Run 5. Outer Step 91. Inner Step 1.

config.json

@@ -1,77 +1,77 @@
 {
-  "_name_or_path": "ncaagcc/gr1",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
-    "102": "NON_PARTICIPATING",
-    "103": "NOT_ALIVE",
+    "102": "NOT_ALIVE",
+    "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
-    "106": "NOT_ALIVE",
+    "105": "NOT_ALIVE",
+    "106": "NON_PARTICIPATING",
     "107": "NOT_ALIVE",
     "108": "NON_PARTICIPATING",
     "109": "NOT_ALIVE",
     "11": "NON_PARTICIPATING",
     "110": "NOT_ALIVE",
     "111": "NON_PARTICIPATING",
-    "112": "NON_PARTICIPATING",
+    "112": "SUCCESS",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "NOT_ALIVE",
+    "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
-    "119": "NOT_ALIVE",
+    "119": "NON_PARTICIPATING",
     "12": "NON_PARTICIPATING",
-    "120": "NOT_ALIVE",
-    "121": "NON_PARTICIPATING",
+    "120": "NON_PARTICIPATING",
+    "121": "NOT_ALIVE",
     "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
-    "124": "NOT_ALIVE",
+    "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
     "126": "NOT_ALIVE",
     "127": "NON_PARTICIPATING",
     "128": "NON_PARTICIPATING",
-    "129": "NON_PARTICIPATING",
+    "129": "NOT_ALIVE",
     "13": "NON_PARTICIPATING",
-    "130": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
-    "132": "SUCCESS",
-    "133": "NON_PARTICIPATING",
+    "132": "NON_PARTICIPATING",
+    "133": "SUCCESS",
     "134": "NON_PARTICIPATING",
     "135": "NON_PARTICIPATING",
-    "136": "SUCCESS",
+    "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
-    "138": "NON_PARTICIPATING",
+    "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
-    "140": "NON_PARTICIPATING",
+    "140": "NOT_ALIVE",
     "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
+    "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
     "147": "NON_PARTICIPATING",
-    "148": "NON_PARTICIPATING",
+    "148": "SUCCESS",
     "149": "NON_PARTICIPATING",
-    "15": "NOT_ALIVE",
+    "15": "NON_PARTICIPATING",
     "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
-    "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
+    "153": "SUCCESS",
+    "154": "NOT_ALIVE",
     "155": "NON_PARTICIPATING",
-    "156": "NON_PARTICIPATING",
+    "156": "SUCCESS",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
     "159": "SUCCESS",
     "16": "NON_PARTICIPATING",
     "160": "NON_PARTICIPATING",
-    "161": "NOT_ALIVE",
+    "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
@@ -80,7 +80,7 @@
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
     "169": "NOT_ALIVE",
-    "17": "NON_PARTICIPATING",
+    "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
     "172": "NON_PARTICIPATING",
@@ -88,18 +88,18 @@
     "174": "NOT_ALIVE",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
+    "177": "NON_PARTICIPATING",
     "178": "NON_PARTICIPATING",
-    "179": "NON_PARTICIPATING",
-    "18": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
+    "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
-    "181": "NON_PARTICIPATING",
+    "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "SUCCESS",
     "184": "NON_PARTICIPATING",
     "185": "NON_PARTICIPATING",
     "186": "NON_PARTICIPATING",
-    "187": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
@@ -109,29 +109,29 @@
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
     "195": "NON_PARTICIPATING",
-    "196": "NOT_ALIVE",
+    "196": "NON_PARTICIPATING",
     "197": "NOT_ALIVE",
     "198": "NON_PARTICIPATING",
     "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
-    "201": "NON_PARTICIPATING",
+    "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "203": "NOT_ALIVE",
     "204": "NON_PARTICIPATING",
-    "205": "NOT_ALIVE",
+    "205": "NON_PARTICIPATING",
     "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
-    "209": "NOT_ALIVE",
-    "21": "NON_PARTICIPATING",
+    "209": "NON_PARTICIPATING",
+    "21": "NOT_ALIVE",
     "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "SUCCESS",
+    "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
@@ -142,66 +142,66 @@
     "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
-    "225": "NOT_ALIVE",
+    "225": "SUCCESS",
     "226": "NOT_ALIVE",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
     "23": "NON_PARTICIPATING",
-    "230": "NON_PARTICIPATING",
+    "230": "NOT_ALIVE",
     "231": "NON_PARTICIPATING",
     "232": "NOT_ALIVE",
-    "233": "NOT_ALIVE",
-    "234": "NON_PARTICIPATING",
+    "233": "NON_PARTICIPATING",
+    "234": "NOT_ALIVE",
     "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
-    "239": "NON_PARTICIPATING",
-    "24": "SUCCESS",
+    "238": "SUCCESS",
+    "239": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
     "240": "NON_PARTICIPATING",
-    "241": "NON_PARTICIPATING",
+    "241": "NOT_ALIVE",
     "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
-    "244": "NOT_ALIVE",
+    "244": "NON_PARTICIPATING",
     "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "246": "NOT_ALIVE",
     "247": "NOT_ALIVE",
-    "248": "NON_PARTICIPATING",
+    "248": "NOT_ALIVE",
     "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
-    "251": "NON_PARTICIPATING",
+    "251": "NOT_ALIVE",
     "252": "NON_PARTICIPATING",
-    "253": "NOT_ALIVE",
+    "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
-    "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
+    "255": "NOT_ALIVE",
+    "26": "NOT_ALIVE",
     "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
     "3": "NON_PARTICIPATING",
-    "30": "SUCCESS",
+    "30": "NON_PARTICIPATING",
     "31": "NON_PARTICIPATING",
-    "32": "NOT_ALIVE",
+    "32": "SUCCESS",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
     "35": "NON_PARTICIPATING",
     "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
     "38": "SUCCESS",
-    "39": "SUCCESS",
+    "39": "NON_PARTICIPATING",
     "4": "NON_PARTICIPATING",
-    "40": "NON_PARTICIPATING",
-    "41": "SUCCESS",
+    "40": "SUCCESS",
+    "41": "NON_PARTICIPATING",
     "42": "NON_PARTICIPATING",
     "43": "NON_PARTICIPATING",
     "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
+    "46": "NOT_ALIVE",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
-    "49": "NOT_ALIVE",
+    "49": "NON_PARTICIPATING",
     "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
@@ -209,45 +209,45 @@
     "53": "NON_PARTICIPATING",
     "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
-    "56": "SUCCESS",
-    "57": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
+    "57": "NON_PARTICIPATING",
     "58": "NON_PARTICIPATING",
-    "59": "NON_PARTICIPATING",
+    "59": "NOT_ALIVE",
     "6": "NON_PARTICIPATING",
-    "60": "NON_PARTICIPATING",
-    "61": "NOT_ALIVE",
+    "60": "NOT_ALIVE",
+    "61": "NON_PARTICIPATING",
     "62": "SUCCESS",
     "63": "NOT_ALIVE",
-    "64": "NON_PARTICIPATING",
+    "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
-    "7": "NON_PARTICIPATING",
+    "7": "SUCCESS",
     "70": "NON_PARTICIPATING",
     "71": "NON_PARTICIPATING",
-    "72": "NON_PARTICIPATING",
+    "72": "NOT_ALIVE",
     "73": "NON_PARTICIPATING",
-    "74": "NOT_ALIVE",
+    "74": "NON_PARTICIPATING",
     "75": "NON_PARTICIPATING",
     "76": "NON_PARTICIPATING",
-    "77": "NOT_ALIVE",
+    "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
     "80": "NON_PARTICIPATING",
     "81": "NON_PARTICIPATING",
-    "82": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
+    "84": "NOT_ALIVE",
     "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
-    "87": "NOT_ALIVE",
+    "87": "NON_PARTICIPATING",
     "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
-    "9": "NON_PARTICIPATING",
-    "90": "NOT_ALIVE",
+    "9": "NOT_ALIVE",
+    "90": "NON_PARTICIPATING",
     "91": "NON_PARTICIPATING",
     "92": "NON_PARTICIPATING",
     "93": "NON_PARTICIPATING",
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    5939941,
-    5939956
+    5940026,
+    5940036
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 32,
+  "inner_step": 1,
   "intermediate_size": 5632,
-  "last_allreduce_block": 5938691,
+  "last_allreduce_block": 5939891,
   "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 sha256:2ba637c0376d1679c6275f27a38e09482f3cb1b0beb73dc977d91618084218ef
+oid sha256:5947836099bce6b027f0522d430e3e96fcf392f34d49b78c2eefd148f0c07962
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f49312a748a148b47ba30a74d98828328e226b8e827d4a2603b6c9c45efc472
+oid sha256:6ba6ddc4da345d40f85517bdff4c20a1d9bd08af53e6fee22c38e06e280967bd
 size 4400216536