Run 5. Outer Step 150. Inner Step 0. Peers 1.

config.json

@@ -2,72 +2,72 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "NOT_ALIVE",
-    "10": "NOT_ALIVE",
+    "1": "NON_PARTICIPATING",
+    "10": "NON_PARTICIPATING",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
-    "102": "NOT_ALIVE",
+    "102": "NON_PARTICIPATING",
     "103": "NON_PARTICIPATING",
-    "104": "NON_PARTICIPATING",
-    "105": "NOT_ALIVE",
-    "106": "NOT_ALIVE",
+    "104": "NOT_ALIVE",
+    "105": "NON_PARTICIPATING",
+    "106": "NON_PARTICIPATING",
     "107": "NON_PARTICIPATING",
     "108": "NOT_ALIVE",
-    "109": "NOT_ALIVE",
-    "11": "NOT_ALIVE",
-    "110": "NOT_ALIVE",
+    "109": "NON_PARTICIPATING",
+    "11": "NON_PARTICIPATING",
+    "110": "NON_PARTICIPATING",
     "111": "NOT_ALIVE",
     "112": "NON_PARTICIPATING",
-    "113": "NOT_ALIVE",
+    "113": "NON_PARTICIPATING",
     "114": "NON_PARTICIPATING",
-    "115": "NOT_ALIVE",
+    "115": "NON_PARTICIPATING",
     "116": "NON_PARTICIPATING",
-    "117": "NOT_ALIVE",
+    "117": "NON_PARTICIPATING",
     "118": "NOT_ALIVE",
     "119": "NOT_ALIVE",
     "12": "NOT_ALIVE",
-    "120": "NOT_ALIVE",
+    "120": "NON_PARTICIPATING",
     "121": "NON_PARTICIPATING",
-    "122": "NOT_ALIVE",
+    "122": "NON_PARTICIPATING",
     "123": "NON_PARTICIPATING",
     "124": "NON_PARTICIPATING",
     "125": "NON_PARTICIPATING",
     "126": "NON_PARTICIPATING",
     "127": "NOT_ALIVE",
     "128": "NON_PARTICIPATING",
-    "129": "NOT_ALIVE",
+    "129": "NON_PARTICIPATING",
     "13": "NOT_ALIVE",
-    "130": "NOT_ALIVE",
+    "130": "NON_PARTICIPATING",
     "131": "NOT_ALIVE",
-    "132": "NOT_ALIVE",
+    "132": "NON_PARTICIPATING",
     "133": "NON_PARTICIPATING",
     "134": "NON_PARTICIPATING",
     "135": "NON_PARTICIPATING",
     "136": "NON_PARTICIPATING",
-    "137": "NOT_ALIVE",
+    "137": "NON_PARTICIPATING",
     "138": "NON_PARTICIPATING",
-    "139": "NOT_ALIVE",
+    "139": "NON_PARTICIPATING",
     "14": "NOT_ALIVE",
     "140": "NOT_ALIVE",
     "141": "NON_PARTICIPATING",
     "142": "NOT_ALIVE",
-    "143": "NOT_ALIVE",
-    "144": "NOT_ALIVE",
-    "145": "NON_PARTICIPATING",
+    "143": "NON_PARTICIPATING",
+    "144": "NON_PARTICIPATING",
+    "145": "NOT_ALIVE",
     "146": "NOT_ALIVE",
-    "147": "NOT_ALIVE",
+    "147": "NON_PARTICIPATING",
     "148": "NON_PARTICIPATING",
     "149": "NOT_ALIVE",
     "15": "NOT_ALIVE",
-    "150": "NOT_ALIVE",
+    "150": "NON_PARTICIPATING",
     "151": "NOT_ALIVE",
     "152": "NON_PARTICIPATING",
     "153": "NOT_ALIVE",
-    "154": "NOT_ALIVE",
+    "154": "NON_PARTICIPATING",
     "155": "NON_PARTICIPATING",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "NOT_ALIVE",
+    "158": "NON_PARTICIPATING",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "NOT_ALIVE",
@@ -76,37 +76,37 @@
     "163": "NOT_ALIVE",
     "164": "NON_PARTICIPATING",
     "165": "NON_PARTICIPATING",
-    "166": "NOT_ALIVE",
+    "166": "NON_PARTICIPATING",
     "167": "NOT_ALIVE",
     "168": "NOT_ALIVE",
     "169": "NOT_ALIVE",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
     "171": "NOT_ALIVE",
-    "172": "NOT_ALIVE",
+    "172": "NON_PARTICIPATING",
     "173": "NOT_ALIVE",
-    "174": "NOT_ALIVE",
-    "175": "NOT_ALIVE",
+    "174": "NON_PARTICIPATING",
+    "175": "NON_PARTICIPATING",
     "176": "NOT_ALIVE",
-    "177": "NOT_ALIVE",
-    "178": "NOT_ALIVE",
-    "179": "NOT_ALIVE",
+    "177": "NON_PARTICIPATING",
+    "178": "NON_PARTICIPATING",
+    "179": "NON_PARTICIPATING",
     "18": "NON_PARTICIPATING",
-    "180": "NOT_ALIVE",
+    "180": "NON_PARTICIPATING",
     "181": "NOT_ALIVE",
     "182": "NOT_ALIVE",
     "183": "NON_PARTICIPATING",
-    "184": "NOT_ALIVE",
+    "184": "NON_PARTICIPATING",
     "185": "NON_PARTICIPATING",
-    "186": "NOT_ALIVE",
-    "187": "NOT_ALIVE",
+    "186": "NON_PARTICIPATING",
+    "187": "NON_PARTICIPATING",
     "188": "NON_PARTICIPATING",
-    "189": "NOT_ALIVE",
-    "19": "NOT_ALIVE",
+    "189": "NON_PARTICIPATING",
+    "19": "NON_PARTICIPATING",
     "190": "NON_PARTICIPATING",
-    "191": "NOT_ALIVE",
-    "192": "NOT_ALIVE",
-    "193": "NON_PARTICIPATING",
+    "191": "NON_PARTICIPATING",
+    "192": "NON_PARTICIPATING",
+    "193": "NOT_ALIVE",
     "194": "NON_PARTICIPATING",
     "195": "NOT_ALIVE",
     "196": "NON_PARTICIPATING",
@@ -115,7 +115,7 @@
     "199": "NON_PARTICIPATING",
     "2": "NON_PARTICIPATING",
     "20": "NOT_ALIVE",
-    "200": "NOT_ALIVE",
+    "200": "NON_PARTICIPATING",
     "201": "NOT_ALIVE",
     "202": "NON_PARTICIPATING",
     "203": "NOT_ALIVE",
@@ -123,37 +123,37 @@
     "205": "NOT_ALIVE",
     "206": "NOT_ALIVE",
     "207": "NOT_ALIVE",
-    "208": "NOT_ALIVE",
+    "208": "NON_PARTICIPATING",
     "209": "NOT_ALIVE",
     "21": "NOT_ALIVE",
     "210": "NOT_ALIVE",
     "211": "NON_PARTICIPATING",
     "212": "NON_PARTICIPATING",
-    "213": "NOT_ALIVE",
-    "214": "NOT_ALIVE",
+    "213": "NON_PARTICIPATING",
+    "214": "NON_PARTICIPATING",
     "215": "NON_PARTICIPATING",
     "216": "NON_PARTICIPATING",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
     "219": "NOT_ALIVE",
     "22": "NON_PARTICIPATING",
-    "220": "NOT_ALIVE",
-    "221": "NOT_ALIVE",
+    "220": "NON_PARTICIPATING",
+    "221": "NON_PARTICIPATING",
     "222": "NON_PARTICIPATING",
     "223": "NON_PARTICIPATING",
     "224": "NOT_ALIVE",
     "225": "NON_PARTICIPATING",
-    "226": "NOT_ALIVE",
-    "227": "NOT_ALIVE",
+    "226": "NON_PARTICIPATING",
+    "227": "NON_PARTICIPATING",
     "228": "NON_PARTICIPATING",
-    "229": "NOT_ALIVE",
-    "23": "NOT_ALIVE",
+    "229": "NON_PARTICIPATING",
+    "23": "NON_PARTICIPATING",
     "230": "NOT_ALIVE",
     "231": "NOT_ALIVE",
     "232": "NOT_ALIVE",
     "233": "NON_PARTICIPATING",
     "234": "NOT_ALIVE",
-    "235": "NOT_ALIVE",
+    "235": "NON_PARTICIPATING",
     "236": "NON_PARTICIPATING",
     "237": "NON_PARTICIPATING",
     "238": "NON_PARTICIPATING",
@@ -161,14 +161,14 @@
     "24": "NOT_ALIVE",
     "240": "NON_PARTICIPATING",
     "241": "NOT_ALIVE",
-    "242": "NOT_ALIVE",
+    "242": "NON_PARTICIPATING",
     "243": "NON_PARTICIPATING",
     "244": "NON_PARTICIPATING",
-    "245": "NOT_ALIVE",
+    "245": "NON_PARTICIPATING",
     "246": "NOT_ALIVE",
     "247": "NON_PARTICIPATING",
     "248": "NOT_ALIVE",
-    "249": "NOT_ALIVE",
+    "249": "NON_PARTICIPATING",
     "25": "SUCCESS",
     "250": "NOT_ALIVE",
     "251": "NOT_ALIVE",
@@ -176,29 +176,29 @@
     "253": "NON_PARTICIPATING",
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
-    "26": "NOT_ALIVE",
+    "26": "NON_PARTICIPATING",
     "27": "NOT_ALIVE",
-    "28": "NOT_ALIVE",
+    "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
-    "3": "NOT_ALIVE",
-    "30": "NON_PARTICIPATING",
+    "3": "NON_PARTICIPATING",
+    "30": "NOT_ALIVE",
     "31": "NON_PARTICIPATING",
     "32": "NON_PARTICIPATING",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
-    "35": "NOT_ALIVE",
+    "35": "NON_PARTICIPATING",
     "36": "NOT_ALIVE",
-    "37": "NOT_ALIVE",
+    "37": "NON_PARTICIPATING",
     "38": "NOT_ALIVE",
     "39": "NON_PARTICIPATING",
-    "4": "NOT_ALIVE",
-    "40": "NOT_ALIVE",
-    "41": "NOT_ALIVE",
+    "4": "NON_PARTICIPATING",
+    "40": "NON_PARTICIPATING",
+    "41": "NON_PARTICIPATING",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
     "44": "NON_PARTICIPATING",
-    "45": "NOT_ALIVE",
-    "46": "NOT_ALIVE",
+    "45": "NON_PARTICIPATING",
+    "46": "NON_PARTICIPATING",
     "47": "NON_PARTICIPATING",
     "48": "NOT_ALIVE",
     "49": "NOT_ALIVE",
@@ -206,47 +206,47 @@
     "50": "NON_PARTICIPATING",
     "51": "NON_PARTICIPATING",
     "52": "NON_PARTICIPATING",
-    "53": "NOT_ALIVE",
+    "53": "NON_PARTICIPATING",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
     "57": "NOT_ALIVE",
     "58": "NON_PARTICIPATING",
-    "59": "NON_PARTICIPATING",
-    "6": "NOT_ALIVE",
+    "59": "NOT_ALIVE",
+    "6": "NON_PARTICIPATING",
     "60": "NOT_ALIVE",
-    "61": "NOT_ALIVE",
-    "62": "NOT_ALIVE",
+    "61": "NON_PARTICIPATING",
+    "62": "NON_PARTICIPATING",
     "63": "NON_PARTICIPATING",
     "64": "NOT_ALIVE",
     "65": "NON_PARTICIPATING",
     "66": "NON_PARTICIPATING",
     "67": "NOT_ALIVE",
-    "68": "NOT_ALIVE",
+    "68": "NON_PARTICIPATING",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "NON_PARTICIPATING",
     "71": "NON_PARTICIPATING",
-    "72": "NOT_ALIVE",
-    "73": "NOT_ALIVE",
+    "72": "NON_PARTICIPATING",
+    "73": "NON_PARTICIPATING",
     "74": "NOT_ALIVE",
-    "75": "NON_PARTICIPATING",
+    "75": "NOT_ALIVE",
     "76": "NOT_ALIVE",
     "77": "NON_PARTICIPATING",
     "78": "NON_PARTICIPATING",
     "79": "NON_PARTICIPATING",
-    "8": "NON_PARTICIPATING",
+    "8": "NOT_ALIVE",
     "80": "NON_PARTICIPATING",
     "81": "NOT_ALIVE",
-    "82": "NOT_ALIVE",
+    "82": "NON_PARTICIPATING",
     "83": "NOT_ALIVE",
-    "84": "NOT_ALIVE",
+    "84": "NON_PARTICIPATING",
     "85": "NOT_ALIVE",
     "86": "NON_PARTICIPATING",
     "87": "NOT_ALIVE",
     "88": "NOT_ALIVE",
     "89": "NOT_ALIVE",
-    "9": "NOT_ALIVE",
+    "9": "NON_PARTICIPATING",
     "90": "NOT_ALIVE",
     "91": "NON_PARTICIPATING",
     "92": "NOT_ALIVE",
@@ -256,7 +256,7 @@
     "96": "NOT_ALIVE",
     "97": "NON_PARTICIPATING",
     "98": "NOT_ALIVE",
-    "99": "NOT_ALIVE"
+    "99": "NON_PARTICIPATING"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -271,7 +271,7 @@
   "initializer_range": 0.02,
   "inner_step": 0,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6059469,
+  "last_allreduce_block": 6062756,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e874fc0fd66f2ada48c03624740df94db0f83e00b16569d4c9b4bd03ea354b1
+oid sha256:344ba423ea26e8f824477bcb0c2e2734bf6a7a42c88b80fc2d50490bb7050ccb
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b906ac7dd1c48143caec8a1b5e4d057b561f68eb6433fb9c7c4e66e16763419
+oid sha256:92005eac2a616dc83df73484eb2039f4f1acc949f17f4663cae21557d479854c
 size 4400216536

outer_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fafd7ad8e66630a098d752f47a55cbd7c3353497ba358183d703f3331e1898a9
+oid sha256:835db1309cfaa53e6a71a506938af7aaf0ae92616f44f8f9a70326bf5057cbb3
 size 4400252568