON_PARTICIPATING",
     "108": "NON_PARTICIPATING",
     "109": "NON_PARTICIPATING",
-    "11": "NON_PARTICIPATING",
+    "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
+    "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NOT_ALIVE",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
-    "118": "NON_PARTICIPATING",
+    "118": "NOT_ALIVE",
     "119": "NOT_ALIVE",
     "12": "NON_PARTICIPATING",
-    "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "120": "SUCCESS",
+    "121": "SUCCESS",
     "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
     "125": "NOT_ALIVE",
     "126": "NON_PARTICIPATING",
-    "127": "NOT_ALIVE",
+    "127": "NON_PARTICIPATING",
     "128": "NOT_ALIVE",
-    "129": "NON_PARTICIPATING",
-    "13": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
+    "13": "SUCCESS",
     "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NON_PARTICIPATING",
     "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "135": "NOT_ALIVE",
     "136": "NON_PARTICIPATING",
     "137": "NOT_ALIVE",
-    "138": "SUCCESS",
+    "138": "NON_PARTICIPATING",
     "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
-    "143": "NOT_ALIVE",
+    "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
-    "147": "NON_PARTICIPATING",
+    "147": "NOT_ALIVE",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
-    "153": "SUCCESS",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NON_PARTICIPATING",
     "156": "NON_PARTICIPATING",
-    "157": "NON_PARTICIPATING",
+    "157": "NOT_ALIVE",
     "158": "NOT_ALIVE",
     "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
-    "160": "NON_PARTICIPATING",
+    "160": "SUCCESS",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
-    "163": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
-    "165": "SUCCESS",
-    "166": "NON_PARTICIPATING",
+    "165": "NON_PARTICIPATING",
+    "166": "SUCCESS",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
     "169": "NON_PARTICIPATING",
     "17": "NON_PARTICIPATING",
-    "170": "NOT_ALIVE",
+    "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
-    "172": "NOT_ALIVE",
-    "173": "SUCCESS",
+    "172": "SUCCESS",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
-    "178": "NON_PARTICIPATING",
+    "177": "NOT_ALIVE",
+    "178": "NOT_ALIVE",
     "179": "NON_PARTICIPATING",
     "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
@@ -102,47 +102,47 @@
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
-    "19": "NON_PARTICIPATING",
+    "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
+    "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
-    "197": "NON_PARTICIPATING",
-    "198": "NOT_ALIVE",
+    "197": "NOT_ALIVE",
+    "198": "FAIL",
     "199": "NOT_ALIVE",
-    "2": "NOT_ALIVE",
+    "2": "SUCCESS",
     "20": "NOT_ALIVE",
-    "200": "NOT_ALIVE",
+    "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
-    "204": "NOT_ALIVE",
+    "204": "NON_PARTICIPATING",
     "205": "NOT_ALIVE",
     "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
+    "208": "SUCCESS",
     "209": "NON_PARTICIPATING",
-    "21": "SUCCESS",
+    "21": "NON_PARTICIPATING",
     "210": "NON_PARTICIPATING",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
-    "213": "SUCCESS",
+    "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "SUCCESS",
+    "215": "NON_PARTICIPATING",
     "216": "NOT_ALIVE",
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
     "219": "NON_PARTICIPATING",
-    "22": "NON_PARTICIPATING",
+    "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
     "221": "NOT_ALIVE",
     "222": "NON_PARTICIPATING",
-    "223": "NOT_ALIVE",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
-    "225": "NOT_ALIVE",
+    "225": "NON_PARTICIPATING",
     "226": "NOT_ALIVE",
     "227": "NOT_ALIVE",
     "228": "NON_PARTICIPATING",
@@ -151,24 +151,24 @@
     "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NON_PARTICIPATING",
-    "234": "NOT_ALIVE",
+    "233": "FAIL",
+    "234": "SUCCESS",
     "235": "NOT_ALIVE",
-    "236": "NOT_ALIVE",
+    "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NOT_ALIVE",
     "239": "NON_PARTICIPATING",
     "24": "NON_PARTICIPATING",
     "240": "NON_PARTICIPATING",
     "241": "NON_PARTICIPATING",
-    "242": "NON_PARTICIPATING",
+    "242": "SUCCESS",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
     "245": "NOT_ALIVE",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "249": "SUCCESS",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
     "251": "NON_PARTICIPATING",
@@ -179,52 +179,52 @@
     "26": "NOT_ALIVE",
     "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
-    "29": "NOT_ALIVE",
+    "29": "SUCCESS",
     "3": "NON_PARTICIPATING",
-    "30": "SUCCESS",
-    "31": "NON_PARTICIPATING",
+    "30": "NON_PARTICIPATING",
+    "31": "SUCCESS",
     "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "NON_PARTICIPATING",
-    "35": "SUCCESS",
+    "34": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
     "36": "NON_PARTICIPATING",
     "37": "NON_PARTICIPATING",
-    "38": "SUCCESS",
-    "39": "NOT_ALIVE",
-    "4": "NOT_ALIVE",
-    "40": "SUCCESS",
-    "41": "NOT_ALIVE",
+    "38": "NON_PARTICIPATING",
+    "39": "NON_PARTICIPATING",
+    "4": "SUCCESS",
+    "40": "NON_PARTICIPATING",
+    "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
     "43": "NON_PARTICIPATING",
-    "44": "NON_PARTICIPATING",
+    "44": "NOT_ALIVE",
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
-    "47": "NON_PARTICIPATING",
+    "47": "SUCCESS",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
-    "5": "SUCCESS",
-    "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
+    "5": "NON_PARTICIPATING",
+    "50": "SUCCESS",
+    "51": "SUCCESS",
     "52": "NOT_ALIVE",
     "53": "NOT_ALIVE",
     "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
     "57": "NON_PARTICIPATING",
-    "58": "NON_PARTICIPATING",
-    "59": "SUCCESS",
-    "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
+    "58": "NOT_ALIVE",
+    "59": "NON_PARTICIPATING",
+    "6": "NON_PARTICIPATING",
+    "60": "NON_PARTICIPATING",
     "61": "NOT_ALIVE",
-    "62": "NOT_ALIVE",
-    "63": "NOT_ALIVE",
+    "62": "SUCCESS",
+    "63": "NON_PARTICIPATING",
     "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
-    "67": "NON_PARTICIPATING",
-    "68": "NOT_ALIVE",
-    "69": "NON_PARTICIPATING",
-    "7": "NON_PARTICIPATING",
+    "67": "NOT_ALIVE",
+    "68": "SUCCESS",
+    "69": "NOT_ALIVE",
+    "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
@@ -236,27 +236,27 @@
     "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "NON_PARTICIPATING",
-    "83": "NOT_ALIVE",
-    "84": "NON_PARTICIPATING",
+    "82": "FAIL",
+    "83": "SUCCESS",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
-    "86": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
     "87": "NON_PARTICIPATING",
-    "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
+    "89": "NOT_ALIVE",
     "9": "NON_PARTICIPATING",
-    "90": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
     "91": "NOT_ALIVE",
-    "92": "NON_PARTICIPATING",
+    "92": "NOT_ALIVE",
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
-    "95": "NOT_ALIVE",
+    "95": "SUCCESS",
     "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
-    "99": "NOT_ALIVE"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    5924918,
-    5924930
+    5924976,
+    5924988
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 187,
+  "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 5922405,
+  "last_allreduce_block": 5924869,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f60f46f2bc48c861b5573f5ba554bd46c61cb33e06f75fbb0d3aaae77c19bbb2
+oid sha256:87af64000a18785342a1f9b445e6a7ff8795275013a49df0484d51c55adb1ce1
 size 8800564561

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4144af9a3c12e6a0288977f5c6558ba85660cc16c5326b70d1d8928436c2fa0
+oid sha256:f1b148ea437223fbb57ce6289570982b06f66281f40eccc7911d2c718cdfcc1a
 size 4400216536