Run 11. Outer Step 380. Inner Step 9.

config.json

@@ -1,63 +1,63 @@
 {
-  "_name_or_path": "trenden/d_m16",
+  "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NON_PARTICIPATING",
+    "10": "FAIL",
     "100": "NOT_ALIVE",
-    "101": "SUCCESS",
+    "101": "FAIL",
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
+    "104": "FAIL",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
-    "107": "SUCCESS",
-    "108": "SUCCESS",
-    "109": "SUCCESS",
+    "107": "FAIL",
+    "108": "FAIL",
+    "109": "FAIL",
     "11": "NOT_ALIVE",
-    "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
-    "113": "SUCCESS",
+    "110": "FAIL",
+    "111": "NOT_ALIVE",
+    "112": "FAIL",
+    "113": "FAIL",
     "114": "NOT_ALIVE",
-    "115": "NOT_ALIVE",
-    "116": "NON_PARTICIPATING",
+    "115": "FAIL",
+    "116": "FAIL",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
-    "119": "NON_PARTICIPATING",
+    "119": "FAIL",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NOT_ALIVE",
     "122": "NOT_ALIVE",
-    "123": "SUCCESS",
+    "123": "FAIL",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
-    "127": "SUCCESS",
-    "128": "NOT_ALIVE",
+    "127": "FAIL",
+    "128": "FAIL",
     "129": "NOT_ALIVE",
-    "13": "SUCCESS",
-    "130": "SUCCESS",
+    "13": "FAIL",
+    "130": "FAIL",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "134": "NOT_ALIVE",
+    "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "SUCCESS",
-    "141": "NON_PARTICIPATING",
+    "140": "NON_PARTICIPATING",
+    "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
     "143": "NON_PARTICIPATING",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "SUCCESS",
-    "148": "SUCCESS",
-    "149": "NON_PARTICIPATING",
+    "147": "FAIL",
+    "148": "NON_PARTICIPATING",
+    "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
@@ -70,35 +70,35 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "SUCCESS",
+    "160": "FAIL",
     "161": "NOT_ALIVE",
-    "162": "NON_PARTICIPATING",
+    "162": "FAIL",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
     "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
-    "169": "NOT_ALIVE",
+    "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NOT_ALIVE",
+    "173": "NON_PARTICIPATING",
     "174": "NOT_ALIVE",
-    "175": "NON_PARTICIPATING",
+    "175": "NOT_ALIVE",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NON_PARTICIPATING",
-    "179": "SUCCESS",
+    "178": "NOT_ALIVE",
+    "179": "FAIL",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "NON_PARTICIPATING",
     "183": "NOT_ALIVE",
-    "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
-    "186": "SUCCESS",
+    "184": "NOT_ALIVE",
+    "185": "NON_PARTICIPATING",
+    "186": "NOT_ALIVE",
     "187": "NON_PARTICIPATING",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -106,28 +106,28 @@
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "SUCCESS",
+    "193": "FAIL",
     "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
+    "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
-    "198": "NON_PARTICIPATING",
-    "199": "SUCCESS",
-    "2": "NON_PARTICIPATING",
-    "20": "SUCCESS",
-    "200": "NON_PARTICIPATING",
-    "201": "NON_PARTICIPATING",
+    "198": "FAIL",
+    "199": "FAIL",
+    "2": "NOT_ALIVE",
+    "20": "FAIL",
+    "200": "FAIL",
+    "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "FAIL",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "SUCCESS",
+    "211": "FAIL",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "NOT_ALIVE",
@@ -143,36 +143,36 @@
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "SUCCESS",
+    "226": "FAIL",
     "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "SUCCESS",
-    "231": "SUCCESS",
+    "230": "FAIL",
+    "231": "FAIL",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
-    "237": "SUCCESS",
-    "238": "SUCCESS",
+    "236": "FAIL",
+    "237": "FAIL",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
-    "24": "SUCCESS",
+    "24": "FAIL",
     "240": "NOT_ALIVE",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "SUCCESS",
+    "245": "NON_PARTICIPATING",
     "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "SUCCESS",
-    "252": "SUCCESS",
+    "251": "FAIL",
+    "252": "FAIL",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
@@ -182,13 +182,13 @@
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "SUCCESS",
+    "31": "FAIL",
     "32": "NOT_ALIVE",
-    "33": "NOT_ALIVE",
+    "33": "NON_PARTICIPATING",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
-    "37": "SUCCESS",
+    "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
@@ -207,56 +207,56 @@
     "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "NOT_ALIVE",
+    "54": "NON_PARTICIPATING",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
-    "57": "SUCCESS",
+    "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
-    "59": "NON_PARTICIPATING",
+    "59": "FAIL",
     "6": "NOT_ALIVE",
-    "60": "NOT_ALIVE",
-    "61": "SUCCESS",
+    "60": "NON_PARTICIPATING",
+    "61": "FAIL",
     "62": "NON_PARTICIPATING",
-    "63": "SUCCESS",
-    "64": "NON_PARTICIPATING",
-    "65": "SUCCESS",
+    "63": "NOT_ALIVE",
+    "64": "FAIL",
+    "65": "NOT_ALIVE",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
-    "72": "NON_PARTICIPATING",
+    "71": "FAIL",
+    "72": "FAIL",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "SUCCESS",
+    "75": "FAIL",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "NON_PARTICIPATING",
+    "79": "FAIL",
     "8": "NON_PARTICIPATING",
-    "80": "NOT_ALIVE",
+    "80": "FAIL",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
-    "84": "NON_PARTICIPATING",
+    "83": "NOT_ALIVE",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
-    "87": "SUCCESS",
-    "88": "SUCCESS",
+    "87": "NOT_ALIVE",
+    "88": "NOT_ALIVE",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "SUCCESS",
-    "92": "SUCCESS",
+    "91": "FAIL",
+    "92": "FAIL",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
-    "98": "NOT_ALIVE",
-    "99": "SUCCESS"
+    "98": "NON_PARTICIPATING",
+    "99": "FAIL"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6734113,
-    6734149
+    6734266,
+    6734294
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 36,
+  "inner_step": 8,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6728726,
+  "last_allreduce_block": 6734234,
   "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:d26c5c23f8d9bd8f733fc075f936ff980bb89a1db89c9c4684b2a2898153b4c6
+oid sha256:d210be8685ccb0920204111eeeceefe850b3069557b7c662f378c1c817e49c84
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f30baca67c86945fe5302356d6f205e72e5e982749b336e73e46db58083ce86
+oid sha256:ec7b6da68658a446c9e3dcef39802999f9f86f08e96f4661da3f9c1db5dddc3e
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7bdb8905c82ceb32c2a48f98301a70ec6dfa4bb62b25bc55411d871a3410cb43
+oid sha256:b1658f995ca827cfdd79b2c9a8f76f5e4531f94f975cea6893f58374f77b6954
 size 4400216536