Run 5. Outer Step 133. Inner Step 1.

config.json

@@ -1,90 +1,90 @@
 {
-  "_name_or_path": "ncaagcc/ter2",
+  "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NON_PARTICIPATING",
-    "102": "NOT_ALIVE",
+    "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
-    "105": "NON_PARTICIPATING",
-    "106": "NON_PARTICIPATING",
+    "104": "NOT_ALIVE",
+    "105": "NOT_ALIVE",
+    "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
-    "108": "SUCCESS",
+    "108": "NOT_ALIVE",
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
     "111": "NON_PARTICIPATING",
-    "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
+    "112": "SUCCESS",
+    "113": "SUCCESS",
     "114": "NON_PARTICIPATING",
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
-    "118": "NOT_ALIVE",
-    "119": "NOT_ALIVE",
-    "12": "NON_PARTICIPATING",
-    "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "118": "NON_PARTICIPATING",
+    "119": "NON_PARTICIPATING",
+    "12": "NOT_ALIVE",
+    "120": "NON_PARTICIPATING",
+    "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
     "123": "NOT_ALIVE",
-    "124": "SUCCESS",
+    "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
-    "126": "NOT_ALIVE",
+    "126": "SUCCESS",
     "127": "NOT_ALIVE",
-    "128": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NON_PARTICIPATING",
-    "134": "NOT_ALIVE",
+    "134": "NON_PARTICIPATING",
     "135": "NON_PARTICIPATING",
-    "136": "NON_PARTICIPATING",
+    "136": "SUCCESS",
     "137": "NON_PARTICIPATING",
     "138": "SUCCESS",
     "139": "NON_PARTICIPATING",
-    "14": "SUCCESS",
+    "14": "NOT_ALIVE",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
-    "142": "NOT_ALIVE",
+    "142": "NON_PARTICIPATING",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
-    "145": "NOT_ALIVE",
-    "146": "NON_PARTICIPATING",
+    "145": "NON_PARTICIPATING",
+    "146": "NOT_ALIVE",
     "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
     "149": "NON_PARTICIPATING",
     "15": "SUCCESS",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
-    "154": "NON_PARTICIPATING",
-    "155": "SUCCESS",
+    "153": "NON_PARTICIPATING",
+    "154": "NOT_ALIVE",
+    "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "NON_PARTICIPATING",
+    "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
-    "16": "NON_PARTICIPATING",
+    "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
-    "161": "NON_PARTICIPATING",
+    "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
     "163": "NOT_ALIVE",
-    "164": "SUCCESS",
+    "164": "NON_PARTICIPATING",
     "165": "SUCCESS",
-    "166": "NOT_ALIVE",
+    "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
     "168": "SUCCESS",
     "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NON_PARTICIPATING",
     "171": "NOT_ALIVE",
-    "172": "NOT_ALIVE",
-    "173": "SUCCESS",
+    "172": "NON_PARTICIPATING",
+    "173": "NON_PARTICIPATING",
     "174": "NON_PARTICIPATING",
     "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
@@ -98,55 +98,55 @@
     "183": "SUCCESS",
     "184": "NON_PARTICIPATING",
     "185": "NOT_ALIVE",
-    "186": "NON_PARTICIPATING",
-    "187": "NOT_ALIVE",
-    "188": "NOT_ALIVE",
+    "186": "NOT_ALIVE",
+    "187": "NON_PARTICIPATING",
+    "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
-    "191": "NOT_ALIVE",
+    "191": "NON_PARTICIPATING",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
-    "195": "NOT_ALIVE",
-    "196": "NON_PARTICIPATING",
-    "197": "SUCCESS",
-    "198": "NOT_ALIVE",
+    "195": "NON_PARTICIPATING",
+    "196": "NOT_ALIVE",
+    "197": "NON_PARTICIPATING",
+    "198": "NON_PARTICIPATING",
     "199": "NON_PARTICIPATING",
-    "2": "NOT_ALIVE",
+    "2": "NON_PARTICIPATING",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
-    "201": "NON_PARTICIPATING",
+    "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
-    "204": "SUCCESS",
-    "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "204": "NOT_ALIVE",
+    "205": "NON_PARTICIPATING",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NON_PARTICIPATING",
     "209": "NOT_ALIVE",
     "21": "NOT_ALIVE",
-    "210": "NOT_ALIVE",
+    "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
-    "212": "NON_PARTICIPATING",
-    "213": "NON_PARTICIPATING",
+    "212": "NOT_ALIVE",
+    "213": "SUCCESS",
     "214": "NON_PARTICIPATING",
-    "215": "SUCCESS",
-    "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
-    "218": "NON_PARTICIPATING",
+    "215": "NON_PARTICIPATING",
+    "216": "NOT_ALIVE",
+    "217": "NOT_ALIVE",
+    "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
-    "221": "NOT_ALIVE",
+    "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "NOT_ALIVE",
-    "227": "NON_PARTICIPATING",
+    "227": "NOT_ALIVE",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "NOT_ALIVE",
     "23": "NON_PARTICIPATING",
     "230": "NON_PARTICIPATING",
     "231": "NOT_ALIVE",
@@ -155,57 +155,57 @@
     "234": "NOT_ALIVE",
     "235": "NON_PARTICIPATING",
     "236": "NOT_ALIVE",
-    "237": "SUCCESS",
+    "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
-    "240": "NOT_ALIVE",
+    "240": "NON_PARTICIPATING",
     "241": "NON_PARTICIPATING",
-    "242": "NOT_ALIVE",
+    "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
     "244": "NOT_ALIVE",
-    "245": "NOT_ALIVE",
-    "246": "NON_PARTICIPATING",
+    "245": "NON_PARTICIPATING",
+    "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NON_PARTICIPATING",
-    "249": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
-    "250": "SUCCESS",
+    "250": "NON_PARTICIPATING",
     "251": "NON_PARTICIPATING",
-    "252": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "SUCCESS",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
-    "29": "NOT_ALIVE",
-    "3": "NOT_ALIVE",
+    "29": "NON_PARTICIPATING",
+    "3": "NON_PARTICIPATING",
     "30": "NOT_ALIVE",
-    "31": "NON_PARTICIPATING",
+    "31": "NOT_ALIVE",
     "32": "SUCCESS",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NON_PARTICIPATING",
-    "36": "NOT_ALIVE",
+    "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
-    "39": "NON_PARTICIPATING",
-    "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
+    "39": "SUCCESS",
+    "4": "NON_PARTICIPATING",
+    "40": "SUCCESS",
     "41": "NON_PARTICIPATING",
-    "42": "NON_PARTICIPATING",
+    "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
     "46": "NOT_ALIVE",
-    "47": "NOT_ALIVE",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NON_PARTICIPATING",
-    "5": "SUCCESS",
+    "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
-    "52": "NOT_ALIVE",
+    "52": "FAIL",
     "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
@@ -214,9 +214,9 @@
     "58": "NON_PARTICIPATING",
     "59": "NON_PARTICIPATING",
     "6": "NON_PARTICIPATING",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "NON_PARTICIPATING",
-    "62": "NOT_ALIVE",
+    "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
@@ -228,34 +228,34 @@
     "70": "NOT_ALIVE",
     "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
-    "73": "NOT_ALIVE",
-    "74": "NON_PARTICIPATING",
+    "73": "NON_PARTICIPATING",
+    "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
+    "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
     "78": "NOT_ALIVE",
-    "79": "NOT_ALIVE",
+    "79": "NON_PARTICIPATING",
     "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
-    "81": "NOT_ALIVE",
+    "81": "NON_PARTICIPATING",
     "82": "NON_PARTICIPATING",
-    "83": "NOT_ALIVE",
+    "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
-    "85": "NOT_ALIVE",
+    "85": "NON_PARTICIPATING",
     "86": "NON_PARTICIPATING",
     "87": "NON_PARTICIPATING",
-    "88": "NON_PARTICIPATING",
+    "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
-    "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
+    "9": "NON_PARTICIPATING",
+    "90": "NOT_ALIVE",
+    "91": "NOT_ALIVE",
     "92": "NON_PARTICIPATING",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NON_PARTICIPATING",
     "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
+    "98": "NOT_ALIVE",
     "99": "NOT_ALIVE"
   },
   "architectures": [
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    5989615,
-    5989648
+    6004034,
+    6004053
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 59,
+  "inner_step": 1,
   "intermediate_size": 5632,
-  "last_allreduce_block": 5986899,
+  "last_allreduce_block": 6003212,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

gradients.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09ed42a0786b77a4e5c9fd75cab6c577cd5ab482be98b5101930bbdc5d8dd233
+size 44585978

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:851ef13f446a58c2079fc45b52e3610808bb418605ff5d591c052d50aee7dbd7
+oid sha256:96d647cd4593af786821b0f1d32e7f4235e71885d5dc1408a6a73dc2e3fed4af
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5267631fbcb370770ca7e433f899ae99f190a2ac76e2b783f4056c76ca84c44e
+oid sha256:39c26ad232b2e598b76f042e1ed6a5011aa87ef6501504e8c7fd97ba62b4b933
 size 4400216536