Run 11. Outer Step 215. Inner Step 9.

config.json

@@ -2,46 +2,46 @@
   "_name_or_path": "dstrbtd/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
-    "10": "SUCCESS",
+    "1": "SUCCESS",
+    "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
-    "101": "NON_PARTICIPATING",
+    "101": "NOT_ALIVE",
     "102": "NOT_ALIVE",
-    "103": "NON_PARTICIPATING",
-    "104": "SUCCESS",
+    "103": "SUCCESS",
+    "104": "NON_PARTICIPATING",
     "105": "NON_PARTICIPATING",
     "106": "SUCCESS",
-    "107": "SUCCESS",
+    "107": "NON_PARTICIPATING",
     "108": "SUCCESS",
     "109": "SUCCESS",
     "11": "NOT_ALIVE",
     "110": "NON_PARTICIPATING",
-    "111": "NOT_ALIVE",
+    "111": "NON_PARTICIPATING",
     "112": "NOT_ALIVE",
     "113": "NON_PARTICIPATING",
     "114": "NOT_ALIVE",
-    "115": "NON_PARTICIPATING",
+    "115": "SUCCESS",
     "116": "NON_PARTICIPATING",
     "117": "NOT_ALIVE",
     "118": "NOT_ALIVE",
     "119": "SUCCESS",
     "12": "NOT_ALIVE",
     "120": "NON_PARTICIPATING",
-    "121": "NON_PARTICIPATING",
+    "121": "SUCCESS",
     "122": "NOT_ALIVE",
     "123": "SUCCESS",
     "124": "NOT_ALIVE",
     "125": "NOT_ALIVE",
     "126": "NOT_ALIVE",
     "127": "SUCCESS",
-    "128": "NON_PARTICIPATING",
-    "129": "NON_PARTICIPATING",
-    "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "128": "SUCCESS",
+    "129": "NOT_ALIVE",
+    "13": "SUCCESS",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
     "133": "NOT_ALIVE",
-    "134": "NON_PARTICIPATING",
+    "134": "SUCCESS",
     "135": "SUCCESS",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
@@ -49,27 +49,27 @@
     "139": "NON_PARTICIPATING",
     "14": "NON_PARTICIPATING",
     "140": "SUCCESS",
-    "141": "SUCCESS",
-    "142": "NON_PARTICIPATING",
-    "143": "NOT_ALIVE",
+    "141": "NOT_ALIVE",
+    "142": "NOT_ALIVE",
+    "143": "SUCCESS",
     "144": "NOT_ALIVE",
     "145": "NOT_ALIVE",
     "146": "SUCCESS",
-    "147": "SUCCESS",
+    "147": "NON_PARTICIPATING",
     "148": "SUCCESS",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NON_PARTICIPATING",
+    "150": "FAIL",
     "151": "NOT_ALIVE",
     "152": "NOT_ALIVE",
-    "153": "NON_PARTICIPATING",
+    "153": "SUCCESS",
     "154": "NOT_ALIVE",
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "SUCCESS",
     "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
-    "16": "NON_PARTICIPATING",
+    "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "SUCCESS",
@@ -84,12 +84,12 @@
     "170": "NOT_ALIVE",
     "171": "NON_PARTICIPATING",
     "172": "NOT_ALIVE",
-    "173": "NON_PARTICIPATING",
+    "173": "SUCCESS",
     "174": "NOT_ALIVE",
-    "175": "SUCCESS",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
     "177": "NOT_ALIVE",
-    "178": "SUCCESS",
+    "178": "NON_PARTICIPATING",
     "179": "NOT_ALIVE",
     "18": "NOT_ALIVE",
     "180": "NOT_ALIVE",
@@ -97,8 +97,8 @@
     "182": "SUCCESS",
     "183": "NOT_ALIVE",
     "184": "NON_PARTICIPATING",
-    "185": "SUCCESS",
-    "186": "NON_PARTICIPATING",
+    "185": "NOT_ALIVE",
+    "186": "SUCCESS",
     "187": "NOT_ALIVE",
     "188": "NOT_ALIVE",
     "189": "NOT_ALIVE",
@@ -108,7 +108,7 @@
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
-    "195": "NON_PARTICIPATING",
+    "195": "SUCCESS",
     "196": "NOT_ALIVE",
     "197": "SUCCESS",
     "198": "NOT_ALIVE",
@@ -116,46 +116,46 @@
     "2": "NOT_ALIVE",
     "20": "SUCCESS",
     "200": "NON_PARTICIPATING",
-    "201": "NON_PARTICIPATING",
+    "201": "NOT_ALIVE",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
     "205": "NOT_ALIVE",
-    "206": "NON_PARTICIPATING",
+    "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
     "208": "NOT_ALIVE",
-    "209": "SUCCESS",
+    "209": "NON_PARTICIPATING",
     "21": "NOT_ALIVE",
-    "210": "SUCCESS",
+    "210": "NON_PARTICIPATING",
     "211": "NOT_ALIVE",
     "212": "NOT_ALIVE",
     "213": "NON_PARTICIPATING",
-    "214": "NOT_ALIVE",
+    "214": "SUCCESS",
     "215": "NOT_ALIVE",
     "216": "NON_PARTICIPATING",
-    "217": "SUCCESS",
+    "217": "NON_PARTICIPATING",
     "218": "NOT_ALIVE",
     "219": "NOT_ALIVE",
     "22": "NOT_ALIVE",
-    "220": "NON_PARTICIPATING",
+    "220": "NOT_ALIVE",
     "221": "NON_PARTICIPATING",
     "222": "NOT_ALIVE",
     "223": "SUCCESS",
     "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
     "226": "SUCCESS",
-    "227": "NON_PARTICIPATING",
+    "227": "NOT_ALIVE",
     "228": "NOT_ALIVE",
     "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "23": "NON_PARTICIPATING",
     "230": "SUCCESS",
-    "231": "NOT_ALIVE",
+    "231": "SUCCESS",
     "232": "NOT_ALIVE",
-    "233": "FAIL",
+    "233": "SUCCESS",
     "234": "NON_PARTICIPATING",
     "235": "NON_PARTICIPATING",
-    "236": "SUCCESS",
-    "237": "NON_PARTICIPATING",
+    "236": "NOT_ALIVE",
+    "237": "SUCCESS",
     "238": "SUCCESS",
     "239": "NOT_ALIVE",
     "24": "SUCCESS",
@@ -164,8 +164,8 @@
     "242": "NOT_ALIVE",
     "243": "NOT_ALIVE",
     "244": "NOT_ALIVE",
-    "245": "NON_PARTICIPATING",
-    "246": "NON_PARTICIPATING",
+    "245": "SUCCESS",
+    "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
@@ -176,7 +176,7 @@
     "253": "NOT_ALIVE",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "SUCCESS",
     "27": "NOT_ALIVE",
     "28": "NOT_ALIVE",
     "29": "NOT_ALIVE",
@@ -189,15 +189,15 @@
     "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "SUCCESS",
-    "38": "SUCCESS",
+    "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
-    "4": "NOT_ALIVE",
+    "4": "NON_PARTICIPATING",
     "40": "NOT_ALIVE",
-    "41": "NOT_ALIVE",
-    "42": "NON_PARTICIPATING",
+    "41": "SUCCESS",
+    "42": "SUCCESS",
     "43": "NOT_ALIVE",
     "44": "NOT_ALIVE",
-    "45": "SUCCESS",
+    "45": "NON_PARTICIPATING",
     "46": "NOT_ALIVE",
     "47": "NOT_ALIVE",
     "48": "NON_PARTICIPATING",
@@ -207,7 +207,7 @@
     "51": "NON_PARTICIPATING",
     "52": "NOT_ALIVE",
     "53": "NON_PARTICIPATING",
-    "54": "SUCCESS",
+    "54": "NON_PARTICIPATING",
     "55": "NON_PARTICIPATING",
     "56": "NOT_ALIVE",
     "57": "NOT_ALIVE",
@@ -216,13 +216,13 @@
     "6": "NOT_ALIVE",
     "60": "NON_PARTICIPATING",
     "61": "SUCCESS",
-    "62": "NOT_ALIVE",
-    "63": "SUCCESS",
+    "62": "NON_PARTICIPATING",
+    "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
     "65": "SUCCESS",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
-    "68": "SUCCESS",
+    "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NOT_ALIVE",
@@ -231,16 +231,16 @@
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
     "75": "NOT_ALIVE",
-    "76": "NON_PARTICIPATING",
+    "76": "SUCCESS",
     "77": "NOT_ALIVE",
     "78": "NOT_ALIVE",
-    "79": "SUCCESS",
+    "79": "NON_PARTICIPATING",
     "8": "SUCCESS",
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
-    "82": "SUCCESS",
-    "83": "SUCCESS",
-    "84": "NON_PARTICIPATING",
+    "82": "NOT_ALIVE",
+    "83": "NOT_ALIVE",
+    "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
     "87": "NOT_ALIVE",
@@ -251,12 +251,12 @@
     "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
     "93": "NOT_ALIVE",
-    "94": "NON_PARTICIPATING",
+    "94": "NOT_ALIVE",
     "95": "NOT_ALIVE",
     "96": "NOT_ALIVE",
     "97": "SUCCESS",
     "98": "SUCCESS",
-    "99": "NON_PARTICIPATING"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,8 +264,8 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6664790,
-    6664834
+    6664977,
+    6665006
   ],
   "bos_token_id": 1,
   "dtype": "float32",
@@ -274,9 +274,9 @@
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 45,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6661467,
+  "last_allreduce_block": 6664911,
   "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 sha256:59e7a7bd94923cc39ea9cab2cd24383a3eec633edb3323efc123e0fcc7aaa4ab
+oid sha256:20d4c814a8041900c04080e6850cb9295d784140dc418d2d7b75535caa34ec56
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe012781bc70bb95e8b07b7dcd3c3b23a3e3d08ff654c9657c03245830512631
+oid sha256:57dbd1f317c1d0a501c0a4e855314743d4745568656d2df5eb9ede2846ce6bdf
 size 4924926906

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12c56d137227dfe1fd951eeba8bf21587e9111947d4067b6ca2aa7b19496ffc6
+oid sha256:629515b208fd84c90e79f0de6070b916d3e5ff2b0b5bab40b377046c6856b9c8
 size 4400216536