Run 11. Outer Step 250. Inner Step 9.

config.json

@@ -3,11 +3,11 @@
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
     "1": "NOT_ALIVE",
-    "10": "NON_PARTICIPATING",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "NOT_ALIVE",
+    "101": "SUCCESS",
     "102": "NOT_ALIVE",
-    "103": "NON_PARTICIPATING",
+    "103": "SUCCESS",
     "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
@@ -15,11 +15,11 @@
     "108": "NON_PARTICIPATING",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "NON_PARTICIPATING",
     "113": "NOT_ALIVE",
-    "114": "SUCCESS",
+    "114": "NON_PARTICIPATING",
     "115": "NOT_ALIVE",
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
@@ -27,14 +27,14 @@
     "119": "NON_PARTICIPATING",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "NOT_ALIVE",
+    "121": "SUCCESS",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
-    "125": "NOT_ALIVE",
+    "125": "NON_PARTICIPATING",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "SUCCESS",
+    "128": "NOT_ALIVE",
     "129": "NOT_ALIVE",
     "13": "SUCCESS",
     "130": "NON_PARTICIPATING",
@@ -42,7 +42,7 @@
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
     "134": "SUCCESS",
-    "135": "NOT_ALIVE",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
@@ -55,14 +55,14 @@
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "NOT_ALIVE",
-    "148": "NON_PARTICIPATING",
+    "147": "SUCCESS",
+    "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NON_PARTICIPATING",
     "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NOT_ALIVE",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -70,34 +70,34 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "SUCCESS",
+    "160": "NON_PARTICIPATING",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
     "166": "NOT_ALIVE",
-    "167": "NOT_ALIVE",
+    "167": "NON_PARTICIPATING",
     "168": "NOT_ALIVE",
     "169": "SUCCESS",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "NOT_ALIVE",
     "174": "NOT_ALIVE",
-    "175": "NOT_ALIVE",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NON_PARTICIPATING",
-    "179": "NOT_ALIVE",
+    "178": "SUCCESS",
+    "179": "FAIL",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
+    "185": "NOT_ALIVE",
     "186": "NON_PARTICIPATING",
     "187": "SUCCESS",
     "188": "NOT_ALIVE",
@@ -106,34 +106,34 @@
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "SUCCESS",
+    "197": "NOT_ALIVE",
     "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "SUCCESS",
+    "20": "NOT_ALIVE",
     "200": "NON_PARTICIPATING",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "SUCCESS",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
     "209": "SUCCESS",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
-    "211": "NON_PARTICIPATING",
+    "211": "SUCCESS",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
     "214": "SUCCESS",
-    "215": "SUCCESS",
+    "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "NON_PARTICIPATING",
+    "217": "SUCCESS",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
@@ -144,10 +144,10 @@
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "SUCCESS",
+    "227": "NON_PARTICIPATING",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "SUCCESS",
     "230": "SUCCESS",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
@@ -215,14 +215,14 @@
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
-    "61": "SUCCESS",
+    "61": "NOT_ALIVE",
     "62": "NON_PARTICIPATING",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NON_PARTICIPATING",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
@@ -231,7 +231,7 @@
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "SUCCESS",
+    "76": "FAIL",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
@@ -239,19 +239,19 @@
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "NON_PARTICIPATING",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
-    "86": "NON_PARTICIPATING",
-    "87": "NON_PARTICIPATING",
+    "86": "NOT_ALIVE",
+    "87": "SUCCESS",
     "88": "NON_PARTICIPATING",
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "FAIL",
+    "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NOT_ALIVE",
+    "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "NOT_ALIVE",
@@ -264,7 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6678209
+    6679394,
+    6679423
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -273,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 0,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6678056,
+  "last_allreduce_block": 6679196,
   "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 sha256:38cb3abcc93a62f093ecc53c483f4385a88e27dc426b314406f253c1b13d0acc
+oid sha256:2ed1f389bff130962f5302e793381c577094b5ede4d01407f157f7cf639bf01b
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2acebd3920299dfb825c274a2acb878c525b5251b6bc6ca64ac83c5e87311f4
+oid sha256:8afc916e3246a5da05126ab20cc4e0455283cff43566f9000fce1333a5d27615
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3ae360065ac934dea0cae2d4d4d4f387f9823ff1c0dfd1d75548f5bbd093e74
+oid sha256:4bd6048ab60509ecc45baa4709831de3f353a66a63389738d918a6809b53974c
 size 4400216536