Run 4. Outer Step 8. Inner Step 0.

config.json

@@ -1,22 +1,22 @@
 {
-  "_name_or_path": "thieu86/SN3807",
+  "_name_or_path": "distributed/optimized-gpt2-1b",
   "activation_function": "gelu_new",
   "all_reduce_scores": {
     "0": "NON_PARTICIPATING",
-    "1": "NON_PARTICIPATING",
-    "10": "NON_PARTICIPATING",
+    "1": "SUCCESS",
+    "10": "FAIL",
     "100": "NON_PARTICIPATING",
-    "101": "NON_PARTICIPATING",
+    "101": "SUCCESS",
     "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
-    "107": "NON_PARTICIPATING",
+    "107": "SUCCESS",
     "108": "NON_PARTICIPATING",
-    "109": "SUCCESS",
-    "11": "NON_PARTICIPATING",
-    "110": "NON_PARTICIPATING",
+    "109": "NON_PARTICIPATING",
+    "11": "SUCCESS",
+    "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
@@ -24,13 +24,13 @@
     "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
     "117": "SUCCESS",
-    "118": "NON_PARTICIPATING",
+    "118": "SUCCESS",
     "119": "NON_PARTICIPATING",
-    "12": "NON_PARTICIPATING",
+    "12": "SUCCESS",
     "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
-    "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
+    "122": "SUCCESS",
+    "123": "SUCCESS",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
@@ -40,9 +40,9 @@
     "13": "NON_PARTICIPATING",
     "130": "NON_PARTICIPATING",
     "131": "NON_PARTICIPATING",
-    "132": "NON_PARTICIPATING",
+    "132": "SUCCESS",
     "133": "NON_PARTICIPATING",
-    "134": "NON_PARTICIPATING",
+    "134": "SUCCESS",
     "135": "NON_PARTICIPATING",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
@@ -51,78 +51,78 @@
     "14": "NON_PARTICIPATING",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
-    "142": "NON_PARTICIPATING",
+    "142": "SUCCESS",
     "143": "NON_PARTICIPATING",
     "144": "NON_PARTICIPATING",
-    "145": "NON_PARTICIPATING",
+    "145": "SUCCESS",
     "146": "NON_PARTICIPATING",
-    "147": "NON_PARTICIPATING",
-    "148": "NON_PARTICIPATING",
-    "149": "NON_PARTICIPATING",
-    "15": "NON_PARTICIPATING",
-    "150": "NON_PARTICIPATING",
+    "147": "SUCCESS",
+    "148": "SUCCESS",
+    "149": "SUCCESS",
+    "15": "FAIL",
+    "150": "SUCCESS",
     "151": "NON_PARTICIPATING",
     "152": "NON_PARTICIPATING",
-    "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
+    "153": "SUCCESS",
+    "154": "SUCCESS",
     "155": "NON_PARTICIPATING",
-    "156": "NON_PARTICIPATING",
+    "156": "FAIL",
     "157": "NON_PARTICIPATING",
     "158": "NON_PARTICIPATING",
-    "159": "NON_PARTICIPATING",
-    "16": "NON_PARTICIPATING",
-    "160": "NON_PARTICIPATING",
+    "159": "SUCCESS",
+    "16": "SUCCESS",
+    "160": "SUCCESS",
     "161": "NON_PARTICIPATING",
     "162": "NON_PARTICIPATING",
     "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
-    "167": "NON_PARTICIPATING",
+    "167": "FAIL",
     "168": "NON_PARTICIPATING",
     "169": "NON_PARTICIPATING",
-    "17": "NON_PARTICIPATING",
+    "17": "SUCCESS",
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
-    "172": "NON_PARTICIPATING",
+    "172": "SUCCESS",
     "173": "NON_PARTICIPATING",
     "174": "NON_PARTICIPATING",
-    "175": "NON_PARTICIPATING",
+    "175": "SUCCESS",
     "176": "NON_PARTICIPATING",
-    "177": "NON_PARTICIPATING",
-    "178": "NON_PARTICIPATING",
+    "177": "FAIL",
+    "178": "FAIL",
     "179": "NON_PARTICIPATING",
     "18": "NON_PARTICIPATING",
     "180": "NON_PARTICIPATING",
-    "181": "SUCCESS",
+    "181": "NON_PARTICIPATING",
     "182": "NON_PARTICIPATING",
     "183": "NON_PARTICIPATING",
-    "184": "SUCCESS",
+    "184": "NON_PARTICIPATING",
     "185": "NON_PARTICIPATING",
-    "186": "NON_PARTICIPATING",
-    "187": "NON_PARTICIPATING",
+    "186": "SUCCESS",
+    "187": "FAIL",
     "188": "NON_PARTICIPATING",
     "189": "NON_PARTICIPATING",
     "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
-    "191": "NON_PARTICIPATING",
+    "191": "FAIL",
     "192": "NON_PARTICIPATING",
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
+    "195": "SUCCESS",
     "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
     "198": "NON_PARTICIPATING",
     "199": "NON_PARTICIPATING",
     "2": "SUCCESS",
-    "20": "NON_PARTICIPATING",
+    "20": "SUCCESS",
     "200": "SUCCESS",
-    "201": "NON_PARTICIPATING",
-    "202": "NON_PARTICIPATING",
-    "203": "NON_PARTICIPATING",
+    "201": "SUCCESS",
+    "202": "SUCCESS",
+    "203": "SUCCESS",
     "204": "NON_PARTICIPATING",
     "205": "NON_PARTICIPATING",
-    "206": "NON_PARTICIPATING",
+    "206": "FAIL",
     "207": "NON_PARTICIPATING",
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
@@ -132,29 +132,29 @@
     "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
-    "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
+    "215": "SUCCESS",
+    "216": "SUCCESS",
     "217": "NON_PARTICIPATING",
-    "218": "NON_PARTICIPATING",
+    "218": "SUCCESS",
     "219": "NON_PARTICIPATING",
     "22": "SUCCESS",
-    "220": "NON_PARTICIPATING",
-    "221": "NON_PARTICIPATING",
-    "222": "NON_PARTICIPATING",
-    "223": "NON_PARTICIPATING",
+    "220": "FAIL",
+    "221": "FAIL",
+    "222": "SUCCESS",
+    "223": "SUCCESS",
     "224": "NON_PARTICIPATING",
-    "225": "NON_PARTICIPATING",
-    "226": "NON_PARTICIPATING",
+    "225": "SUCCESS",
+    "226": "SUCCESS",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
     "229": "NON_PARTICIPATING",
-    "23": "SUCCESS",
+    "23": "NON_PARTICIPATING",
     "230": "NON_PARTICIPATING",
     "231": "NON_PARTICIPATING",
     "232": "NON_PARTICIPATING",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
-    "235": "SUCCESS",
+    "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
@@ -167,58 +167,58 @@
     "244": "NON_PARTICIPATING",
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
+    "247": "SUCCESS",
     "248": "NON_PARTICIPATING",
     "249": "SUCCESS",
     "25": "SUCCESS",
-    "250": "NON_PARTICIPATING",
+    "250": "SUCCESS",
     "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
-    "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
+    "255": "SUCCESS",
+    "26": "FAIL",
+    "27": "SUCCESS",
     "28": "NON_PARTICIPATING",
     "29": "NON_PARTICIPATING",
     "3": "NON_PARTICIPATING",
-    "30": "NON_PARTICIPATING",
+    "30": "SUCCESS",
     "31": "NON_PARTICIPATING",
-    "32": "NON_PARTICIPATING",
+    "32": "FAIL",
     "33": "NON_PARTICIPATING",
-    "34": "NON_PARTICIPATING",
-    "35": "NON_PARTICIPATING",
+    "34": "SUCCESS",
+    "35": "SUCCESS",
     "36": "NON_PARTICIPATING",
-    "37": "NON_PARTICIPATING",
+    "37": "SUCCESS",
     "38": "NON_PARTICIPATING",
-    "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
-    "40": "NON_PARTICIPATING",
-    "41": "NON_PARTICIPATING",
+    "39": "FAIL",
+    "4": "FAIL",
+    "40": "FAIL",
+    "41": "SUCCESS",
     "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
-    "44": "NON_PARTICIPATING",
-    "45": "NON_PARTICIPATING",
-    "46": "NON_PARTICIPATING",
+    "43": "SUCCESS",
+    "44": "SUCCESS",
+    "45": "SUCCESS",
+    "46": "SUCCESS",
     "47": "NON_PARTICIPATING",
     "48": "NON_PARTICIPATING",
-    "49": "NON_PARTICIPATING",
+    "49": "SUCCESS",
     "5": "NON_PARTICIPATING",
-    "50": "NON_PARTICIPATING",
-    "51": "NON_PARTICIPATING",
+    "50": "SUCCESS",
+    "51": "SUCCESS",
     "52": "NON_PARTICIPATING",
-    "53": "NON_PARTICIPATING",
+    "53": "SUCCESS",
     "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
+    "55": "SUCCESS",
+    "56": "SUCCESS",
     "57": "NON_PARTICIPATING",
-    "58": "NON_PARTICIPATING",
+    "58": "SUCCESS",
     "59": "NON_PARTICIPATING",
-    "6": "NON_PARTICIPATING",
-    "60": "NON_PARTICIPATING",
+    "6": "SUCCESS",
+    "60": "SUCCESS",
     "61": "NON_PARTICIPATING",
     "62": "SUCCESS",
-    "63": "NON_PARTICIPATING",
+    "63": "SUCCESS",
     "64": "NON_PARTICIPATING",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
@@ -226,38 +226,38 @@
     "68": "NON_PARTICIPATING",
     "69": "NON_PARTICIPATING",
     "7": "NON_PARTICIPATING",
-    "70": "NON_PARTICIPATING",
+    "70": "SUCCESS",
     "71": "NON_PARTICIPATING",
     "72": "NON_PARTICIPATING",
     "73": "NON_PARTICIPATING",
     "74": "NON_PARTICIPATING",
-    "75": "NON_PARTICIPATING",
+    "75": "SUCCESS",
     "76": "NON_PARTICIPATING",
     "77": "NON_PARTICIPATING",
     "78": "SUCCESS",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
+    "80": "SUCCESS",
     "81": "NON_PARTICIPATING",
-    "82": "NON_PARTICIPATING",
+    "82": "SUCCESS",
     "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
+    "84": "SUCCESS",
+    "85": "SUCCESS",
     "86": "NON_PARTICIPATING",
-    "87": "NON_PARTICIPATING",
+    "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
+    "89": "FAIL",
     "9": "NON_PARTICIPATING",
-    "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
+    "90": "SUCCESS",
+    "91": "FAIL",
     "92": "NON_PARTICIPATING",
     "93": "NON_PARTICIPATING",
     "94": "NON_PARTICIPATING",
-    "95": "NON_PARTICIPATING",
-    "96": "NON_PARTICIPATING",
+    "95": "SUCCESS",
+    "96": "SUCCESS",
     "97": "NON_PARTICIPATING",
     "98": "NON_PARTICIPATING",
-    "99": "NON_PARTICIPATING"
+    "99": "SUCCESS"
   },
   "architectures": [
     "GPTOptim"
@@ -268,20 +268,20 @@
     "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
   },
   "block_list": [
-    5330923,
-    5330927,
-    5330930,
-    5330934,
-    5330938
+    5501234,
+    5501238,
+    5501242,
+    5501246,
+    5501249
   ],
   "block_size": 1024,
   "bos_token_id": 50256,
   "embd_pdrop": 0.1,
   "eos_token_id": 50256,
   "initializer_range": 0.02,
-  "inner_step": 53,
+  "inner_step": 0,
   "inner_steps": 0,
-  "last_allreduce_block": 5323221,
+  "last_allreduce_block": 5499548,
   "layer_norm_epsilon": 1e-05,
   "model_type": "gpt_optimized",
   "n_embd": 1280,

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5dcdfc41987c641e5900b57051f8c3e7b9fe076e05c18b555d0e5d3d7a407c66
+oid sha256:1a606695e0adec9477b43c25b1e1268bab930e40a7fa6c5c73bcb3688e34e5e8
 size 8081782026

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3484a6e54a3fa19e458d5220895f984f58bf20ca3bab36277d335a32cddbaf8
+oid sha256:3c1ad854fc14e509aa4a615119d509af3a3d9894ec41b0d8805f1d7f5c626682
 size 4040701744