Run 10. Outer Step 2. Inner Step 0. Peers 35.

config.json

@@ -1,153 +1,153 @@
 {
-  "_name_or_path": "jssky/sn38-132",
+  "_name_or_path": "sglangperf/dara3",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "SUCCESS",
+    "1": "NOT_ALIVE",
     "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "NON_PARTICIPATING",
+    "102": "FAIL",
     "103": "NON_PARTICIPATING",
-    "104": "SUCCESS",
+    "104": "NOT_ALIVE",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "NON_PARTICIPATING",
+    "109": "FAIL",
     "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
-    "111": "NON_PARTICIPATING",
-    "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
-    "114": "NON_PARTICIPATING",
+    "110": "NON_PARTICIPATING",
+    "111": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
+    "113": "FAIL",
+    "114": "NOT_ALIVE",
     "115": "SUCCESS",
     "116": "NON_PARTICIPATING",
-    "117": "NOT_ALIVE",
+    "117": "NON_PARTICIPATING",
     "118": "NON_PARTICIPATING",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
-    "120": "NOT_ALIVE",
+    "120": "FAIL",
     "121": "NOT_ALIVE",
-    "122": "SUCCESS",
-    "123": "SUCCESS",
-    "124": "SUCCESS",
+    "122": "FAIL",
+    "123": "NON_PARTICIPATING",
+    "124": "FAIL",
     "125": "NON_PARTICIPATING",
-    "126": "SUCCESS",
+    "126": "NON_PARTICIPATING",
     "127": "NON_PARTICIPATING",
-    "128": "SUCCESS",
-    "129": "NOT_ALIVE",
-    "13": "SUCCESS",
+    "128": "NON_PARTICIPATING",
+    "129": "NON_PARTICIPATING",
+    "13": "NOT_ALIVE",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NON_PARTICIPATING",
-    "133": "SUCCESS",
-    "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "133": "NON_PARTICIPATING",
+    "134": "FAIL",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NON_PARTICIPATING",
     "139": "NOT_ALIVE",
-    "14": "NOT_ALIVE",
+    "14": "NON_PARTICIPATING",
     "140": "NOT_ALIVE",
     "141": "NOT_ALIVE",
     "142": "SUCCESS",
     "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NOT_ALIVE",
-    "147": "NON_PARTICIPATING",
-    "148": "SUCCESS",
+    "146": "NON_PARTICIPATING",
+    "147": "FAIL",
+    "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
-    "15": "SUCCESS",
+    "15": "NON_PARTICIPATING",
     "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NON_PARTICIPATING",
-    "154": "NOT_ALIVE",
-    "155": "NON_PARTICIPATING",
+    "154": "NON_PARTICIPATING",
+    "155": "FAIL",
     "156": "NON_PARTICIPATING",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
-    "159": "SUCCESS",
+    "159": "NON_PARTICIPATING",
     "16": "NON_PARTICIPATING",
     "160": "NOT_ALIVE",
     "161": "NOT_ALIVE",
-    "162": "SUCCESS",
+    "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
-    "165": "NON_PARTICIPATING",
-    "166": "NOT_ALIVE",
-    "167": "SUCCESS",
+    "165": "NOT_ALIVE",
+    "166": "NON_PARTICIPATING",
+    "167": "NON_PARTICIPATING",
     "168": "NON_PARTICIPATING",
-    "169": "SUCCESS",
-    "17": "NON_PARTICIPATING",
-    "170": "NON_PARTICIPATING",
+    "169": "FAIL",
+    "17": "NOT_ALIVE",
+    "170": "SUCCESS",
     "171": "NOT_ALIVE",
-    "172": "NON_PARTICIPATING",
-    "173": "SUCCESS",
-    "174": "NON_PARTICIPATING",
-    "175": "NOT_ALIVE",
+    "172": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
+    "174": "FAIL",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NON_PARTICIPATING",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
-    "18": "NOT_ALIVE",
+    "179": "NOT_ALIVE",
+    "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
     "181": "NON_PARTICIPATING",
-    "182": "NON_PARTICIPATING",
-    "183": "NON_PARTICIPATING",
+    "182": "SUCCESS",
+    "183": "SUCCESS",
     "184": "NON_PARTICIPATING",
-    "185": "NON_PARTICIPATING",
-    "186": "SUCCESS",
-    "187": "NON_PARTICIPATING",
+    "185": "SUCCESS",
+    "186": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
-    "19": "NON_PARTICIPATING",
+    "19": "NOT_ALIVE",
     "190": "NON_PARTICIPATING",
     "191": "NON_PARTICIPATING",
     "192": "NON_PARTICIPATING",
-    "193": "NON_PARTICIPATING",
-    "194": "SUCCESS",
-    "195": "NON_PARTICIPATING",
+    "193": "FAIL",
+    "194": "NON_PARTICIPATING",
+    "195": "FAIL",
     "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
-    "198": "NON_PARTICIPATING",
-    "199": "NOT_ALIVE",
+    "198": "NOT_ALIVE",
+    "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
-    "200": "NON_PARTICIPATING",
-    "201": "NOT_ALIVE",
+    "20": "SUCCESS",
+    "200": "SUCCESS",
+    "201": "SUCCESS",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "SUCCESS",
-    "206": "NON_PARTICIPATING",
+    "205": "NON_PARTICIPATING",
+    "206": "SUCCESS",
     "207": "NOT_ALIVE",
-    "208": "NON_PARTICIPATING",
+    "208": "NOT_ALIVE",
     "209": "NON_PARTICIPATING",
     "21": "SUCCESS",
     "210": "NON_PARTICIPATING",
     "211": "NON_PARTICIPATING",
-    "212": "NOT_ALIVE",
+    "212": "NON_PARTICIPATING",
     "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
     "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
-    "217": "SUCCESS",
+    "217": "FAIL",
     "218": "NOT_ALIVE",
-    "219": "SUCCESS",
+    "219": "NON_PARTICIPATING",
     "22": "NOT_ALIVE",
-    "220": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
     "221": "SUCCESS",
-    "222": "SUCCESS",
-    "223": "NON_PARTICIPATING",
+    "222": "NON_PARTICIPATING",
+    "223": "FAIL",
     "224": "NOT_ALIVE",
     "225": "SUCCESS",
-    "226": "NON_PARTICIPATING",
+    "226": "SUCCESS",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
+    "229": "NON_PARTICIPATING",
+    "23": "NOT_ALIVE",
     "230": "NON_PARTICIPATING",
     "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
@@ -160,80 +160,80 @@
     "239": "NON_PARTICIPATING",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
-    "242": "NOT_ALIVE",
+    "241": "NON_PARTICIPATING",
+    "242": "NON_PARTICIPATING",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
-    "248": "NON_PARTICIPATING",
+    "247": "FAIL",
+    "248": "SUCCESS",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "FAIL",
     "251": "NOT_ALIVE",
-    "252": "NON_PARTICIPATING",
-    "253": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
+    "253": "SUCCESS",
     "254": "NON_PARTICIPATING",
-    "255": "SUCCESS",
-    "26": "NON_PARTICIPATING",
+    "255": "NON_PARTICIPATING",
+    "26": "NOT_ALIVE",
     "27": "NON_PARTICIPATING",
     "28": "NON_PARTICIPATING",
     "29": "SUCCESS",
     "3": "NON_PARTICIPATING",
     "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
-    "32": "NOT_ALIVE",
+    "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
-    "34": "SUCCESS",
-    "35": "NON_PARTICIPATING",
-    "36": "NOT_ALIVE",
+    "34": "NOT_ALIVE",
+    "35": "FAIL",
+    "36": "NON_PARTICIPATING",
     "37": "NOT_ALIVE",
-    "38": "SUCCESS",
+    "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
-    "40": "SUCCESS",
+    "4": "SUCCESS",
+    "40": "NOT_ALIVE",
     "41": "NOT_ALIVE",
-    "42": "SUCCESS",
+    "42": "FAIL",
     "43": "NOT_ALIVE",
-    "44": "SUCCESS",
+    "44": "FAIL",
     "45": "NOT_ALIVE",
-    "46": "NON_PARTICIPATING",
-    "47": "SUCCESS",
-    "48": "NOT_ALIVE",
+    "46": "NOT_ALIVE",
+    "47": "FAIL",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NON_PARTICIPATING",
     "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
-    "52": "NOT_ALIVE",
-    "53": "NOT_ALIVE",
+    "51": "SUCCESS",
+    "52": "NON_PARTICIPATING",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
-    "55": "NOT_ALIVE",
-    "56": "SUCCESS",
-    "57": "NON_PARTICIPATING",
-    "58": "NOT_ALIVE",
+    "55": "NON_PARTICIPATING",
+    "56": "NON_PARTICIPATING",
+    "57": "SUCCESS",
+    "58": "NON_PARTICIPATING",
     "59": "SUCCESS",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
-    "61": "NON_PARTICIPATING",
-    "62": "NON_PARTICIPATING",
+    "60": "NON_PARTICIPATING",
+    "61": "SUCCESS",
+    "62": "NOT_ALIVE",
     "63": "SUCCESS",
     "64": "SUCCESS",
     "65": "NOT_ALIVE",
-    "66": "NOT_ALIVE",
+    "66": "NON_PARTICIPATING",
     "67": "NON_PARTICIPATING",
     "68": "NOT_ALIVE",
     "69": "SUCCESS",
-    "7": "SUCCESS",
-    "70": "NOT_ALIVE",
+    "7": "NOT_ALIVE",
+    "70": "NON_PARTICIPATING",
     "71": "NOT_ALIVE",
-    "72": "SUCCESS",
-    "73": "NON_PARTICIPATING",
-    "74": "NON_PARTICIPATING",
-    "75": "NON_PARTICIPATING",
+    "72": "NON_PARTICIPATING",
+    "73": "FAIL",
+    "74": "SUCCESS",
+    "75": "FAIL",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
-    "78": "SUCCESS",
+    "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
     "80": "NOT_ALIVE",
@@ -241,21 +241,21 @@
     "82": "NOT_ALIVE",
     "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
-    "86": "SUCCESS",
-    "87": "NON_PARTICIPATING",
+    "85": "NOT_ALIVE",
+    "86": "NOT_ALIVE",
+    "87": "NOT_ALIVE",
     "88": "NON_PARTICIPATING",
-    "89": "SUCCESS",
+    "89": "FAIL",
     "9": "NON_PARTICIPATING",
-    "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
+    "90": "SUCCESS",
+    "91": "SUCCESS",
     "92": "SUCCESS",
     "93": "NOT_ALIVE",
-    "94": "SUCCESS",
+    "94": "FAIL",
     "95": "NOT_ALIVE",
-    "96": "NON_PARTICIPATING",
-    "97": "NON_PARTICIPATING",
-    "98": "NOT_ALIVE",
+    "96": "NOT_ALIVE",
+    "97": "NOT_ALIVE",
+    "98": "NON_PARTICIPATING",
     "99": "NON_PARTICIPATING"
   },
   "architectures": [
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6463065,
+  "last_allreduce_block": 6464382,
   "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 sha256:54224150c3a3fba1f93862882ff4245aa8c8871fcd00eac0e61fee21ef3d0e84
+oid sha256:c4d6c6d975439e58ed4857403c8d58888ad67061f36503592a0b7c4ac558af7c
 size 4924926842

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05704a6122f3bff0f169ab01fbd32de6bfe1eb24071a9cf536d6268cd8686867
+oid sha256:50be98aadfc253da2c90fe7dc71ef3792149ffcf4cb58f09c5af20f263cb31ea
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59c4e61992a58a817c5ac159dc6b0caedef9eb947256257cb7c3687b2363b036
+oid sha256:ab85bbcd749ce97ce103337b48b3cb266a555979cda40392bd9fc660a1ebb4b5
 size 4400252568