Run 11. Outer Step 54. Inner Step 9.

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "mujerry/bt_v2",
+  "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
@@ -8,7 +8,7 @@
     "101": "NOT_ALIVE",
     "102": "SUCCESS",
     "103": "NOT_ALIVE",
-    "104": "SUCCESS",
+    "104": "FAIL",
     "105": "SUCCESS",
     "106": "SUCCESS",
     "107": "NOT_ALIVE",
@@ -22,7 +22,7 @@
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
     "116": "NOT_ALIVE",
-    "117": "NOT_ALIVE",
+    "117": "SUCCESS",
     "118": "SUCCESS",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
@@ -31,18 +31,18 @@
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "SUCCESS",
-    "125": "NOT_ALIVE",
-    "126": "NOT_ALIVE",
+    "125": "NON_PARTICIPATING",
+    "126": "NON_PARTICIPATING",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
     "13": "NON_PARTICIPATING",
-    "130": "NON_PARTICIPATING",
+    "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
-    "135": "NON_PARTICIPATING",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -68,65 +68,65 @@
     "156": "NOT_ALIVE",
     "157": "NON_PARTICIPATING",
     "158": "SUCCESS",
-    "159": "SUCCESS",
-    "16": "SUCCESS",
+    "159": "FAIL",
+    "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
     "161": "NON_PARTICIPATING",
     "162": "NOT_ALIVE",
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
-    "166": "NOT_ALIVE",
+    "166": "NON_PARTICIPATING",
     "167": "NOT_ALIVE",
     "168": "SUCCESS",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "FAIL",
     "174": "NON_PARTICIPATING",
     "175": "SUCCESS",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
     "179": "NOT_ALIVE",
-    "18": "NON_PARTICIPATING",
+    "18": "SUCCESS",
     "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
     "184": "SUCCESS",
     "185": "SUCCESS",
-    "186": "NOT_ALIVE",
+    "186": "SUCCESS",
     "187": "NOT_ALIVE",
     "188": "NON_PARTICIPATING",
     "189": "NOT_ALIVE",
     "19": "NON_PARTICIPATING",
-    "190": "SUCCESS",
+    "190": "NON_PARTICIPATING",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
     "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
-    "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
+    "197": "SUCCESS",
+    "198": "NOT_ALIVE",
     "199": "NOT_ALIVE",
     "2": "NOT_ALIVE",
-    "20": "NOT_ALIVE",
+    "20": "NON_PARTICIPATING",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
-    "205": "NON_PARTICIPATING",
+    "205": "NOT_ALIVE",
     "206": "SUCCESS",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "NOT_ALIVE",
-    "21": "SUCCESS",
-    "210": "SUCCESS",
+    "21": "NON_PARTICIPATING",
+    "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
@@ -147,41 +147,41 @@
     "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NON_PARTICIPATING",
+    "23": "NOT_ALIVE",
     "230": "NOT_ALIVE",
-    "231": "NON_PARTICIPATING",
+    "231": "NOT_ALIVE",
     "232": "NON_PARTICIPATING",
     "233": "NOT_ALIVE",
-    "234": "NOT_ALIVE",
+    "234": "SUCCESS",
     "235": "NOT_ALIVE",
     "236": "NON_PARTICIPATING",
-    "237": "SUCCESS",
-    "238": "SUCCESS",
+    "237": "NOT_ALIVE",
+    "238": "NON_PARTICIPATING",
     "239": "NOT_ALIVE",
-    "24": "SUCCESS",
+    "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
     "241": "NOT_ALIVE",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
     "245": "NOT_ALIVE",
-    "246": "NOT_ALIVE",
+    "246": "NON_PARTICIPATING",
     "247": "NOT_ALIVE",
-    "248": "SUCCESS",
+    "248": "FAIL",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "SUCCESS",
     "251": "NOT_ALIVE",
-    "252": "NOT_ALIVE",
+    "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "FAIL",
     "27": "SUCCESS",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
-    "30": "SUCCESS",
+    "30": "NOT_ALIVE",
     "31": "SUCCESS",
     "32": "NOT_ALIVE",
     "33": "NOT_ALIVE",
@@ -189,11 +189,11 @@
     "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
-    "38": "NON_PARTICIPATING",
-    "39": "SUCCESS",
+    "38": "SUCCESS",
+    "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
     "40": "NOT_ALIVE",
-    "41": "NON_PARTICIPATING",
+    "41": "SUCCESS",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
@@ -201,62 +201,62 @@
     "46": "SUCCESS",
     "47": "NOT_ALIVE",
     "48": "SUCCESS",
-    "49": "NOT_ALIVE",
+    "49": "SUCCESS",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NON_PARTICIPATING",
-    "52": "SUCCESS",
-    "53": "NON_PARTICIPATING",
+    "52": "NON_PARTICIPATING",
+    "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
-    "58": "NON_PARTICIPATING",
+    "57": "FAIL",
+    "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "NOT_ALIVE",
+    "61": "NON_PARTICIPATING",
     "62": "SUCCESS",
     "63": "NON_PARTICIPATING",
-    "64": "NON_PARTICIPATING",
+    "64": "SUCCESS",
     "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
-    "69": "SUCCESS",
-    "7": "NON_PARTICIPATING",
+    "69": "NON_PARTICIPATING",
+    "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
     "71": "NOT_ALIVE",
     "72": "NOT_ALIVE",
     "73": "SUCCESS",
     "74": "NON_PARTICIPATING",
     "75": "SUCCESS",
-    "76": "NON_PARTICIPATING",
+    "76": "SUCCESS",
     "77": "NOT_ALIVE",
-    "78": "NON_PARTICIPATING",
-    "79": "SUCCESS",
-    "8": "NOT_ALIVE",
+    "78": "NOT_ALIVE",
+    "79": "NON_PARTICIPATING",
+    "8": "SUCCESS",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "FAIL",
     "84": "NON_PARTICIPATING",
-    "85": "SUCCESS",
-    "86": "NOT_ALIVE",
-    "87": "SUCCESS",
-    "88": "NON_PARTICIPATING",
+    "85": "NON_PARTICIPATING",
+    "86": "SUCCESS",
+    "87": "NOT_ALIVE",
+    "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
-    "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
+    "90": "SUCCESS",
+    "91": "NOT_ALIVE",
     "92": "SUCCESS",
-    "93": "NON_PARTICIPATING",
+    "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
-    "96": "SUCCESS",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "SUCCESS",
-    "99": "SUCCESS"
+    "99": "NOT_ALIVE"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6573092,
-    6573126
+    6573266,
+    6573295
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 54,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6570982,
+  "last_allreduce_block": 6573223,
   "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 sha256:24c896a96e29584e827875c5955884dc2471e2a718138bc3c7679e56af225135
+oid sha256:16bcc1cfd1b9c49e7b4e3c3374cdad4315b15b5e7f82ce76689292cfeb6c0a52
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:522d366421ddae5b79da71d6810d1e3cbb8eea478fcad72612a33b168a003d40
+oid sha256:df667d6546f4680106e658081d981a3c6f69b84b00f8d3069ae478203e5c4a7a
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0d6e7360bda6fae3152f9b1bb4ac5f2da09930ab981a52d86509c243b729854
+oid sha256:d5452707928f36cd4f81b0d38018d318527b2bc1b709be6d2817485649305239
 size 4400216536