Run 4. Outer Step 32. Inner Step 0.

config.json

@@ -1,23 +1,23 @@
 {
-  "_name_or_path": "Mhair/onethreesix",
+  "_name_or_path": "distributed/optimized-gpt2-1b",
   "activation_function": "gelu_new",
   "all_reduce_scores": {
     "0": "NON_PARTICIPATING",
-    "1": "NON_PARTICIPATING",
+    "1": "SUCCESS",
     "10": "NON_PARTICIPATING",
     "100": "NON_PARTICIPATING",
-    "101": "NON_PARTICIPATING",
+    "101": "SUCCESS",
     "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
-    "106": "NON_PARTICIPATING",
+    "106": "SUCCESS",
     "107": "NON_PARTICIPATING",
-    "108": "SUCCESS",
-    "109": "NON_PARTICIPATING",
+    "108": "NON_PARTICIPATING",
+    "109": "SUCCESS",
     "11": "NON_PARTICIPATING",
-    "110": "NON_PARTICIPATING",
-    "111": "NON_PARTICIPATING",
+    "110": "SUCCESS",
+    "111": "SUCCESS",
     "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
@@ -30,25 +30,25 @@
     "120": "NON_PARTICIPATING",
     "121": "SUCCESS",
     "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
     "127": "NON_PARTICIPATING",
     "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
-    "13": "NON_PARTICIPATING",
+    "13": "SUCCESS",
     "130": "NON_PARTICIPATING",
     "131": "NON_PARTICIPATING",
     "132": "NON_PARTICIPATING",
     "133": "NON_PARTICIPATING",
-    "134": "SUCCESS",
-    "135": "NON_PARTICIPATING",
+    "134": "NON_PARTICIPATING",
+    "135": "SUCCESS",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
-    "138": "NON_PARTICIPATING",
+    "138": "SUCCESS",
     "139": "NON_PARTICIPATING",
-    "14": "NON_PARTICIPATING",
+    "14": "SUCCESS",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
@@ -56,32 +56,32 @@
     "144": "NON_PARTICIPATING",
     "145": "NON_PARTICIPATING",
     "146": "NON_PARTICIPATING",
-    "147": "SUCCESS",
+    "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
     "149": "NON_PARTICIPATING",
-    "15": "NON_PARTICIPATING",
+    "15": "SUCCESS",
     "150": "NON_PARTICIPATING",
     "151": "NON_PARTICIPATING",
     "152": "NON_PARTICIPATING",
     "153": "NON_PARTICIPATING",
-    "154": "NON_PARTICIPATING",
+    "154": "SUCCESS",
     "155": "NON_PARTICIPATING",
-    "156": "NON_PARTICIPATING",
+    "156": "SUCCESS",
     "157": "NON_PARTICIPATING",
-    "158": "NON_PARTICIPATING",
+    "158": "SUCCESS",
     "159": "NON_PARTICIPATING",
-    "16": "NON_PARTICIPATING",
+    "16": "SUCCESS",
     "160": "NON_PARTICIPATING",
     "161": "NON_PARTICIPATING",
-    "162": "NON_PARTICIPATING",
+    "162": "SUCCESS",
     "163": "NON_PARTICIPATING",
     "164": "NON_PARTICIPATING",
-    "165": "NON_PARTICIPATING",
+    "165": "SUCCESS",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
-    "17": "NON_PARTICIPATING",
+    "17": "SUCCESS",
     "170": "NON_PARTICIPATING",
     "171": "NON_PARTICIPATING",
     "172": "NON_PARTICIPATING",
@@ -92,15 +92,15 @@
     "177": "NON_PARTICIPATING",
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
-    "18": "NON_PARTICIPATING",
+    "18": "SUCCESS",
     "180": "NON_PARTICIPATING",
     "181": "NON_PARTICIPATING",
     "182": "NON_PARTICIPATING",
     "183": "NON_PARTICIPATING",
-    "184": "NON_PARTICIPATING",
+    "184": "SUCCESS",
     "185": "NON_PARTICIPATING",
-    "186": "NON_PARTICIPATING",
-    "187": "SUCCESS",
+    "186": "SUCCESS",
+    "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
     "189": "NON_PARTICIPATING",
     "19": "NON_PARTICIPATING",
@@ -109,15 +109,15 @@
     "192": "NON_PARTICIPATING",
     "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
-    "195": "NON_PARTICIPATING",
-    "196": "SUCCESS",
+    "195": "SUCCESS",
+    "196": "NON_PARTICIPATING",
     "197": "NON_PARTICIPATING",
     "198": "NON_PARTICIPATING",
     "199": "NON_PARTICIPATING",
-    "2": "NON_PARTICIPATING",
-    "20": "NON_PARTICIPATING",
+    "2": "SUCCESS",
+    "20": "SUCCESS",
     "200": "NON_PARTICIPATING",
-    "201": "NON_PARTICIPATING",
+    "201": "SUCCESS",
     "202": "NON_PARTICIPATING",
     "203": "NON_PARTICIPATING",
     "204": "NON_PARTICIPATING",
@@ -126,138 +126,138 @@
     "207": "NON_PARTICIPATING",
     "208": "NON_PARTICIPATING",
     "209": "NON_PARTICIPATING",
-    "21": "NON_PARTICIPATING",
-    "210": "NON_PARTICIPATING",
+    "21": "SUCCESS",
+    "210": "SUCCESS",
     "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
-    "213": "SUCCESS",
-    "214": "NON_PARTICIPATING",
+    "213": "NON_PARTICIPATING",
+    "214": "SUCCESS",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
     "218": "NON_PARTICIPATING",
-    "219": "NON_PARTICIPATING",
-    "22": "NON_PARTICIPATING",
+    "219": "SUCCESS",
+    "22": "SUCCESS",
     "220": "NON_PARTICIPATING",
     "221": "NON_PARTICIPATING",
     "222": "NON_PARTICIPATING",
-    "223": "NON_PARTICIPATING",
+    "223": "SUCCESS",
     "224": "NON_PARTICIPATING",
-    "225": "NON_PARTICIPATING",
+    "225": "SUCCESS",
     "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "SUCCESS",
     "23": "NON_PARTICIPATING",
     "230": "NON_PARTICIPATING",
     "231": "NON_PARTICIPATING",
-    "232": "SUCCESS",
+    "232": "NON_PARTICIPATING",
     "233": "NON_PARTICIPATING",
-    "234": "SUCCESS",
-    "235": "SUCCESS",
+    "234": "NON_PARTICIPATING",
+    "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
-    "237": "SUCCESS",
+    "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
     "239": "NON_PARTICIPATING",
-    "24": "NON_PARTICIPATING",
+    "24": "SUCCESS",
     "240": "NON_PARTICIPATING",
-    "241": "SUCCESS",
+    "241": "NON_PARTICIPATING",
     "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
-    "244": "NON_PARTICIPATING",
+    "244": "SUCCESS",
     "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
-    "247": "NON_PARTICIPATING",
+    "247": "SUCCESS",
     "248": "NON_PARTICIPATING",
-    "249": "NON_PARTICIPATING",
+    "249": "SUCCESS",
     "25": "SUCCESS",
     "250": "NON_PARTICIPATING",
     "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
-    "253": "SUCCESS",
+    "253": "NON_PARTICIPATING",
     "254": "NON_PARTICIPATING",
     "255": "NON_PARTICIPATING",
-    "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
-    "28": "NON_PARTICIPATING",
-    "29": "NON_PARTICIPATING",
-    "3": "NON_PARTICIPATING",
+    "26": "SUCCESS",
+    "27": "SUCCESS",
+    "28": "SUCCESS",
+    "29": "SUCCESS",
+    "3": "SUCCESS",
     "30": "NON_PARTICIPATING",
-    "31": "NON_PARTICIPATING",
-    "32": "NON_PARTICIPATING",
+    "31": "SUCCESS",
+    "32": "SUCCESS",
     "33": "NON_PARTICIPATING",
     "34": "NON_PARTICIPATING",
-    "35": "NON_PARTICIPATING",
+    "35": "SUCCESS",
     "36": "NON_PARTICIPATING",
-    "37": "NON_PARTICIPATING",
-    "38": "NON_PARTICIPATING",
-    "39": "NON_PARTICIPATING",
-    "4": "NON_PARTICIPATING",
-    "40": "NON_PARTICIPATING",
-    "41": "NON_PARTICIPATING",
+    "37": "SUCCESS",
+    "38": "SUCCESS",
+    "39": "SUCCESS",
+    "4": "SUCCESS",
+    "40": "SUCCESS",
+    "41": "SUCCESS",
     "42": "NON_PARTICIPATING",
-    "43": "NON_PARTICIPATING",
-    "44": "NON_PARTICIPATING",
-    "45": "NON_PARTICIPATING",
+    "43": "SUCCESS",
+    "44": "SUCCESS",
+    "45": "SUCCESS",
     "46": "NON_PARTICIPATING",
     "47": "SUCCESS",
     "48": "NON_PARTICIPATING",
-    "49": "NON_PARTICIPATING",
+    "49": "SUCCESS",
     "5": "NON_PARTICIPATING",
-    "50": "NON_PARTICIPATING",
+    "50": "SUCCESS",
     "51": "SUCCESS",
-    "52": "NON_PARTICIPATING",
+    "52": "SUCCESS",
     "53": "NON_PARTICIPATING",
-    "54": "NON_PARTICIPATING",
-    "55": "NON_PARTICIPATING",
-    "56": "NON_PARTICIPATING",
-    "57": "NON_PARTICIPATING",
+    "54": "SUCCESS",
+    "55": "SUCCESS",
+    "56": "SUCCESS",
+    "57": "SUCCESS",
     "58": "NON_PARTICIPATING",
-    "59": "NON_PARTICIPATING",
+    "59": "SUCCESS",
     "6": "NON_PARTICIPATING",
-    "60": "NON_PARTICIPATING",
+    "60": "FAIL",
     "61": "NON_PARTICIPATING",
-    "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "62": "SUCCESS",
+    "63": "SUCCESS",
     "64": "NON_PARTICIPATING",
     "65": "SUCCESS",
-    "66": "NON_PARTICIPATING",
+    "66": "SUCCESS",
     "67": "SUCCESS",
-    "68": "NON_PARTICIPATING",
-    "69": "SUCCESS",
-    "7": "NON_PARTICIPATING",
+    "68": "SUCCESS",
+    "69": "NON_PARTICIPATING",
+    "7": "SUCCESS",
     "70": "NON_PARTICIPATING",
     "71": "NON_PARTICIPATING",
-    "72": "NON_PARTICIPATING",
+    "72": "SUCCESS",
     "73": "NON_PARTICIPATING",
-    "74": "NON_PARTICIPATING",
-    "75": "NON_PARTICIPATING",
-    "76": "NON_PARTICIPATING",
+    "74": "SUCCESS",
+    "75": "SUCCESS",
+    "76": "SUCCESS",
     "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
     "8": "NON_PARTICIPATING",
-    "80": "NON_PARTICIPATING",
-    "81": "NON_PARTICIPATING",
-    "82": "NON_PARTICIPATING",
+    "80": "SUCCESS",
+    "81": "SUCCESS",
+    "82": "SUCCESS",
     "83": "NON_PARTICIPATING",
-    "84": "NON_PARTICIPATING",
-    "85": "NON_PARTICIPATING",
+    "84": "SUCCESS",
+    "85": "SUCCESS",
     "86": "NON_PARTICIPATING",
-    "87": "NON_PARTICIPATING",
-    "88": "NON_PARTICIPATING",
-    "89": "NON_PARTICIPATING",
+    "87": "SUCCESS",
+    "88": "SUCCESS",
+    "89": "SUCCESS",
     "9": "NON_PARTICIPATING",
     "90": "SUCCESS",
-    "91": "NON_PARTICIPATING",
-    "92": "NON_PARTICIPATING",
+    "91": "SUCCESS",
+    "92": "SUCCESS",
     "93": "NON_PARTICIPATING",
-    "94": "NON_PARTICIPATING",
+    "94": "SUCCESS",
     "95": "SUCCESS",
-    "96": "NON_PARTICIPATING",
-    "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
-    "99": "NON_PARTICIPATING"
+    "96": "SUCCESS",
+    "97": "SUCCESS",
+    "98": "SUCCESS",
+    "99": "SUCCESS"
   },
   "architectures": [
     "GPTOptim"
@@ -268,20 +268,20 @@
     "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
   },
   "block_list": [
-    5123714,
-    5123721,
-    5123731,
-    5123739,
-    5123749
+    5543366,
+    5543371,
+    5543376,
+    5543380,
+    5543385
   ],
   "block_size": 1024,
   "bos_token_id": 50256,
   "embd_pdrop": 0.1,
   "eos_token_id": 50256,
   "initializer_range": 0.02,
-  "inner_step": 5767,
+  "inner_step": 0,
   "inner_steps": 0,
-  "last_allreduce_block": 5072405,
+  "last_allreduce_block": 5542700,
   "layer_norm_epsilon": 1e-05,
   "model_type": "gpt_optimized",
   "n_embd": 1280,

inner_optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:946eaa1371cae7c13287501cdaaa224ee41c1cfe8d886a2cbb643048c698eed3
+size 8081782026

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4620474527da90e3a89002859d39c26db706434a7bd495620c40999afba7e49
+oid sha256:067bffef08b49efdb47728efabad96e810260b26ae0e8fff0063c24ac4a5ab6e
 size 4040701744