Run 11. Outer Step 328. Inner Step 9.

config.json

@@ -9,49 +9,49 @@
     "102": "NOT_ALIVE",
     "103": "NOT_ALIVE",
     "104": "SUCCESS",
-    "105": "NON_PARTICIPATING",
+    "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
     "107": "SUCCESS",
-    "108": "NOT_ALIVE",
-    "109": "SUCCESS",
+    "108": "SUCCESS",
+    "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
-    "111": "NON_PARTICIPATING",
-    "112": "SUCCESS",
+    "111": "NOT_ALIVE",
+    "112": "NON_PARTICIPATING",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
     "115": "NON_PARTICIPATING",
-    "116": "SUCCESS",
+    "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NOT_ALIVE",
-    "121": "SUCCESS",
+    "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
     "123": "NON_PARTICIPATING",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
+    "128": "SUCCESS",
     "129": "NON_PARTICIPATING",
     "13": "SUCCESS",
-    "130": "NON_PARTICIPATING",
+    "130": "SUCCESS",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NON_PARTICIPATING",
-    "135": "NON_PARTICIPATING",
+    "134": "SUCCESS",
+    "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
     "138": "NOT_ALIVE",
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
-    "140": "NON_PARTICIPATING",
-    "141": "NOT_ALIVE",
+    "140": "SUCCESS",
+    "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
-    "143": "NON_PARTICIPATING",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
@@ -62,7 +62,7 @@
     "150": "NOT_ALIVE",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "SUCCESS",
+    "153": "NON_PARTICIPATING",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
@@ -70,9 +70,9 @@
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
-    "160": "NOT_ALIVE",
+    "160": "SUCCESS",
     "161": "NOT_ALIVE",
-    "162": "SUCCESS",
+    "162": "NON_PARTICIPATING",
     "163": "NOT_ALIVE",
     "164": "NOT_ALIVE",
     "165": "NOT_ALIVE",
@@ -94,19 +94,19 @@
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
-    "182": "SUCCESS",
+    "182": "NOT_ALIVE",
     "183": "NOT_ALIVE",
-    "184": "SUCCESS",
+    "184": "NON_PARTICIPATING",
     "185": "SUCCESS",
-    "186": "NOT_ALIVE",
-    "187": "NON_PARTICIPATING",
+    "186": "NON_PARTICIPATING",
+    "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
+    "193": "SUCCESS",
     "194": "NOT_ALIVE",
     "195": "NOT_ALIVE",
     "196": "NOT_ALIVE",
@@ -114,17 +114,17 @@
     "198": "SUCCESS",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
-    "20": "NOT_ALIVE",
+    "20": "SUCCESS",
     "200": "SUCCESS",
     "201": "NON_PARTICIPATING",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NOT_ALIVE",
+    "206": "NON_PARTICIPATING",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "SUCCESS",
@@ -132,13 +132,13 @@
     "213": "NON_PARTICIPATING",
     "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
-    "216": "NOT_ALIVE",
-    "217": "SUCCESS",
+    "216": "NON_PARTICIPATING",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
     "220": "NOT_ALIVE",
-    "221": "SUCCESS",
+    "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
     "224": "NOT_ALIVE",
@@ -148,43 +148,43 @@
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
     "23": "NOT_ALIVE",
-    "230": "NON_PARTICIPATING",
+    "230": "SUCCESS",
     "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "SUCCESS",
+    "233": "NON_PARTICIPATING",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "NON_PARTICIPATING",
-    "237": "NON_PARTICIPATING",
-    "238": "NON_PARTICIPATING",
+    "236": "SUCCESS",
+    "237": "SUCCESS",
+    "238": "SUCCESS",
     "239": "NOT_ALIVE",
-    "24": "NOT_ALIVE",
+    "24": "NON_PARTICIPATING",
     "240": "NOT_ALIVE",
     "241": "NON_PARTICIPATING",
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "NOT_ALIVE",
+    "245": "SUCCESS",
     "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "SUCCESS",
+    "251": "NOT_ALIVE",
     "252": "SUCCESS",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
     "3": "NOT_ALIVE",
     "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
+    "31": "NON_PARTICIPATING",
     "32": "NOT_ALIVE",
-    "33": "NON_PARTICIPATING",
+    "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
@@ -216,47 +216,47 @@
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
+    "62": "SUCCESS",
     "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "NON_PARTICIPATING",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "NOT_ALIVE",
+    "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
-    "75": "NOT_ALIVE",
+    "75": "NON_PARTICIPATING",
     "76": "NOT_ALIVE",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "SUCCESS",
     "8": "SUCCESS",
-    "80": "NON_PARTICIPATING",
-    "81": "NOT_ALIVE",
+    "80": "SUCCESS",
+    "81": "NON_PARTICIPATING",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
-    "87": "NOT_ALIVE",
-    "88": "NOT_ALIVE",
+    "87": "SUCCESS",
+    "88": "NON_PARTICIPATING",
     "89": "NON_PARTICIPATING",
     "9": "NOT_ALIVE",
     "90": "NOT_ALIVE",
-    "91": "NOT_ALIVE",
+    "91": "SUCCESS",
     "92": "SUCCESS",
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NOT_ALIVE",
+    "97": "NON_PARTICIPATING",
     "98": "SUCCESS",
-    "99": "NON_PARTICIPATING"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6713408,
-    6713441
+    6713611,
+    6713655
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 36,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6713169,
+  "last_allreduce_block": 6713550,
   "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 sha256:83cced60966d936ebf735be370bf0327a79380c36d09bcb99c2cd9d51b8dca86
+oid sha256:dadd415b9c51eaefce949e4db6b751a647a524ce4f5e14163266b3c3f16f3384
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:adc005d1937dd03f34dad1046ea9ef487d334f2d8536c76f55e59c921481bd17
+oid sha256:c04ae828532e3d0ca25251811827783ce8fbf883f1bf664199cfe2240b010dc5
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f52e0bfc99e0356a44cb194c64f46c02f9d01a20e3c887827a303b83729a654
+oid sha256:030458617e975d6bb2cb1d03f09e6affa9248a536dd28502ca005e6a7f5bfb68
 size 4400216536