Run 6. Outer Step 30. Inner Step 5.

config.json

@@ -2,60 +2,60 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "SUCCESS",
+    "1": "NOT_ALIVE",
     "10": "NOT_ALIVE",
-    "100": "NOT_ALIVE",
+    "100": "NON_PARTICIPATING",
     "101": "NOT_ALIVE",
     "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NON_PARTICIPATING",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
-    "108": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
-    "111": "NON_PARTICIPATING",
-    "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
+    "111": "SUCCESS",
+    "112": "SUCCESS",
+    "113": "SUCCESS",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
     "116": "NOT_ALIVE",
-    "117": "NOT_ALIVE",
-    "118": "NOT_ALIVE",
-    "119": "NOT_ALIVE",
+    "117": "NON_PARTICIPATING",
+    "118": "NON_PARTICIPATING",
+    "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
-    "120": "NON_PARTICIPATING",
-    "121": "NOT_ALIVE",
+    "120": "NOT_ALIVE",
+    "121": "SUCCESS",
     "122": "NON_PARTICIPATING",
-    "123": "SUCCESS",
-    "124": "NON_PARTICIPATING",
+    "123": "NON_PARTICIPATING",
+    "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
-    "126": "NOT_ALIVE",
+    "126": "SUCCESS",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "SUCCESS",
-    "134": "NON_PARTICIPATING",
-    "135": "NOT_ALIVE",
-    "136": "NON_PARTICIPATING",
+    "133": "NOT_ALIVE",
+    "134": "NOT_ALIVE",
+    "135": "NON_PARTICIPATING",
+    "136": "NOT_ALIVE",
     "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
-    "139": "NON_PARTICIPATING",
+    "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
-    "140": "SUCCESS",
+    "140": "FAIL",
     "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
-    "145": "SUCCESS",
+    "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
-    "147": "NOT_ALIVE",
+    "147": "SUCCESS",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "SUCCESS",
@@ -63,8 +63,8 @@
     "151": "NON_PARTICIPATING",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
-    "154": "NON_PARTICIPATING",
-    "155": "NOT_ALIVE",
+    "154": "NOT_ALIVE",
+    "155": "SUCCESS",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
     "158": "NON_PARTICIPATING",
@@ -72,41 +72,41 @@
     "16": "NOT_ALIVE",
     "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
-    "162": "NOT_ALIVE",
+    "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
-    "167": "NOT_ALIVE",
-    "168": "NON_PARTICIPATING",
+    "167": "NON_PARTICIPATING",
+    "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
-    "170": "NON_PARTICIPATING",
+    "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
-    "173": "NON_PARTICIPATING",
-    "174": "NOT_ALIVE",
-    "175": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
+    "174": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
-    "18": "SUCCESS",
+    "179": "NOT_ALIVE",
+    "18": "NOT_ALIVE",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
-    "182": "NON_PARTICIPATING",
+    "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
-    "184": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
     "185": "NOT_ALIVE",
-    "186": "NOT_ALIVE",
-    "187": "SUCCESS",
-    "188": "NOT_ALIVE",
-    "189": "SUCCESS",
+    "186": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
+    "188": "SUCCESS",
+    "189": "NON_PARTICIPATING",
     "19": "NOT_ALIVE",
-    "190": "SUCCESS",
-    "191": "NON_PARTICIPATING",
+    "190": "NOT_ALIVE",
+    "191": "NOT_ALIVE",
     "192": "NON_PARTICIPATING",
-    "193": "SUCCESS",
+    "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
@@ -115,98 +115,98 @@
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
-    "200": "NOT_ALIVE",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
-    "202": "NON_PARTICIPATING",
+    "202": "SUCCESS",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "NOT_ALIVE",
+    "205": "NON_PARTICIPATING",
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NOT_ALIVE",
     "21": "SUCCESS",
     "210": "NOT_ALIVE",
-    "211": "NOT_ALIVE",
+    "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
-    "214": "NON_PARTICIPATING",
+    "214": "NOT_ALIVE",
     "215": "NON_PARTICIPATING",
     "216": "SUCCESS",
-    "217": "SUCCESS",
+    "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
-    "219": "SUCCESS",
+    "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
     "221": "NOT_ALIVE",
-    "222": "NOT_ALIVE",
+    "222": "NON_PARTICIPATING",
     "223": "NOT_ALIVE",
     "224": "NON_PARTICIPATING",
     "225": "NON_PARTICIPATING",
     "226": "NOT_ALIVE",
-    "227": "NOT_ALIVE",
+    "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "SUCCESS",
+    "23": "NOT_ALIVE",
     "230": "SUCCESS",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NOT_ALIVE",
     "234": "NON_PARTICIPATING",
-    "235": "NOT_ALIVE",
+    "235": "NON_PARTICIPATING",
     "236": "NOT_ALIVE",
-    "237": "NOT_ALIVE",
-    "238": "NON_PARTICIPATING",
+    "237": "NON_PARTICIPATING",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "SUCCESS",
     "243": "NOT_ALIVE",
-    "244": "NON_PARTICIPATING",
-    "245": "NOT_ALIVE",
+    "244": "SUCCESS",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "NOT_ALIVE",
-    "252": "NON_PARTICIPATING",
+    "251": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "NON_PARTICIPATING",
-    "254": "NOT_ALIVE",
+    "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
+    "26": "NOT_ALIVE",
     "27": "NON_PARTICIPATING",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "SUCCESS",
-    "32": "SUCCESS",
+    "31": "NOT_ALIVE",
+    "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
     "34": "NON_PARTICIPATING",
-    "35": "NOT_ALIVE",
-    "36": "SUCCESS",
-    "37": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
+    "36": "NON_PARTICIPATING",
+    "37": "NON_PARTICIPATING",
     "38": "NON_PARTICIPATING",
-    "39": "SUCCESS",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "NON_PARTICIPATING",
-    "41": "NOT_ALIVE",
+    "40": "NOT_ALIVE",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NON_PARTICIPATING",
-    "44": "NOT_ALIVE",
+    "44": "NON_PARTICIPATING",
     "45": "NOT_ALIVE",
-    "46": "SUCCESS",
+    "46": "NON_PARTICIPATING",
     "47": "NOT_ALIVE",
-    "48": "NON_PARTICIPATING",
+    "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "SUCCESS",
     "51": "NOT_ALIVE",
-    "52": "NOT_ALIVE",
-    "53": "NON_PARTICIPATING",
+    "52": "NON_PARTICIPATING",
+    "53": "NOT_ALIVE",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NON_PARTICIPATING",
@@ -217,29 +217,29 @@
     "60": "NON_PARTICIPATING",
     "61": "NON_PARTICIPATING",
     "62": "NOT_ALIVE",
-    "63": "NON_PARTICIPATING",
-    "64": "NON_PARTICIPATING",
-    "65": "NON_PARTICIPATING",
+    "63": "NOT_ALIVE",
+    "64": "NOT_ALIVE",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
-    "70": "NON_PARTICIPATING",
+    "70": "NOT_ALIVE",
     "71": "SUCCESS",
-    "72": "NON_PARTICIPATING",
+    "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NON_PARTICIPATING",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "SUCCESS",
+    "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "FAIL",
+    "83": "NON_PARTICIPATING",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
@@ -250,10 +250,10 @@
     "90": "NOT_ALIVE",
     "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
-    "93": "NON_PARTICIPATING",
-    "94": "NOT_ALIVE",
+    "93": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NON_PARTICIPATING",
-    "96": "NON_PARTICIPATING",
+    "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
     "98": "NON_PARTICIPATING",
     "99": "SUCCESS"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6148138,
-    6148179
+    6148345,
+    6148390
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 107,
+  "inner_step": 5,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6146001,
+  "last_allreduce_block": 6148186,
   "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 sha256:6531944ce12a14437e62d512ffa19abdedf1e9fd6b66e7e4f7e9a940704d78e3
+oid sha256:4a21626591f61a9d59b21c4969d33d41f4886627bbc4aeb86b59a5a0dd85a76e
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4957ef910241b58d6d21e13616a698d7465357706ff729cd8358120f8c98a312
+oid sha256:7a85e0db7c0eb21c1f692826c00ea7c3d81f7bf0d9d25c092c684d67a376867a
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb12350f67f6acc58b0b02b290435aac21a916e9c4b551729030449f852a54a6
+oid sha256:588bac008e4a4713cb2244a42d36280d1209ce9aa71394ff5664416fc189333f
 size 4400216536