Run 6. Outer Step 46. Inner Step 5.

config.json

@@ -9,7 +9,7 @@
     "102": "NOT_ALIVE",
     "103": "NON_PARTICIPATING",
     "104": "NOT_ALIVE",
-    "105": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
     "108": "SUCCESS",
@@ -35,7 +35,7 @@
     "126": "NOT_ALIVE",
     "127": "NOT_ALIVE",
     "128": "NOT_ALIVE",
-    "129": "SUCCESS",
+    "129": "NOT_ALIVE",
     "13": "NOT_ALIVE",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
@@ -45,10 +45,10 @@
     "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
-    "138": "SUCCESS",
+    "138": "NON_PARTICIPATING",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
-    "140": "NOT_ALIVE",
+    "140": "SUCCESS",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
     "143": "NOT_ALIVE",
@@ -63,7 +63,7 @@
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "NOT_ALIVE",
-    "154": "SUCCESS",
+    "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
@@ -73,10 +73,10 @@
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "NON_PARTICIPATING",
+    "163": "NOT_ALIVE",
     "164": "SUCCESS",
-    "165": "NOT_ALIVE",
-    "166": "SUCCESS",
+    "165": "NON_PARTICIPATING",
+    "166": "NOT_ALIVE",
     "167": "NOT_ALIVE",
     "168": "NOT_ALIVE",
     "169": "NOT_ALIVE",
@@ -90,7 +90,7 @@
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
     "178": "NOT_ALIVE",
-    "179": "NON_PARTICIPATING",
+    "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
@@ -106,13 +106,13 @@
     "190": "SUCCESS",
     "191": "SUCCESS",
     "192": "NOT_ALIVE",
-    "193": "NOT_ALIVE",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NOT_ALIVE",
-    "199": "SUCCESS",
+    "199": "NOT_ALIVE",
     "2": "SUCCESS",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
@@ -128,26 +128,26 @@
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "NOT_ALIVE",
-    "212": "SUCCESS",
+    "212": "NOT_ALIVE",
     "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "NON_PARTICIPATING",
-    "217": "NOT_ALIVE",
-    "218": "NOT_ALIVE",
+    "216": "SUCCESS",
+    "217": "FAIL",
+    "218": "NON_PARTICIPATING",
     "219": "SUCCESS",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
     "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
-    "224": "NON_PARTICIPATING",
+    "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "NON_PARTICIPATING",
     "227": "NOT_ALIVE",
     "228": "SUCCESS",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "SUCCESS",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
@@ -155,12 +155,12 @@
     "234": "NOT_ALIVE",
     "235": "NOT_ALIVE",
     "236": "NOT_ALIVE",
-    "237": "NON_PARTICIPATING",
+    "237": "NOT_ALIVE",
     "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "NOT_ALIVE",
     "240": "NOT_ALIVE",
-    "241": "NOT_ALIVE",
+    "241": "NON_PARTICIPATING",
     "242": "SUCCESS",
     "243": "NOT_ALIVE",
     "244": "SUCCESS",
@@ -200,13 +200,13 @@
     "45": "NOT_ALIVE",
     "46": "SUCCESS",
     "47": "NOT_ALIVE",
-    "48": "NOT_ALIVE",
+    "48": "NON_PARTICIPATING",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
     "51": "NOT_ALIVE",
     "52": "NOT_ALIVE",
-    "53": "SUCCESS",
+    "53": "NOT_ALIVE",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
     "56": "NOT_ALIVE",
@@ -236,7 +236,7 @@
     "78": "SUCCESS",
     "79": "NOT_ALIVE",
     "8": "NOT_ALIVE",
-    "80": "NOT_ALIVE",
+    "80": "SUCCESS",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
     "83": "SUCCESS",
@@ -253,7 +253,7 @@
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NON_PARTICIPATING",
-    "96": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "NOT_ALIVE",
     "99": "SUCCESS"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6159732,
-    6159751
+    6159865,
+    6159884
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 70,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6159037,
+  "last_allreduce_block": 6159733,
   "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 sha256:2000d4df881c1fa53537957dd03b5c709151fcfddebbbb1f8d3a3968e978c9c2
+oid sha256:89a27cedce56abc31e7931757930b964ad82a91acd994ec417de8e8e71efc5ce
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6c34a9f534e1dbde4a8567012d7c857d2446efb5587aef732a9b5ec9b40fdea
+oid sha256:00fb2ecd6e94d99dc76c64449b5fc1598cb8a062a2ddca03a623da6541f918c2
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de9d9aab51994af8a4c1fbd6d39b56080bb5603b90d2ed377cf3e43f76757852
+oid sha256:d29d9b892773c6e12a29b88220c11a9891258167233ccb2816a64e4fccb627e4
 size 4400216536