Run 11. Outer Step 212. Inner Step 9.

config.json

@@ -2,26 +2,26 @@
   "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NON_PARTICIPATING",
-    "10": "NOT_ALIVE",
+    "1": "NOT_ALIVE",
+    "10": "SUCCESS",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
+    "101": "SUCCESS",
     "102": "NOT_ALIVE",
-    "103": "NOT_ALIVE",
-    "104": "NON_PARTICIPATING",
+    "103": "NON_PARTICIPATING",
+    "104": "SUCCESS",
     "105": "NON_PARTICIPATING",
-    "106": "SUCCESS",
-    "107": "NON_PARTICIPATING",
+    "106": "NOT_ALIVE",
+    "107": "SUCCESS",
     "108": "SUCCESS",
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "SUCCESS",
     "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
-    "113": "NON_PARTICIPATING",
+    "113": "NOT_ALIVE",
     "114": "NOT_ALIVE",
     "115": "SUCCESS",
-    "116": "NON_PARTICIPATING",
+    "116": "SUCCESS",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
@@ -29,14 +29,14 @@
     "120": "SUCCESS",
     "121": "NON_PARTICIPATING",
     "122": "NOT_ALIVE",
-    "123": "SUCCESS",
+    "123": "NOT_ALIVE",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NOT_ALIVE",
+    "128": "NON_PARTICIPATING",
     "129": "NON_PARTICIPATING",
-    "13": "NOT_ALIVE",
+    "13": "SUCCESS",
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
@@ -50,26 +50,26 @@
     "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
     "141": "SUCCESS",
-    "142": "NON_PARTICIPATING",
-    "143": "NON_PARTICIPATING",
+    "142": "NOT_ALIVE",
+    "143": "NOT_ALIVE",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
-    "146": "NOT_ALIVE",
+    "146": "SUCCESS",
     "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NON_PARTICIPATING",
+    "150": "SUCCESS",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
     "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
-    "157": "SUCCESS",
+    "157": "NOT_ALIVE",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
-    "16": "NON_PARTICIPATING",
+    "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
@@ -77,7 +77,7 @@
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
     "166": "NOT_ALIVE",
-    "167": "NON_PARTICIPATING",
+    "167": "SUCCESS",
     "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
@@ -89,16 +89,16 @@
     "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "NON_PARTICIPATING",
+    "178": "SUCCESS",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
     "181": "NOT_ALIVE",
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
-    "184": "NOT_ALIVE",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
+    "184": "SUCCESS",
+    "185": "NOT_ALIVE",
+    "186": "SUCCESS",
     "187": "SUCCESS",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -106,11 +106,11 @@
     "190": "NOT_ALIVE",
     "191": "NOT_ALIVE",
     "192": "NOT_ALIVE",
-    "193": "SUCCESS",
+    "193": "NOT_ALIVE",
     "194": "NOT_ALIVE",
     "195": "NON_PARTICIPATING",
     "196": "NOT_ALIVE",
-    "197": "NOT_ALIVE",
+    "197": "SUCCESS",
     "198": "NOT_ALIVE",
     "199": "SUCCESS",
     "2": "NOT_ALIVE",
@@ -130,7 +130,7 @@
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NOT_ALIVE",
+    "214": "SUCCESS",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
     "217": "SUCCESS",
@@ -143,20 +143,20 @@
     "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "SUCCESS",
+    "226": "NOT_ALIVE",
     "227": "SUCCESS",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "SUCCESS",
+    "23": "NON_PARTICIPATING",
     "230": "SUCCESS",
-    "231": "NON_PARTICIPATING",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
     "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
+    "236": "NOT_ALIVE",
     "237": "SUCCESS",
-    "238": "SUCCESS",
+    "238": "NOT_ALIVE",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
     "240": "NOT_ALIVE",
@@ -164,15 +164,15 @@
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "NOT_ALIVE",
+    "245": "SUCCESS",
     "246": "NON_PARTICIPATING",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
-    "251": "SUCCESS",
-    "252": "SUCCESS",
+    "251": "NON_PARTICIPATING",
+    "252": "NOT_ALIVE",
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
@@ -189,7 +189,7 @@
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
-    "38": "NON_PARTICIPATING",
+    "38": "SUCCESS",
     "39": "NOT_ALIVE",
     "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
@@ -197,10 +197,10 @@
     "42": "SUCCESS",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
-    "45": "SUCCESS",
+    "45": "NON_PARTICIPATING",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
-    "48": "NON_PARTICIPATING",
+    "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
     "5": "NOT_ALIVE",
     "50": "NOT_ALIVE",
@@ -208,7 +208,7 @@
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
     "54": "SUCCESS",
-    "55": "NON_PARTICIPATING",
+    "55": "SUCCESS",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
     "58": "NOT_ALIVE",
@@ -216,22 +216,22 @@
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NON_PARTICIPATING",
-    "63": "NON_PARTICIPATING",
+    "62": "NOT_ALIVE",
+    "63": "SUCCESS",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "SUCCESS",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
-    "71": "SUCCESS",
+    "71": "NON_PARTICIPATING",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
+    "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
     "79": "NON_PARTICIPATING",
@@ -239,7 +239,7 @@
     "80": "SUCCESS",
     "81": "NOT_ALIVE",
     "82": "SUCCESS",
-    "83": "SUCCESS",
+    "83": "NON_PARTICIPATING",
     "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -248,15 +248,15 @@
     "89": "NOT_ALIVE",
     "9": "NOT_ALIVE",
     "90": "NON_PARTICIPATING",
-    "91": "NON_PARTICIPATING",
+    "91": "NOT_ALIVE",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
     "94": "NON_PARTICIPATING",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
-    "97": "NON_PARTICIPATING",
-    "98": "NON_PARTICIPATING",
-    "99": "NOT_ALIVE"
+    "97": "SUCCESS",
+    "98": "SUCCESS",
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6663672,
-    6663698
+    6663750,
+    6663821
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 59,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6663352,
+  "last_allreduce_block": 6663738,
   "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:f544c0f7ed24ccba0a14dbc715d9bc6ed01ca3358f64ac471e8ada61935b63d5
+oid sha256:c0d76f30b59b8d92c1fedc8c314329e859576891785acdba74e74e6055198425
 size 88390091

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc4110c2c4b8404231a96854a6cf5f575958cada67b67c7922dc4e1e5fee17de
+oid sha256:4676571e3c9ec4bc76fe6194c294b356c3fab4e2137ce8be256c4231e1f0f694
 size 4924928111

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df93bf3870a166f392726649a81567d31f549428c7a91fa80662686280e99af0
+oid sha256:0167f2cca2b0ce37ea960c3b39b223976547fc04ba8a56bbe06257b59e603a45
 size 4400216536