Run 6. Outer Step 38. Inner Step 10.

config.json

@@ -2,7 +2,7 @@
   "_name_or_path": "distributed/llama-1b",
   "all_reduce_scores": {
     "0": "NOT_ALIVE",
-    "1": "FAIL",
+    "1": "SUCCESS",
     "10": "NOT_ALIVE",
     "100": "NOT_ALIVE",
     "101": "NOT_ALIVE",
@@ -12,7 +12,7 @@
     "105": "NOT_ALIVE",
     "106": "NOT_ALIVE",
     "107": "NOT_ALIVE",
-    "108": "NON_PARTICIPATING",
+    "108": "NOT_ALIVE",
     "109": "NOT_ALIVE",
     "11": "NOT_ALIVE",
     "110": "NOT_ALIVE",
@@ -40,21 +40,21 @@
     "130": "NOT_ALIVE",
     "131": "NOT_ALIVE",
     "132": "NOT_ALIVE",
-    "133": "FAIL",
+    "133": "NOT_ALIVE",
     "134": "NOT_ALIVE",
     "135": "NOT_ALIVE",
     "136": "NOT_ALIVE",
     "137": "NOT_ALIVE",
-    "138": "NON_PARTICIPATING",
+    "138": "SUCCESS",
     "139": "NOT_ALIVE",
     "14": "NOT_ALIVE",
-    "140": "SUCCESS",
+    "140": "FAIL",
     "141": "NOT_ALIVE",
     "142": "NOT_ALIVE",
     "143": "NOT_ALIVE",
     "144": "SUCCESS",
     "145": "NOT_ALIVE",
-    "146": "FAIL",
+    "146": "NOT_ALIVE",
     "147": "NOT_ALIVE",
     "148": "NOT_ALIVE",
     "149": "NOT_ALIVE",
@@ -67,22 +67,22 @@
     "155": "NOT_ALIVE",
     "156": "NOT_ALIVE",
     "157": "NOT_ALIVE",
-    "158": "NON_PARTICIPATING",
+    "158": "NOT_ALIVE",
     "159": "NOT_ALIVE",
     "16": "NOT_ALIVE",
     "160": "SUCCESS",
     "161": "NOT_ALIVE",
     "162": "NOT_ALIVE",
-    "163": "NOT_ALIVE",
+    "163": "NON_PARTICIPATING",
     "164": "SUCCESS",
-    "165": "NON_PARTICIPATING",
-    "166": "NOT_ALIVE",
+    "165": "SUCCESS",
+    "166": "SUCCESS",
     "167": "NOT_ALIVE",
     "168": "NOT_ALIVE",
     "169": "NON_PARTICIPATING",
     "17": "NOT_ALIVE",
     "170": "NOT_ALIVE",
-    "171": "NON_PARTICIPATING",
+    "171": "NOT_ALIVE",
     "172": "NOT_ALIVE",
     "173": "NOT_ALIVE",
     "174": "NOT_ALIVE",
@@ -103,8 +103,8 @@
     "188": "NOT_ALIVE",
     "189": "SUCCESS",
     "19": "NOT_ALIVE",
-    "190": "SUCCESS",
-    "191": "NOT_ALIVE",
+    "190": "NON_PARTICIPATING",
+    "191": "NON_PARTICIPATING",
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
@@ -112,11 +112,11 @@
     "196": "NOT_ALIVE",
     "197": "NOT_ALIVE",
     "198": "NOT_ALIVE",
-    "199": "SUCCESS",
-    "2": "SUCCESS",
+    "199": "NOT_ALIVE",
+    "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
-    "201": "NOT_ALIVE",
+    "201": "SUCCESS",
     "202": "NOT_ALIVE",
     "203": "NOT_ALIVE",
     "204": "NOT_ALIVE",
@@ -131,7 +131,7 @@
     "212": "SUCCESS",
     "213": "NOT_ALIVE",
     "214": "NON_PARTICIPATING",
-    "215": "NON_PARTICIPATING",
+    "215": "NOT_ALIVE",
     "216": "NOT_ALIVE",
     "217": "NOT_ALIVE",
     "218": "NON_PARTICIPATING",
@@ -141,9 +141,9 @@
     "221": "NOT_ALIVE",
     "222": "NOT_ALIVE",
     "223": "NOT_ALIVE",
-    "224": "NON_PARTICIPATING",
+    "224": "NOT_ALIVE",
     "225": "NOT_ALIVE",
-    "226": "NOT_ALIVE",
+    "226": "NON_PARTICIPATING",
     "227": "NOT_ALIVE",
     "228": "SUCCESS",
     "229": "NOT_ALIVE",
@@ -169,7 +169,7 @@
     "247": "NOT_ALIVE",
     "248": "NOT_ALIVE",
     "249": "NOT_ALIVE",
-    "25": "SUCCESS",
+    "25": "NOT_ALIVE",
     "250": "NOT_ALIVE",
     "251": "NOT_ALIVE",
     "252": "NOT_ALIVE",
@@ -177,12 +177,12 @@
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
+    "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
     "30": "NOT_ALIVE",
-    "31": "NOT_ALIVE",
+    "31": "SUCCESS",
     "32": "SUCCESS",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
@@ -192,7 +192,7 @@
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "NOT_ALIVE",
+    "40": "SUCCESS",
     "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
@@ -209,7 +209,7 @@
     "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NOT_ALIVE",
+    "56": "NON_PARTICIPATING",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
@@ -219,13 +219,13 @@
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
-    "70": "FAIL",
+    "70": "SUCCESS",
     "71": "SUCCESS",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
@@ -239,7 +239,7 @@
     "80": "NOT_ALIVE",
     "81": "NOT_ALIVE",
     "82": "NOT_ALIVE",
-    "83": "SUCCESS",
+    "83": "NOT_ALIVE",
     "84": "NOT_ALIVE",
     "85": "NOT_ALIVE",
     "86": "NOT_ALIVE",
@@ -256,7 +256,7 @@
     "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "NOT_ALIVE",
-    "99": "FAIL"
+    "99": "SUCCESS"
   },
   "architectures": [
     "LlamaForCausalLM"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6154364,
-    6154407
+    6154584,
+    6154624
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 82,
+  "inner_step": 9,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6153736,
+  "last_allreduce_block": 6154397,
   "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:e091897c817a3795d8e190d41c4bf7772ba9a5fe1cca496d5969bd01035da568
+oid sha256:9056e4b33f0896dcbbf89bf8d89fb7faeea5fa7362c8b529056508bede7411f6
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37005871dcb226036b1de3489ed249baf610b88051ae86758e6c631d1f0bb8f8
+oid sha256:aa96cdb568416f885eb46a8da5673f88b0423fd27ba9e892751354037b305bda
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ca12b25174f366a055dbe9bebb8b31f9d89c5f11411b80a4ecc2c02f4361675
+oid sha256:4930efca1233f3f9b617669bf7b4e1ca35085a554e016ae05cb15f60fefedb99
 size 4400216536