Run 10. Outer Step 3. Inner Step 0. Peers 13.

config.json

@@ -1,48 +1,48 @@
 {
-  "_name_or_path": "sglangperf/dara3",
+  "_name_or_path": "STR-NaN/llama-hot-8",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
     "102": "FAIL",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NON_PARTICIPATING",
-    "106": "NON_PARTICIPATING",
+    "105": "FAIL",
+    "106": "FAIL",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
     "109": "FAIL",
-    "11": "NOT_ALIVE",
+    "11": "NON_PARTICIPATING",
     "110": "NON_PARTICIPATING",
-    "111": "NOT_ALIVE",
+    "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
-    "113": "FAIL",
-    "114": "NOT_ALIVE",
-    "115": "SUCCESS",
+    "113": "NOT_ALIVE",
+    "114": "NON_PARTICIPATING",
+    "115": "FAIL",
     "116": "NON_PARTICIPATING",
     "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "FAIL",
+    "120": "SUCCESS",
     "121": "NOT_ALIVE",
     "122": "FAIL",
     "123": "NON_PARTICIPATING",
     "124": "FAIL",
     "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
-    "127": "NON_PARTICIPATING",
-    "128": "NON_PARTICIPATING",
+    "127": "NOT_ALIVE",
+    "128": "FAIL",
     "129": "NON_PARTICIPATING",
-    "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "13": "NON_PARTICIPATING",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NON_PARTICIPATING",
     "133": "NON_PARTICIPATING",
-    "134": "FAIL",
-    "135": "SUCCESS",
+    "134": "SUCCESS",
+    "135": "FAIL",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
@@ -50,82 +50,82 @@
     "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
-    "142": "SUCCESS",
-    "143": "NOT_ALIVE",
+    "142": "FAIL",
+    "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NON_PARTICIPATING",
-    "147": "FAIL",
+    "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
-    "150": "SUCCESS",
+    "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
-    "155": "FAIL",
+    "154": "FAIL",
+    "155": "NOT_ALIVE",
     "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
-    "158": "NON_PARTICIPATING",
+    "158": "NOT_ALIVE",
     "159": "NON_PARTICIPATING",
-    "16": "NON_PARTICIPATING",
+    "16": "SUCCESS",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
-    "165": "NOT_ALIVE",
+    "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "FAIL",
-    "17": "NOT_ALIVE",
-    "170": "SUCCESS",
+    "169": "NON_PARTICIPATING",
+    "17": "NON_PARTICIPATING",
+    "170": "FAIL",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NON_PARTICIPATING",
     "174": "FAIL",
-    "175": "NON_PARTICIPATING",
+    "175": "FAIL",
     "176": "NOT_ALIVE",
-    "177": "NON_PARTICIPATING",
+    "177": "FAIL",
     "178": "NOT_ALIVE",
-    "179": "NOT_ALIVE",
+    "179": "FAIL",
     "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
     "181": "NON_PARTICIPATING",
-    "182": "SUCCESS",
-    "183": "SUCCESS",
+    "182": "NON_PARTICIPATING",
+    "183": "NON_PARTICIPATING",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
+    "185": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
-    "19": "NOT_ALIVE",
-    "190": "NON_PARTICIPATING",
+    "19": "NON_PARTICIPATING",
+    "190": "SUCCESS",
     "191": "NON_PARTICIPATING",
     "192": "NON_PARTICIPATING",
     "193": "FAIL",
-    "194": "NON_PARTICIPATING",
-    "195": "FAIL",
+    "194": "FAIL",
+    "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
-    "198": "NOT_ALIVE",
-    "199": "NON_PARTICIPATING",
+    "198": "NON_PARTICIPATING",
+    "199": "FAIL",
     "2": "NON_PARTICIPATING",
-    "20": "SUCCESS",
-    "200": "SUCCESS",
+    "20": "NOT_ALIVE",
+    "200": "NON_PARTICIPATING",
     "201": "SUCCESS",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "NON_PARTICIPATING",
-    "206": "SUCCESS",
+    "205": "FAIL",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
+    "208": "SUCCESS",
     "209": "NON_PARTICIPATING",
-    "21": "SUCCESS",
+    "21": "FAIL",
     "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
@@ -133,128 +133,128 @@
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
-    "217": "FAIL",
+    "217": "NOT_ALIVE",
     "218": "NOT_ALIVE",
     "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
     "220": "NON_PARTICIPATING",
-    "221": "SUCCESS",
+    "221": "FAIL",
     "222": "NON_PARTICIPATING",
-    "223": "FAIL",
+    "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
-    "225": "SUCCESS",
-    "226": "SUCCESS",
+    "225": "FAIL",
+    "226": "NOT_ALIVE",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
-    "23": "NOT_ALIVE",
+    "229": "NOT_ALIVE",
+    "23": "NON_PARTICIPATING",
     "230": "NON_PARTICIPATING",
     "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
-    "233": "NON_PARTICIPATING",
-    "234": "NON_PARTICIPATING",
+    "233": "FAIL",
+    "234": "FAIL",
     "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
     "239": "NON_PARTICIPATING",
-    "24": "SUCCESS",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
-    "247": "FAIL",
-    "248": "SUCCESS",
+    "245": "FAIL",
+    "246": "FAIL",
+    "247": "NON_PARTICIPATING",
+    "248": "NON_PARTICIPATING",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "FAIL",
     "251": "NOT_ALIVE",
-    "252": "NOT_ALIVE",
-    "253": "SUCCESS",
+    "252": "FAIL",
+    "253": "FAIL",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "NON_PARTICIPATING",
-    "28": "NON_PARTICIPATING",
-    "29": "SUCCESS",
-    "3": "NON_PARTICIPATING",
+    "28": "FAIL",
+    "29": "FAIL",
+    "3": "FAIL",
     "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
-    "32": "NON_PARTICIPATING",
+    "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
-    "34": "NOT_ALIVE",
+    "34": "NON_PARTICIPATING",
     "35": "FAIL",
     "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
-    "4": "SUCCESS",
-    "40": "NOT_ALIVE",
+    "4": "NON_PARTICIPATING",
+    "40": "NON_PARTICIPATING",
     "41": "NOT_ALIVE",
-    "42": "FAIL",
+    "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "FAIL",
     "45": "NOT_ALIVE",
-    "46": "NOT_ALIVE",
-    "47": "FAIL",
+    "46": "FAIL",
+    "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
-    "51": "SUCCESS",
+    "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
     "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
-    "57": "SUCCESS",
-    "58": "NON_PARTICIPATING",
-    "59": "SUCCESS",
+    "56": "NOT_ALIVE",
+    "57": "FAIL",
+    "58": "NOT_ALIVE",
+    "59": "FAIL",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
+    "61": "FAIL",
     "62": "NOT_ALIVE",
-    "63": "SUCCESS",
+    "63": "NON_PARTICIPATING",
     "64": "SUCCESS",
     "65": "NOT_ALIVE",
-    "66": "NON_PARTICIPATING",
+    "66": "NOT_ALIVE",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "SUCCESS",
-    "7": "NOT_ALIVE",
+    "7": "NON_PARTICIPATING",
     "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
     "72": "NON_PARTICIPATING",
-    "73": "FAIL",
+    "73": "NON_PARTICIPATING",
     "74": "SUCCESS",
-    "75": "FAIL",
+    "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NON_PARTICIPATING",
-    "79": "NON_PARTICIPATING",
+    "79": "SUCCESS",
     "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
-    "85": "NOT_ALIVE",
-    "86": "NOT_ALIVE",
-    "87": "NOT_ALIVE",
+    "84": "NOT_ALIVE",
+    "85": "NON_PARTICIPATING",
+    "86": "FAIL",
+    "87": "NON_PARTICIPATING",
     "88": "NON_PARTICIPATING",
     "89": "FAIL",
     "9": "NON_PARTICIPATING",
-    "90": "SUCCESS",
-    "91": "SUCCESS",
-    "92": "SUCCESS",
+    "90": "FAIL",
+    "91": "NON_PARTICIPATING",
+    "92": "FAIL",
     "93": "NOT_ALIVE",
     "94": "FAIL",
     "95": "NOT_ALIVE",
-    "96": "NOT_ALIVE",
-    "97": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
+    "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6464382,
+  "last_allreduce_block": 6464980,
   "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:c4d6c6d975439e58ed4857403c8d58888ad67061f36503592a0b7c4ac558af7c
-size 4924926842
+oid sha256:92539ed27c047485fb15337eee88cc9d44c4c6f00433321dad6f7bb02e0bec70
+size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50be98aadfc253da2c90fe7dc71ef3792149ffcf4cb58f09c5af20f263cb31ea
+oid sha256:1c0b0098f3a2b707843d9e25ca8fc23d033e491e32f9837f68d1ab3669f4f8f7
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab85bbcd749ce97ce103337b48b3cb266a555979cda40392bd9fc660a1ebb4b5
+oid sha256:88280e48d278d6b8788d2fcf91f02c8211359ec6bbffcead3a2ca5d74354064a
 size 4400252568