Run 4. Outer Step 49. Inner Step 1.

config.json

@@ -1,19 +1,19 @@
 {
-  "_name_or_path": "jockibaba/jockibaba7",
+  "_name_or_path": "distributed/optimized-gpt2-1b",
   "activation_function": "gelu_new",
   "all_reduce_scores": {
     "0": "NON_PARTICIPATING",
     "1": "SUCCESS",
     "10": "NON_PARTICIPATING",
     "100": "NON_PARTICIPATING",
-    "101": "FAIL",
+    "101": "NON_PARTICIPATING",
     "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
     "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "NON_PARTICIPATING",
-    "107": "SUCCESS",
-    "108": "NON_PARTICIPATING",
+    "107": "NON_PARTICIPATING",
+    "108": "SUCCESS",
     "109": "NON_PARTICIPATING",
     "11": "SUCCESS",
     "110": "NON_PARTICIPATING",
@@ -28,13 +28,13 @@
     "119": "NON_PARTICIPATING",
     "12": "SUCCESS",
     "120": "NON_PARTICIPATING",
-    "121": "SUCCESS",
+    "121": "NON_PARTICIPATING",
     "122": "NON_PARTICIPATING",
-    "123": "NON_PARTICIPATING",
+    "123": "SUCCESS",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
     "126": "SUCCESS",
-    "127": "SUCCESS",
+    "127": "NON_PARTICIPATING",
     "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
@@ -46,9 +46,9 @@
     "135": "NON_PARTICIPATING",
     "136": "NON_PARTICIPATING",
     "137": "NON_PARTICIPATING",
-    "138": "SUCCESS",
+    "138": "NON_PARTICIPATING",
     "139": "NON_PARTICIPATING",
-    "14": "SUCCESS",
+    "14": "NON_PARTICIPATING",
     "140": "NON_PARTICIPATING",
     "141": "NON_PARTICIPATING",
     "142": "NON_PARTICIPATING",
@@ -59,11 +59,11 @@
     "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",
+    "153": "SUCCESS",
     "154": "NON_PARTICIPATING",
     "155": "NON_PARTICIPATING",
     "156": "NON_PARTICIPATING",
@@ -79,11 +79,11 @@
     "165": "NON_PARTICIPATING",
     "166": "NON_PARTICIPATING",
     "167": "NON_PARTICIPATING",
-    "168": "NON_PARTICIPATING",
+    "168": "SUCCESS",
     "169": "NON_PARTICIPATING",
     "17": "SUCCESS",
     "170": "NON_PARTICIPATING",
-    "171": "NON_PARTICIPATING",
+    "171": "SUCCESS",
     "172": "NON_PARTICIPATING",
     "173": "SUCCESS",
     "174": "NON_PARTICIPATING",
@@ -93,8 +93,8 @@
     "178": "NON_PARTICIPATING",
     "179": "NON_PARTICIPATING",
     "18": "SUCCESS",
-    "180": "NON_PARTICIPATING",
-    "181": "SUCCESS",
+    "180": "SUCCESS",
+    "181": "NON_PARTICIPATING",
     "182": "NON_PARTICIPATING",
     "183": "NON_PARTICIPATING",
     "184": "NON_PARTICIPATING",
@@ -104,10 +104,10 @@
     "188": "NON_PARTICIPATING",
     "189": "NON_PARTICIPATING",
     "19": "SUCCESS",
-    "190": "SUCCESS",
+    "190": "NON_PARTICIPATING",
     "191": "NON_PARTICIPATING",
-    "192": "SUCCESS",
-    "193": "SUCCESS",
+    "192": "NON_PARTICIPATING",
+    "193": "NON_PARTICIPATING",
     "194": "NON_PARTICIPATING",
     "195": "NON_PARTICIPATING",
     "196": "NON_PARTICIPATING",
@@ -124,10 +124,10 @@
     "205": "NON_PARTICIPATING",
     "206": "NON_PARTICIPATING",
     "207": "NON_PARTICIPATING",
-    "208": "NON_PARTICIPATING",
-    "209": "NON_PARTICIPATING",
+    "208": "SUCCESS",
+    "209": "SUCCESS",
     "21": "SUCCESS",
-    "210": "NON_PARTICIPATING",
+    "210": "SUCCESS",
     "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
     "213": "NON_PARTICIPATING",
@@ -135,43 +135,43 @@
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
     "217": "NON_PARTICIPATING",
-    "218": "NON_PARTICIPATING",
+    "218": "SUCCESS",
     "219": "NON_PARTICIPATING",
     "22": "SUCCESS",
-    "220": "NON_PARTICIPATING",
-    "221": "NON_PARTICIPATING",
+    "220": "SUCCESS",
+    "221": "SUCCESS",
     "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NON_PARTICIPATING",
-    "225": "SUCCESS",
-    "226": "SUCCESS",
+    "225": "NON_PARTICIPATING",
+    "226": "NON_PARTICIPATING",
     "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NON_PARTICIPATING",
+    "229": "SUCCESS",
     "23": "SUCCESS",
     "230": "NON_PARTICIPATING",
-    "231": "NON_PARTICIPATING",
+    "231": "SUCCESS",
     "232": "NON_PARTICIPATING",
     "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "NON_PARTICIPATING",
+    "236": "SUCCESS",
     "237": "NON_PARTICIPATING",
     "238": "SUCCESS",
     "239": "NON_PARTICIPATING",
     "24": "SUCCESS",
-    "240": "SUCCESS",
-    "241": "SUCCESS",
-    "242": "NON_PARTICIPATING",
+    "240": "NON_PARTICIPATING",
+    "241": "NON_PARTICIPATING",
+    "242": "SUCCESS",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "SUCCESS",
+    "245": "NON_PARTICIPATING",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NON_PARTICIPATING",
     "249": "SUCCESS",
     "25": "SUCCESS",
-    "250": "SUCCESS",
+    "250": "NON_PARTICIPATING",
     "251": "NON_PARTICIPATING",
     "252": "NON_PARTICIPATING",
     "253": "NON_PARTICIPATING",
@@ -182,7 +182,7 @@
     "28": "SUCCESS",
     "29": "SUCCESS",
     "3": "SUCCESS",
-    "30": "SUCCESS",
+    "30": "NON_PARTICIPATING",
     "31": "SUCCESS",
     "32": "SUCCESS",
     "33": "NON_PARTICIPATING",
@@ -195,7 +195,7 @@
     "4": "SUCCESS",
     "40": "SUCCESS",
     "41": "SUCCESS",
-    "42": "SUCCESS",
+    "42": "NON_PARTICIPATING",
     "43": "SUCCESS",
     "44": "SUCCESS",
     "45": "SUCCESS",
@@ -209,7 +209,7 @@
     "52": "SUCCESS",
     "53": "SUCCESS",
     "54": "SUCCESS",
-    "55": "NON_PARTICIPATING",
+    "55": "SUCCESS",
     "56": "SUCCESS",
     "57": "SUCCESS",
     "58": "NON_PARTICIPATING",
@@ -268,16 +268,20 @@
     "AutoModelForCausalLM": "distributed/optimized-gpt2-500m--modeling_gpt_optimized.GPTOptim"
   },
   "block_list": [
-    5567583
+    5571227,
+    5571231,
+    5571235,
+    5571240,
+    5571244
   ],
   "block_size": 1024,
   "bos_token_id": 50256,
   "embd_pdrop": 0.1,
   "eos_token_id": 50256,
   "initializer_range": 0.02,
-  "inner_step": 0,
+  "inner_step": 1,
   "inner_steps": 0,
-  "last_allreduce_block": 5559633,
+  "last_allreduce_block": 5570305,
   "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:d9b395cbf42f7c23fcdc3022fc9742cb74ae74f4c2fda93f0bd864d5be5eb776
+oid sha256:7202d2b8c41c6557eb278b04542d5fb4b4f31b6e3cfa9c5f5f17b449a7765d21
 size 8081782503

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7337cb6da47e88bd0db5c21e07c018d4eef5ebc3e07ef5642c5e6ab31cae4513
+oid sha256:073f755caf3e392635f430c462f8c8780371c2acea990ffca0cf67ed25f36c64
 size 4040701744