diff --git a/.gitattributes b/.gitattributes index 92f49f902f0886a316eed817dd4a085dbd4bd36e..0bef38fa853db23966105e519d59ffc79c311736 100644 --- a/.gitattributes +++ b/.gitattributes @@ -61,3 +61,4 @@ zain/Activation/wandb/run-20260814_211407-i85cngas/run-i85cngas.wandb filter=lfs zain/Activation/wandb/run-20260814_211507-54wnyexz/run-54wnyexz.wandb filter=lfs diff=lfs merge=lfs -text zain/Activation/wandb/run-20260814_212208-n6fjpg9o/run-n6fjpg9o.wandb filter=lfs diff=lfs merge=lfs -text zain/Activation/out/glu-waleed10-100L_trash_run/training_log.jsonl filter=lfs diff=lfs merge=lfs -text +zain/Activation/wandb/run-20260814_212835-gqskne0x/run-gqskne0x.wandb filter=lfs diff=lfs merge=lfs -text diff --git a/zain/Activation/wandb/run-20260814_195151-ix6wcbwv/logs/debug-core.log b/zain/Activation/wandb/run-20260814_195151-ix6wcbwv/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_195151-ix6wcbwv/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_195151-ix6wcbwv/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_200425-npumf6lz/logs/debug-core.log b/zain/Activation/wandb/run-20260814_200425-npumf6lz/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_200425-npumf6lz/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_200425-npumf6lz/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_200503-bdhnr27j/logs/debug-core.log b/zain/Activation/wandb/run-20260814_200503-bdhnr27j/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_200503-bdhnr27j/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_200503-bdhnr27j/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_201615-w8mr4w6k/logs/debug-core.log b/zain/Activation/wandb/run-20260814_201615-w8mr4w6k/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_201615-w8mr4w6k/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_201615-w8mr4w6k/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_202128-twgo1e89/logs/debug-core.log b/zain/Activation/wandb/run-20260814_202128-twgo1e89/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_202128-twgo1e89/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_202128-twgo1e89/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_202806-g2ch84q3/logs/debug-core.log b/zain/Activation/wandb/run-20260814_202806-g2ch84q3/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_202806-g2ch84q3/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_202806-g2ch84q3/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_203803-7p5140bq/logs/debug-core.log b/zain/Activation/wandb/run-20260814_203803-7p5140bq/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_203803-7p5140bq/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_203803-7p5140bq/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_203945-j86czemo/logs/debug-core.log b/zain/Activation/wandb/run-20260814_203945-j86czemo/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_203945-j86czemo/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_203945-j86czemo/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_205135-0vnbmyac/logs/debug-core.log b/zain/Activation/wandb/run-20260814_205135-0vnbmyac/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_205135-0vnbmyac/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_205135-0vnbmyac/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_210109-qauftiuh/logs/debug-core.log b/zain/Activation/wandb/run-20260814_210109-qauftiuh/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_210109-qauftiuh/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_210109-qauftiuh/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_210155-rxpj34pj/logs/debug-core.log b/zain/Activation/wandb/run-20260814_210155-rxpj34pj/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_210155-rxpj34pj/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_210155-rxpj34pj/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_210809-a6574zx8/logs/debug-core.log b/zain/Activation/wandb/run-20260814_210809-a6574zx8/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_210809-a6574zx8/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_210809-a6574zx8/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/files/config.yaml b/zain/Activation/wandb/run-20260814_211407-i85cngas/files/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..87a5f7e814d50e038ab07bf4c7ce2cb4cd5ccc45 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/files/config.yaml @@ -0,0 +1,439 @@ +_name_or_path: + value: "" +_wandb: + value: + cli_version: 0.28.1 + e: + unk3a3lapicd752rmq1uqlvx1kx26i7e: + args: + - --config + - /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline.yaml + - --variants + - glu-waleed10 + - glu-silu-waleed10 + - mlp-waleed10 + - mlp-silu-waleed10 + - glu-waleed + - glu-situglu_low + - glu-waleedglu_low + codePath: sweep.py + codePathLocal: sweep.py + cpu_count: 112 + cpu_count_logical: 224 + cudaVersion: "12.4" + disk: + /: + total: "1560765693952" + used: "701346242560" + email: deepnevro@gmail.com + executable: /mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python + git: + commit: 3fb47c6943d6927eafc39bd399a3d3a520bae74a + remote: https://github.com/w-ahmad1a10/Activation.git + gpu: NVIDIA H100 80GB HBM3 + gpu_count: 8 + gpu_nvidia: + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-39c684a5-fde6-83d7-1663-0859795881ae + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-132944c4-b689-2b5f-89a4-d730401677ab + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bc6c3e3c-9b90-09ca-c034-774961847c54 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea + host: deeplens-k3s-node1 + memory: + total: "2164089937920" + os: Linux-5.15.0-126-generic-x86_64-with-glibc2.35 + program: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py + python: CPython 3.11.15 + root: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation + startedAt: "2026-08-14T21:14:07.908466Z" + writerId: unk3a3lapicd752rmq1uqlvx1kx26i7e + m: + - "1": train/global_step + "6": + - 3 + "7": [] + - "2": '*' + "5": 1 + "6": + - 1 + "7": [] + python_version: 3.11.15 + t: + "1": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "2": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "3": + - 2 + - 7 + - 13 + - 19 + - 62 + - 66 + "4": 3.11.15 + "5": 0.28.1 + "6": 5.16.0.dev0 + "9": + "1": transformers_trainer + "12": 0.28.1 + "13": linux-x86_64 +accelerator_config: + value: + dispatch_batches: null + even_batches: true + gradient_accumulation_kwargs: null + non_blocking: false + split_batches: false + use_seedable_sampler: true +activation: + value: waleedglu_low +adam_beta1: + value: 0.9 +adam_beta2: + value: 0.999 +adam_epsilon: + value: 1e-08 +architectures: + value: null +attention_bias: + value: false +attention_dropout: + value: 0 +auto_find_batch_size: + value: false +average_tokens_across_devices: + value: true +batch_eval_metrics: + value: false +bf16: + value: true +bf16_full_eval: + value: false +bos_token_id: + value: 1 +chunk_size_feed_forward: + value: 0 +data_seed: + value: 42 +dataloader_drop_last: + value: false +dataloader_in_order: + value: true +dataloader_multiprocessing_context: + value: null +dataloader_num_workers: + value: 0 +dataloader_persistent_workers: + value: false +dataloader_pin_memory: + value: true +dataloader_prefetch_factor: + value: null +ddp_backend: + value: null +ddp_broadcast_buffers: + value: null +ddp_bucket_cap_mb: + value: null +ddp_find_unused_parameters: + value: null +ddp_static_graph: + value: null +ddp_timeout: + value: 1800 +debug: + value: [] +deepspeed: + value: null +disable_tqdm: + value: false +do_eval: + value: true +do_predict: + value: false +do_train: + value: false +dtype: + value: null +enable_jit_checkpoint: + value: false +eos_token_id: + value: 2 +eval_accumulation_steps: + value: null +eval_delay: + value: 0 +eval_do_concat_batches: + value: true +eval_on_start: + value: false +eval_steps: + value: 60000 +eval_strategy: + value: steps +eval_use_gather_object: + value: false +fp16: + value: false +fp16_full_eval: + value: false +fsdp: + value: null +fsdp_config: + value: null +full_determinism: + value: false +gradient_accumulation_steps: + value: 1 +gradient_checkpointing: + value: false +gradient_checkpointing_kwargs: + value: null +greater_is_better: + value: null +head_dim: + value: 32 +hidden_act: + value: silu +hidden_size: + value: 128 +hub_always_push: + value: false +hub_model_id: + value: w-ahmad/A-glu-waleedglu_low-9L +hub_private_repo: + value: null +hub_revision: + value: null +hub_strategy: + value: every_save +hub_token: + value: +id2label: + value: + "0": LABEL_0 + "1": LABEL_1 +ignore_data_skip: + value: false +include_for_metrics: + value: [] +include_num_input_tokens_seen: + value: "no" +initializer_range: + value: 0.02 +intermediate_size: + value: 256 +is_encoder_decoder: + value: false +label_names: + value: null +label_smoothing_factor: + value: 0 +label2id: + value: + LABEL_0: 0 + LABEL_1: 1 +learning_rate: + value: 0.001 +length_column_name: + value: length +liger_kernel_config: + value: null +load_best_model_at_end: + value: false +local_rank: + value: -1 +log_level: + value: passive +log_level_replica: + value: warning +log_on_each_node: + value: true +logging_first_step: + value: false +logging_nan_inf_filter: + value: true +logging_steps: + value: 20 +logging_strategy: + value: steps +lr_scheduler_kwargs: + value: null +lr_scheduler_type: + value: constant +max_grad_norm: + value: 1 +max_position_embeddings: + value: 512 +max_steps: + value: 1000 +metric_for_best_model: + value: null +mlp_bias: + value: false +mlp_type: + value: glu +model/num_parameters: + value: 2001280 +model_type: + value: tiny_llama +neftune_noise_alpha: + value: null +num_attention_heads: + value: 4 +num_hidden_layers: + value: 9 +num_key_value_heads: + value: 4 +num_train_epochs: + value: 1 +optim: + value: adamw_torch +optim_args: + value: null +optim_target_modules: + value: null +output_attentions: + value: false +output_dir: + value: out/glu-waleedglu_low-9L_run +output_hidden_states: + value: false +pad_token_id: + value: 0 +parallelism_config: + value: null +per_device_eval_batch_size: + value: 512 +per_device_train_batch_size: + value: 512 +prediction_loss_only: + value: false +pretraining_tp: + value: 1 +problem_type: + value: null +project: + value: huggingface +push_to_hub: + value: false +remove_unused_columns: + value: false +report_to: + value: + - wandb +restore_callback_states_from_checkpoint: + value: false +resume_from_checkpoint: + value: null +return_dict: + value: true +rms_norm_eps: + value: 1e-06 +rope_parameters: + value: + rope_theta: 10000 + rope_type: default +run_name: + value: LM-glu-waleedglu_low-9L-2.0M-20260814-211407 +save_on_each_node: + value: false +save_only_model: + value: false +save_steps: + value: 100 +save_strategy: + value: steps +save_total_limit: + value: null +seed: + value: 42 +skip_memory_metrics: + value: true +tf32: + value: null +tie_word_embeddings: + value: true +tokenizer_name: + value: w-ahmad/tiny-stories-tokenizer +torch_compile: + value: false +torch_compile_backend: + value: null +torch_compile_mode: + value: null +torch_empty_cache_steps: + value: null +trackio_bucket_id: + value: null +trackio_space_id: + value: null +trackio_static_space_id: + value: null +train_sampling_strategy: + value: random +transformers_version: + value: 5.16.0.dev0 +use_cache: + value: false +use_cpu: + value: false +use_liger_kernel: + value: false +vocab_size: + value: 4096 +waleed_beta: + value: 10 +warmup_steps: + value: 0 +weight_decay: + value: 0.01 diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/files/output.log b/zain/Activation/wandb/run-20260814_211407-i85cngas/files/output.log index 8752a24fd58d388fdafa347f6fa41ad6d0733749..b2122bd2a98ca2fcaaaa522b40de41dd80466f89 100644 --- a/zain/Activation/wandb/run-20260814_211407-i85cngas/files/output.log +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/files/output.log @@ -88,6 +88,23 @@ Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 155 - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). - If you are not the owner of the model architecture class, please contact the model code owner to update it. Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 156.87it/s] - 95%|█████████▍| 949/1000 [10:54<00:34, 1.47it/s]?, ?it/s] +100%|██████████| 1000/1000 [11:40<00:00, 1.33i[transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. {'loss': '2.268', 'grad_norm': '0.9023', 'learning_rate': '0.001', 'epoch': '0.496', 'train/total_time_seconds': '284.7', 'train/time_per_step_avg': '0.3259', 'train/epoch_time_elapsed': '634.1', 'train/estimated_remaining_minutes': '0.4126'} {'loss': '2.264', 'grad_norm': '0.6562', 'learning_rate': '0.001', 'epoch': '0.5067', 'train/total_time_seconds': '291.5', 'train/time_per_step_avg': '0.3278', 'train/epoch_time_elapsed': '648.6', 'train/estimated_remaining_minutes': '0.3102'} +{'loss': '2.258', 'grad_norm': '0.6406', 'learning_rate': '0.001', 'epoch': '0.5175', 'train/total_time_seconds': '298', 'train/time_per_step_avg': '0.3264', 'train/epoch_time_elapsed': '662.8', 'train/estimated_remaining_minutes': '0.207'} +{'loss': '2.251', 'grad_norm': '0.6484', 'learning_rate': '0.001', 'epoch': '0.5283', 'train/total_time_seconds': '304.5', 'train/time_per_step_avg': '0.3226', 'train/epoch_time_elapsed': '676.9', 'train/estimated_remaining_minutes': '0.1036'} +{'loss': '2.245', 'grad_norm': '0.5703', 'learning_rate': '0.001', 'epoch': '0.5391', 'train/total_time_seconds': '311.1', 'train/time_per_step_avg': '0.3265', 'train/epoch_time_elapsed': '691.1', 'train/estimated_remaining_minutes': '0'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes +{'eval_loss': '2.239', 'eval_runtime': '9.298', 'eval_samples_per_second': '1025', 'eval_steps_per_second': '2.044', 'epoch': '0.5391', 'train/total_time_seconds': '311.1', 'train/time_per_step_avg': '0.3265', 'train/epoch_time_elapsed': '700.4', 'train/estimated_remaining_minutes': '0'} + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 158.67it/s] +100%|██████████| 1000/1000 [11:40<00:00, 1.43it/s], ?it/s] +{'train_runtime': '701.3', 'train_samples_per_second': '730.1', 'train_steps_per_second': '1.426', 'train_loss': '2.97', 'epoch': '0.5391', 'train/total_time_seconds': '311.1', 'train/time_per_step_avg': '0.3265', 'train/epoch_time_elapsed': '700.4', 'train/estimated_remaining_minutes': '0'} +100%|██████████| 19/19 [00:08<00:00, 2.17it/s] +[transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 154.24it/s] +>>> FINISHED glu-waleedglu_low-9L successfully diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/files/wandb-summary.json b/zain/Activation/wandb/run-20260814_211407-i85cngas/files/wandb-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..08d24936d5aa60d4f6f0029ce082d561f2a6275c --- /dev/null +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/files/wandb-summary.json @@ -0,0 +1 @@ +{"total_flos":2.323107545088e+15,"eval/runtime":9.5433,"_timestamp":1.786742758641734e+09,"train_samples_per_second":730.125,"train_loss":2.969754108428955,"train/learning_rate":0.001,"train/epoch":0.5390835579514824,"_step":52,"_wandb":{"runtime":710},"train/grad_norm":0.5703125,"train/global_step":1000,"eval/steps_per_second":1.991,"train_steps_per_second":1.426,"eval/loss":2.23911452293396,"_runtime":710,"eval/samples_per_second":998.289,"train_runtime":701.2502,"train/loss":2.245043182373047} \ No newline at end of file diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-core.log b/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-core.log index 85f25ca46da3f66cefb4fda2eaf3022519621f49..cc8455fab0160c8e2ec5d6a753bd34b72beb1d2c 100644 --- a/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-core.log @@ -47,3 +47,16 @@ {"time":"2026-08-14T21:14:07.911109006Z","level":"INFO","msg":"handleInformInit: received","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:08.168219613Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"i85cngas","id":"1(@)"} {"time":"2026-08-14T21:14:13.672910923Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:58.658206835Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.189862536Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"d8p4y9v5n05m"} +{"time":"2026-08-14T21:25:59.191285843Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.192218621Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"i85cngas","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492428474Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492473057Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492482505Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492489625Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:25:59.492505848Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492527496Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492557564Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:25:59.492679293Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-409188-410559-1836801570/socket","Net":"unix"}} +{"time":"2026-08-14T21:25:59.492751296Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-internal.log b/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-internal.log index 316cacd6ac21a7e66bea46338c66d086bd4c1f25..56b9086a13991a0a93e4e418dc4bfb6c8f93c3b5 100644 --- a/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-internal.log +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug-internal.log @@ -93,3 +93,19 @@ {"time":"2026-08-14T21:24:38.835784304Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} {"time":"2026-08-14T21:24:53.676283225Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":45,"history_lines":1,"events_offset":83,"events_lines":2,"console_offset":82,"console_lines":10} {"time":"2026-08-14T21:24:53.807963308Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:08.67577254Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":46,"history_lines":1,"events_offset":85,"events_lines":2,"console_offset":90,"console_lines":1} +{"time":"2026-08-14T21:25:08.813221459Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:23.675748872Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":47,"history_lines":1,"events_offset":87,"events_lines":2,"console_offset":90,"console_lines":1} +{"time":"2026-08-14T21:25:23.862250657Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:38.675553902Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":48,"history_lines":1,"events_offset":89,"events_lines":2,"console_offset":90,"console_lines":1} +{"time":"2026-08-14T21:25:38.897653817Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:53.676490134Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":49,"history_lines":3,"events_offset":91,"events_lines":2,"console_offset":90,"console_lines":1} +{"time":"2026-08-14T21:25:53.925875438Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:59.040344747Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"} +{"time":"2026-08-14T21:25:59.040864542Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":52,"history_lines":1,"events_offset":93,"events_lines":1,"console_offset":92,"console_lines":18,"uploaded_len":3,"complete":true,"exit_code":0} +{"time":"2026-08-14T21:25:59.187205462Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:59.188550421Z","level":"INFO","msg":"handler: operation stats","stats":{}} +{"time":"2026-08-14T21:25:59.191308472Z","level":"INFO","msg":"stream: finishing up"} +{"time":"2026-08-14T21:25:59.191335194Z","level":"INFO","msg":"handler: closed"} +{"time":"2026-08-14T21:25:59.191433704Z","level":"INFO","msg":"sender: closed"} +{"time":"2026-08-14T21:25:59.191438728Z","level":"INFO","msg":"stream: all finished"} diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug.log b/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug.log index c9a4dd26582c67173b34b22d6930b818ee06c3ef..a30373aab68f4af4a38aee47f1396e820c91cda1 100644 --- a/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug.log +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/logs/debug.log @@ -17,3 +17,8 @@ config: {'_wandb': {}} 2026-08-14 21:14:08,672 INFO MainThread:409188 [wandb_run.py:_config_callback():1346] config_cb None None {'transformers_version': '5.16.0.dev0', 'architectures': None, 'output_hidden_states': False, 'return_dict': True, 'dtype': None, 'chunk_size_feed_forward': 0, 'is_encoder_decoder': False, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'problem_type': None, 'vocab_size': 4096, 'hidden_size': 128, 'intermediate_size': 256, 'num_hidden_layers': 9, 'num_attention_heads': 4, 'num_key_value_heads': 4, 'hidden_act': 'silu', 'max_position_embeddings': 512, 'initializer_range': 0.02, 'rms_norm_eps': 1e-06, 'use_cache': False, 'pad_token_id': 0, 'bos_token_id': 1, 'eos_token_id': 2, 'pretraining_tp': 1, 'tie_word_embeddings': True, 'rope_parameters': {'rope_theta': 10000.0, 'rope_type': 'default'}, 'attention_bias': False, 'attention_dropout': 0.0, 'mlp_bias': False, 'head_dim': 32, '_name_or_path': '', 'tokenizer_name': 'w-ahmad/tiny-stories-tokenizer', 'mlp_type': 'glu', 'activation': 'waleedglu_low', 'waleed_beta': 10.0, 'model_type': 'tiny_llama', 'output_attentions': False, 'output_dir': 'out/glu-waleedglu_low-9L_run', 'per_device_train_batch_size': 512, 'num_train_epochs': 1, 'max_steps': 1000, 'learning_rate': 0.001, 'lr_scheduler_type': 'constant', 'lr_scheduler_kwargs': None, 'warmup_steps': 0, 'optim': 'adamw_torch', 'optim_args': None, 'weight_decay': 0.01, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'optim_target_modules': None, 'gradient_accumulation_steps': 1, 'average_tokens_across_devices': True, 'max_grad_norm': 1.0, 'label_smoothing_factor': 0.0, 'bf16': True, 'fp16': False, 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': None, 'gradient_checkpointing': False, 'gradient_checkpointing_kwargs': None, 'torch_compile': False, 'torch_compile_backend': None, 'torch_compile_mode': None, 'use_liger_kernel': False, 'liger_kernel_config': None, 'neftune_noise_alpha': None, 'torch_empty_cache_steps': None, 'auto_find_batch_size': False, 'logging_strategy': 'steps', 'logging_steps': 20, 'logging_first_step': False, 'log_on_each_node': True, 'logging_nan_inf_filter': True, 'include_num_input_tokens_seen': 'no', 'log_level': 'passive', 'log_level_replica': 'warning', 'disable_tqdm': False, 'report_to': ['wandb'], 'run_name': 'LM-glu-waleedglu_low-9L-2.0M-20260814-211407', 'project': 'huggingface', 'trackio_space_id': None, 'trackio_bucket_id': None, 'trackio_static_space_id': None, 'eval_strategy': 'steps', 'eval_steps': 60000, 'eval_delay': 0, 'per_device_eval_batch_size': 512, 'prediction_loss_only': False, 'eval_on_start': False, 'eval_do_concat_batches': True, 'eval_use_gather_object': False, 'eval_accumulation_steps': None, 'include_for_metrics': [], 'batch_eval_metrics': False, 'save_only_model': False, 'save_strategy': 'steps', 'save_steps': 100, 'save_on_each_node': False, 'save_total_limit': None, 'enable_jit_checkpoint': False, 'push_to_hub': False, 'hub_token': '', 'hub_private_repo': None, 'hub_model_id': 'w-ahmad/A-glu-waleedglu_low-9L', 'hub_strategy': 'every_save', 'hub_always_push': False, 'hub_revision': None, 'load_best_model_at_end': False, 'metric_for_best_model': None, 'greater_is_better': None, 'ignore_data_skip': False, 'restore_callback_states_from_checkpoint': False, 'full_determinism': False, 'seed': 42, 'data_seed': 42, 'use_cpu': False, 'accelerator_config': {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}, 'parallelism_config': None, 'dataloader_drop_last': False, 'dataloader_num_workers': 0, 'dataloader_pin_memory': True, 'dataloader_persistent_workers': False, 'dataloader_prefetch_factor': None, 'dataloader_multiprocessing_context': None, 'dataloader_in_order': True, 'remove_unused_columns': False, 'label_names': None, 'train_sampling_strategy': 'random', 'length_column_name': 'length', 'ddp_find_unused_parameters': None, 'ddp_bucket_cap_mb': None, 'ddp_broadcast_buffers': None, 'ddp_static_graph': None, 'ddp_backend': None, 'ddp_timeout': 1800, 'fsdp': None, 'fsdp_config': None, 'deepspeed': None, 'debug': [], 'skip_memory_metrics': True, 'do_train': False, 'do_eval': True, 'do_predict': False, 'resume_from_checkpoint': None, 'local_rank': -1} 2026-08-14 21:14:08,674 INFO MainThread:409188 [wandb_config.py:__setitem__():155] [no run ID] config set model/num_parameters = 2001280 - > 2026-08-14 21:14:08,674 INFO MainThread:409188 [wandb_run.py:_config_callback():1346] config_cb model/num_parameters 2001280 None +2026-08-14 21:25:58,657 INFO MainThread:409188 [wandb_run.py:_finish():2383] finishing run deepnevro-deepnevro/research1/i85cngas +2026-08-14 21:25:58,657 INFO MainThread:409188 [wandb_run.py:_atexit_cleanup():2588] got exitcode: 0 +2026-08-14 21:25:58,658 INFO MainThread:409188 [wandb_run.py:_restore():2570] restore +2026-08-14 21:25:58,658 INFO MainThread:409188 [wandb_run.py:_restore():2576] restore done +2026-08-14 21:25:59,190 INFO MainThread:409188 [wandb_run.py:_footer_sync_info():3993] logging synced files diff --git a/zain/Activation/wandb/run-20260814_211407-i85cngas/run-i85cngas.wandb b/zain/Activation/wandb/run-20260814_211407-i85cngas/run-i85cngas.wandb index bab1d7085462de38dcf89a1327aad574d544092a..67eb671f504e21f3089b2909b1245bf10d0b9d61 100644 --- a/zain/Activation/wandb/run-20260814_211407-i85cngas/run-i85cngas.wandb +++ b/zain/Activation/wandb/run-20260814_211407-i85cngas/run-i85cngas.wandb @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b0ebd0512cdc3282aa723f19ce33fc677dc5c04ef4430505d970ad61518c4800 -size 458752 +oid sha256:c4205d484f8c5c9946f91a69fae62658e7117137decdea8fec4aa2fb4748d2f1 +size 525167 diff --git a/zain/Activation/wandb/run-20260814_211507-54wnyexz/logs/debug-core.log b/zain/Activation/wandb/run-20260814_211507-54wnyexz/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_211507-54wnyexz/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_211507-54wnyexz/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/config.yaml b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..42e7f72faec490b4631ca5048498e291861e6eb6 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/config.yaml @@ -0,0 +1,439 @@ +_name_or_path: + value: "" +_wandb: + value: + cli_version: 0.28.1 + e: + dwqr38wvfx2faq20nlxzk1aeem4hvapm: + args: + - --config + - /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline100l.yaml + - --variants + - glu-relu + - glu-gelu + - glu-sigmoid + - glu-waleed10 + - glu-silu-waleed10 + codePath: sweep.py + codePathLocal: sweep.py + cpu_count: 112 + cpu_count_logical: 224 + cudaVersion: "12.4" + disk: + /: + total: "1560765693952" + used: "706493382656" + email: deepnevro@gmail.com + executable: /mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python + git: + commit: 3fb47c6943d6927eafc39bd399a3d3a520bae74a + remote: https://github.com/w-ahmad1a10/Activation.git + gpu: NVIDIA H100 80GB HBM3 + gpu_count: 8 + gpu_nvidia: + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-39c684a5-fde6-83d7-1663-0859795881ae + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-132944c4-b689-2b5f-89a4-d730401677ab + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bc6c3e3c-9b90-09ca-c034-774961847c54 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea + host: deeplens-k3s-node1 + memory: + total: "2164089937920" + os: Linux-5.15.0-126-generic-x86_64-with-glibc2.35 + program: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py + python: CPython 3.11.15 + root: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation + startedAt: "2026-08-14T21:22:08.715735Z" + writerId: dwqr38wvfx2faq20nlxzk1aeem4hvapm + m: + - "1": train/global_step + "6": + - 3 + "7": [] + - "2": '*' + "5": 1 + "6": + - 1 + "7": [] + python_version: 3.11.15 + t: + "1": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "2": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "3": + - 2 + - 7 + - 13 + - 19 + - 41 + - 61 + - 62 + - 66 + "4": 3.11.15 + "5": 0.28.1 + "6": 5.16.0.dev0 + "9": + "1": transformers_trainer + "12": 0.28.1 + "13": linux-x86_64 +accelerator_config: + value: + dispatch_batches: null + even_batches: true + gradient_accumulation_kwargs: null + non_blocking: false + split_batches: false + use_seedable_sampler: true +activation: + value: waleed10 +adam_beta1: + value: 0.9 +adam_beta2: + value: 0.999 +adam_epsilon: + value: 1e-08 +architectures: + value: null +attention_bias: + value: false +attention_dropout: + value: 0 +auto_find_batch_size: + value: false +average_tokens_across_devices: + value: true +batch_eval_metrics: + value: false +bf16: + value: true +bf16_full_eval: + value: false +bos_token_id: + value: 1 +chunk_size_feed_forward: + value: 0 +data_seed: + value: 42 +dataloader_drop_last: + value: false +dataloader_in_order: + value: true +dataloader_multiprocessing_context: + value: null +dataloader_num_workers: + value: 0 +dataloader_persistent_workers: + value: false +dataloader_pin_memory: + value: true +dataloader_prefetch_factor: + value: null +ddp_backend: + value: null +ddp_broadcast_buffers: + value: null +ddp_bucket_cap_mb: + value: null +ddp_find_unused_parameters: + value: null +ddp_static_graph: + value: null +ddp_timeout: + value: 1800 +debug: + value: [] +deepspeed: + value: null +disable_tqdm: + value: false +do_eval: + value: true +do_predict: + value: false +do_train: + value: false +dtype: + value: null +enable_jit_checkpoint: + value: false +eos_token_id: + value: 2 +eval_accumulation_steps: + value: null +eval_delay: + value: 0 +eval_do_concat_batches: + value: true +eval_on_start: + value: false +eval_steps: + value: 2498 +eval_strategy: + value: steps +eval_use_gather_object: + value: false +fp16: + value: false +fp16_full_eval: + value: false +fsdp: + value: null +fsdp_config: + value: null +full_determinism: + value: false +gradient_accumulation_steps: + value: 1 +gradient_checkpointing: + value: false +gradient_checkpointing_kwargs: + value: null +greater_is_better: + value: null +head_dim: + value: 32 +hidden_act: + value: silu +hidden_size: + value: 128 +hub_always_push: + value: false +hub_model_id: + value: w-ahmad/6L-glu-waleed10-100L +hub_private_repo: + value: null +hub_revision: + value: null +hub_strategy: + value: every_save +hub_token: + value: +id2label: + value: + "0": LABEL_0 + "1": LABEL_1 +ignore_data_skip: + value: false +include_for_metrics: + value: [] +include_num_input_tokens_seen: + value: "no" +initializer_range: + value: 0.02 +intermediate_size: + value: 256 +is_encoder_decoder: + value: false +label_names: + value: null +label_smoothing_factor: + value: 0 +label2id: + value: + LABEL_0: 0 + LABEL_1: 1 +learning_rate: + value: 9e-05 +length_column_name: + value: length +liger_kernel_config: + value: null +load_best_model_at_end: + value: false +local_rank: + value: -1 +log_level: + value: passive +log_level_replica: + value: warning +log_on_each_node: + value: true +logging_first_step: + value: false +logging_nan_inf_filter: + value: true +logging_steps: + value: 20 +logging_strategy: + value: steps +lr_scheduler_kwargs: + value: null +lr_scheduler_type: + value: constant_with_warmup +max_grad_norm: + value: 1 +max_position_embeddings: + value: 512 +max_steps: + value: 3500 +metric_for_best_model: + value: null +mlp_bias: + value: false +mlp_type: + value: glu +model/num_parameters: + value: 16934016 +model_type: + value: tiny_llama +neftune_noise_alpha: + value: null +num_attention_heads: + value: 4 +num_hidden_layers: + value: 100 +num_key_value_heads: + value: 4 +num_train_epochs: + value: 1 +optim: + value: adamw_torch_fused +optim_args: + value: null +optim_target_modules: + value: null +output_attentions: + value: false +output_dir: + value: out/glu-waleed10-100L_trash_run +output_hidden_states: + value: false +pad_token_id: + value: 0 +parallelism_config: + value: null +per_device_eval_batch_size: + value: 1024 +per_device_train_batch_size: + value: 64 +prediction_loss_only: + value: false +pretraining_tp: + value: 1 +problem_type: + value: null +project: + value: huggingface +push_to_hub: + value: false +remove_unused_columns: + value: false +report_to: + value: + - wandb +restore_callback_states_from_checkpoint: + value: false +resume_from_checkpoint: + value: null +return_dict: + value: true +rms_norm_eps: + value: 1e-06 +rope_parameters: + value: + rope_theta: 10000 + rope_type: default +run_name: + value: LM-glu-waleed10-100L-16.9M-20260814-212207 +save_on_each_node: + value: false +save_only_model: + value: false +save_steps: + value: 50 +save_strategy: + value: steps +save_total_limit: + value: null +seed: + value: 42 +skip_memory_metrics: + value: true +tf32: + value: null +tie_word_embeddings: + value: true +tokenizer_name: + value: w-ahmad/tiny-stories-tokenizer +torch_compile: + value: false +torch_compile_backend: + value: null +torch_compile_mode: + value: null +torch_empty_cache_steps: + value: null +trackio_bucket_id: + value: null +trackio_space_id: + value: null +trackio_static_space_id: + value: null +train_sampling_strategy: + value: random +transformers_version: + value: 5.16.0.dev0 +use_cache: + value: false +use_cpu: + value: false +use_liger_kernel: + value: false +vocab_size: + value: 4096 +waleed_beta: + value: 4 +warmup_steps: + value: 50 +weight_decay: + value: 0 diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/output.log b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/output.log index 5a9e3f495e79065519c17dfbde0646afd60af879..9ed66e87f102702fd6f485744ef456ef8441b0c5 100644 --- a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/output.log +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/output.log @@ -39,6 +39,73 @@ Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 18. - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). - If you are not the owner of the model architecture class, please contact the model code owner to update it. Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 14.39it/s] - 84%|████████▍ | 2955/3500 [02:49<03:39, 2.48it/s], ?it/s] + 86%|████████▌ | 3000/3500 [03:05<02:55, 2.85it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. {'loss': '7.945', 'grad_norm': '2.851e+06', 'learning_rate': '9e-05', 'epoch': '0.1968', 'train/total_time_seconds': '133.2', 'train/time_per_step_avg': '0.3165', 'train/epoch_time_elapsed': '156.7', 'train/estimated_remaining_minutes': '0.4411', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.3', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001747', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.6', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001678', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3906', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3164', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3906', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.707', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4447', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.009949', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.539', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.156', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.719', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.156', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '11.88', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4448', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.006439', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.539', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.438', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.375', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.438', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '10.81', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '6.545e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '24.88', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '101', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.6769', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '1.109', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '101', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '99.89', 'train/tensor_act_model_layers_0_mlp/norm': '7920', 'train/tensor_act_model_layers_0_mlp/mean': '3.859', 'train/tensor_act_model_layers_0_mlp/std': '0.3281', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '1.078', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '2.922', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '7926', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '3.859', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.3418', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_lay {'loss': '7.946', 'grad_norm': '2.064e+06', 'learning_rate': '9e-05', 'epoch': '0.1982', 'train/total_time_seconds': '139.5', 'train/time_per_step_avg': '0.3138', 'train/epoch_time_elapsed': '164', 'train/estimated_remaining_minutes': '0.443'} +{'loss': '7.938', 'grad_norm': '1.499e+06', 'learning_rate': '9e-05', 'epoch': '0.1995', 'train/total_time_seconds': '146', 'train/time_per_step_avg': '0.3136', 'train/epoch_time_elapsed': '171.6', 'train/estimated_remaining_minutes': '0.4438', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.3', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001625', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.7', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001595', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3906', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3418', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3906', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.7324', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4585', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.01001', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.586', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.312', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.719', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.312', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.03', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4587', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.005768', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.586', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.75', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.469', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.75', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.22', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '7.202e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '28.38', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20.75', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '107', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.8107', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '1.93', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '107', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '105.1', 'train/tensor_act_model_layers_0_mlp/norm': '8099', 'train/tensor_act_model_layers_0_mlp/mean': '3.953', 'train/tensor_act_model_layers_0_mlp/std': '0.1328', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '1.789', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '2.211', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8104', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '3.953', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.1621', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.375', 'train/tensor_act_model_layer +{'loss': '7.943', 'grad_norm': '1.057e+06', 'learning_rate': '9e-05', 'epoch': '0.2009', 'train/total_time_seconds': '152', 'train/time_per_step_avg': '0.3155', 'train/epoch_time_elapsed': '178.7', 'train/estimated_remaining_minutes': '0.442'} +{'loss': '7.944', 'grad_norm': '8.069e+05', 'learning_rate': '9e-05', 'epoch': '0.2022', 'train/total_time_seconds': '158', 'train/time_per_step_avg': '0.3142', 'train/epoch_time_elapsed': '185.7', 'train/estimated_remaining_minutes': '0.4388'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 15.48it/s] + 89%|████████▊ | 3100/3500 [03:41<01:56, 3.44it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '7.943', 'grad_norm': '6.226e+05', 'learning_rate': '9e-05', 'epoch': '0.2036', 'train/total_time_seconds': '164.9', 'train/time_per_step_avg': '0.3162', 'train/epoch_time_elapsed': '193.9', 'train/estimated_remaining_minutes': '0.4367', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.5', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001663', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.7', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.00164', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3496', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3066', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3496', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6562', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4633', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.009521', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.602', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.531', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.75', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.531', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.28', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4635', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.004761', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.602', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.5', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.344', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.5', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '10.84', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '7.552e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '30.25', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '21', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '108', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.8997', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '2.891', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '108', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '105.1', 'train/tensor_act_model_layers_0_mlp/norm': '8153', 'train/tensor_act_model_layers_0_mlp/mean': '3.984', 'train/tensor_act_model_layers_0_mlp/std': '0.05884', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '2.469', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '1.531', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8159', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '3.984', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.1094', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_layers_ +{'loss': '7.948', 'grad_norm': '4.669e+05', 'learning_rate': '9e-05', 'epoch': '0.2049', 'train/total_time_seconds': '170.6', 'train/time_per_step_avg': '0.3105', 'train/epoch_time_elapsed': '200.7', 'train/estimated_remaining_minutes': '0.4302'} +{'loss': '7.942', 'grad_norm': '3.645e+05', 'learning_rate': '9e-05', 'epoch': '0.2063', 'train/total_time_seconds': '177.6', 'train/time_per_step_avg': '0.316', 'train/epoch_time_elapsed': '208.8', 'train/estimated_remaining_minutes': '0.4256', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.2', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001671', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.5', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.00161', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.0918', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3301', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3281', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3301', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6582', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4695', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.009033', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.531', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.844', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.531', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.38', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4694', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.007019', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.406', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.03', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '7.853e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '32', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '21.12', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '114.5', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9573', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '4.219', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '114.5', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '110.3', 'train/tensor_act_model_layers_0_mlp/norm': '8175', 'train/tensor_act_model_layers_0_mlp/mean': '3.984', 'train/tensor_act_model_layers_0_mlp/std': '0.02795', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.141', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.8594', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8180', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09619', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_la +{'loss': '7.942', 'grad_norm': '3.052e+05', 'learning_rate': '9e-05', 'epoch': '0.2076', 'train/total_time_seconds': '183', 'train/time_per_step_avg': '0.3101', 'train/epoch_time_elapsed': '215.2', 'train/estimated_remaining_minutes': '0.4158'} +{'loss': '7.946', 'grad_norm': '2.437e+05', 'learning_rate': '9e-05', 'epoch': '0.209', 'train/total_time_seconds': '188', 'train/time_per_step_avg': '0.3008', 'train/epoch_time_elapsed': '221.3', 'train/estimated_remaining_minutes': '0.4044'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 18.72it/s] + 91%|█████████▏| 3200/3500 [04:10<01:21, 3.69it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '7.948', 'grad_norm': '1.976e+05', 'learning_rate': '9e-05', 'epoch': '0.2103', 'train/total_time_seconds': '193.7', 'train/time_per_step_avg': '0.2887', 'train/epoch_time_elapsed': '228.3', 'train/estimated_remaining_minutes': '0.3933', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.4', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001686', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.6', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001694', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3262', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3262', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3203', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6465', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4683', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.008606', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.438', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.875', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.438', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.31', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4678', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.005646', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.656', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.344', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.656', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '7.929e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '32.5', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '21', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '116', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9792', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '5.5', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '116', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '110.5', 'train/tensor_act_model_layers_0_mlp/norm': '8181', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.01721', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.516', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.4844', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8187', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09375', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_layers_0_resi +{'loss': '7.943', 'grad_norm': '1.649e+05', 'learning_rate': '9e-05', 'epoch': '0.2117', 'train/total_time_seconds': '198.4', 'train/time_per_step_avg': '0.278', 'train/epoch_time_elapsed': '234', 'train/estimated_remaining_minutes': '0.3791'} +{'loss': '7.944', 'grad_norm': '1.505e+05', 'learning_rate': '9e-05', 'epoch': '0.213', 'train/total_time_seconds': '203.3', 'train/time_per_step_avg': '0.2569', 'train/epoch_time_elapsed': '240', 'train/estimated_remaining_minutes': '0.3645', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.3', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001633', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.5', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001633', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3418', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3418', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3301', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6719', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4703', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.009766', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.344', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.719', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.344', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.06', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4697', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.006012', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.406', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.03', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.064e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '33.25', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '21', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '116.5', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9919', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '6.625', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '116.5', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '109.9', 'train/tensor_act_model_layers_0_mlp/norm': '8186', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.01105', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.719', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.2812', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8191', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09277', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.344', 'train/tensor_act_model_layers +{'loss': '7.941', 'grad_norm': '1.249e+05', 'learning_rate': '9e-05', 'epoch': '0.2144', 'train/total_time_seconds': '207.7', 'train/time_per_step_avg': '0.2475', 'train/epoch_time_elapsed': '245.5', 'train/estimated_remaining_minutes': '0.3484'} +{'loss': '7.943', 'grad_norm': '1.162e+05', 'learning_rate': '9e-05', 'epoch': '0.2157', 'train/total_time_seconds': '212.2', 'train/time_per_step_avg': '0.2413', 'train/epoch_time_elapsed': '251', 'train/estimated_remaining_minutes': '0.3315'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 17.05it/s] + 94%|█████████▍| 3300/3500 [04:39<00:54, 3.69it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '7.946', 'grad_norm': '9.216e+04', 'learning_rate': '9e-05', 'epoch': '0.2171', 'train/total_time_seconds': '217', 'train/time_per_step_avg': '0.2331', 'train/epoch_time_elapsed': '257.1', 'train/estimated_remaining_minutes': '0.3146', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.5', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001595', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.8', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001617', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3418', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3418', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3281', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6699', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4696', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.009766', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.469', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.875', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.469', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.34', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4690', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.006653', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.594', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.5', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.594', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.09', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.094e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '33.5', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20.88', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '114.5', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9958', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '7.406', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '114.5', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '107.1', 'train/tensor_act_model_layers_0_mlp/norm': '8188', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.007935', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.812', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.1875', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8193', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09277', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_laye +{'loss': '7.944', 'grad_norm': '8.96e+04', 'learning_rate': '9e-05', 'epoch': '0.2184', 'train/total_time_seconds': '221.5', 'train/time_per_step_avg': '0.2309', 'train/epoch_time_elapsed': '262.9', 'train/estimated_remaining_minutes': '0.2962'} +{'loss': '7.945', 'grad_norm': '8.09e+04', 'learning_rate': '9e-05', 'epoch': '0.2198', 'train/total_time_seconds': '226.4', 'train/time_per_step_avg': '0.2313', 'train/epoch_time_elapsed': '269', 'train/estimated_remaining_minutes': '0.2778', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.5', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001648', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.0918', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.8', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.00164', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3262', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3242', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3262', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6504', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4682', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.00766', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.469', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-6.031', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.469', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.5', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4675', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.007874', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.688', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.344', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.688', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.03', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.103e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '33.75', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20.88', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '110.5', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9976', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '8', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '110.5', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '102.5', 'train/tensor_act_model_layers_0_mlp/norm': '8189', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.006287', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.859', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.1406', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8194', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_layers_0_r +{'loss': '7.945', 'grad_norm': '7.117e+04', 'learning_rate': '9e-05', 'epoch': '0.2211', 'train/total_time_seconds': '230.8', 'train/time_per_step_avg': '0.2309', 'train/epoch_time_elapsed': '274.5', 'train/estimated_remaining_minutes': '0.258'} +{'loss': '7.945', 'grad_norm': '6.605e+04', 'learning_rate': '9e-05', 'epoch': '0.2224', 'train/total_time_seconds': '235.3', 'train/time_per_step_avg': '0.2309', 'train/epoch_time_elapsed': '279.9', 'train/estimated_remaining_minutes': '0.2376'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 18.23it/s] + 97%|█████████▋| 3400/3500 [05:08<00:27, 3.68it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '7.941', 'grad_norm': '5.555e+04', 'learning_rate': '9e-05', 'epoch': '0.2238', 'train/total_time_seconds': '240.1', 'train/time_per_step_avg': '0.2306', 'train/epoch_time_elapsed': '286.1', 'train/estimated_remaining_minutes': '0.217', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.3', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001633', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.5', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001617', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.0918', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3223', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3223', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3203', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6426', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4695', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.008484', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.219', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.906', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.219', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.12', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4687', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.006805', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.344', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '10.97', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.162e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '34', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20.75', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '111.5', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9984', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '8.062', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '111.5', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '103.4', 'train/tensor_act_model_layers_0_mlp/norm': '8190', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.005157', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.859', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.1406', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8195', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_laye +{'loss': '7.943', 'grad_norm': '5.094e+04', 'learning_rate': '9e-05', 'epoch': '0.2251', 'train/total_time_seconds': '244.5', 'train/time_per_step_avg': '0.2306', 'train/epoch_time_elapsed': '291.5', 'train/estimated_remaining_minutes': '0.1952'} +{'loss': '7.945', 'grad_norm': '4.506e+04', 'learning_rate': '9e-05', 'epoch': '0.2265', 'train/total_time_seconds': '249.4', 'train/time_per_step_avg': '0.2299', 'train/epoch_time_elapsed': '297.5', 'train/estimated_remaining_minutes': '0.1732', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.1', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001587', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.4', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001564', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.0918', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3906', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3281', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3906', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.7188', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4705', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.008057', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.375', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.812', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.375', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.19', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4697', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.008301', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.562', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.19', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.235e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '34.5', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20.75', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '112', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9994', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '8.75', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '112', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '103.2', 'train/tensor_act_model_layers_0_mlp/norm': '8190', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.004181', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.906', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.09375', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8195', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.375', 'train/tensor_act_model_layer +{'loss': '7.945', 'grad_norm': '4.147e+04', 'learning_rate': '9e-05', 'epoch': '0.2278', 'train/total_time_seconds': '253.9', 'train/time_per_step_avg': '0.2303', 'train/epoch_time_elapsed': '303', 'train/estimated_remaining_minutes': '0.1502'} +{'loss': '7.947', 'grad_norm': '3.61e+04', 'learning_rate': '9e-05', 'epoch': '0.2292', 'train/total_time_seconds': '258.3', 'train/time_per_step_avg': '0.2302', 'train/epoch_time_elapsed': '308.4', 'train/estimated_remaining_minutes': '0.1266'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 15.84it/s] +100%|██████████| 3500/3500 [05:52<00:00, 3.66i[transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '7.943', 'grad_norm': '3.098e+04', 'learning_rate': '9e-05', 'epoch': '0.2305', 'train/total_time_seconds': '263.1', 'train/time_per_step_avg': '0.2301', 'train/epoch_time_elapsed': '314.6', 'train/estimated_remaining_minutes': '0.1026', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.3', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001648', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.6', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001602', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3125', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3125', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3105', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.623', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4703', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.008362', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.625', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.469', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.938', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.469', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.41', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4690', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.00705', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.812', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.812', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.44', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.271e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '34.5', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '20.88', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '112', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0.9999', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '9.375', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '112', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '102.6', 'train/tensor_act_model_layers_0_mlp/norm': '8191', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.003555', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.922', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.07812', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8196', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.312', 'train/tensor_act_model_layer +{'loss': '7.943', 'grad_norm': '2.598e+04', 'learning_rate': '9e-05', 'epoch': '0.2319', 'train/total_time_seconds': '267.6', 'train/time_per_step_avg': '0.2306', 'train/epoch_time_elapsed': '320.1', 'train/estimated_remaining_minutes': '0.07779'} +{'loss': '7.947', 'grad_norm': '2.125e+04', 'learning_rate': '9e-05', 'epoch': '0.2332', 'train/total_time_seconds': '272.5', 'train/time_per_step_avg': '0.2308', 'train/epoch_time_elapsed': '326.1', 'train/estimated_remaining_minutes': '0.0525', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '187.5', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001755', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.2754', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2715', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5469', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '188.8', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '0.001747', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.373', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3418', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.373', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.7148', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '4689', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '-0.007751', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '6.25', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-5.844', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '6.25', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '12.09', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '4675', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-0.008545', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '1.617', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-5.5', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '5.625', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '11.12', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '8.314e+04', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '34.75', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '21', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '112', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '1', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '8.188', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '112', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '103.8', 'train/tensor_act_model_layers_0_mlp/norm': '8191', 'train/tensor_act_model_layers_0_mlp/mean': '4', 'train/tensor_act_model_layers_0_mlp/std': '0.002655', 'train/tensor_act_model_layers_0_mlp/max_abs': '4', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '3.875', 'train/tensor_act_model_layers_0_mlp/max': '4', 'train/tensor_act_model_layers_0_mlp/range': '0.125', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '8196', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '4', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.09229', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '4.375', 'train/tensor_act_model_layers_0_residual_ +{'loss': '7.947', 'grad_norm': '1.933e+04', 'learning_rate': '9e-05', 'epoch': '0.2346', 'train/total_time_seconds': '277', 'train/time_per_step_avg': '0.2314', 'train/epoch_time_elapsed': '331.7', 'train/estimated_remaining_minutes': '0.02653'} +{'loss': '7.949', 'grad_norm': '1.389e+04', 'learning_rate': '9e-05', 'epoch': '0.2359', 'train/total_time_seconds': '281.5', 'train/time_per_step_avg': '0.232', 'train/epoch_time_elapsed': '337.2', 'train/estimated_remaining_minutes': '0'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes +{'eval_loss': '7.945', 'eval_runtime': '15.6', 'eval_samples_per_second': '610.9', 'eval_steps_per_second': '0.641', 'epoch': '0.2359', 'train/total_time_seconds': '281.5', 'train/time_per_step_avg': '0.232', 'train/epoch_time_elapsed': '352.8', 'train/estimated_remaining_minutes': '0'} + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 14.20it/s] +100%|██████████| 3500/3500 [05:53<00:00, 9.91it/s], ?it/s] +{'train_runtime': '353.9', 'train_samples_per_second': '632.9', 'train_steps_per_second': '9.889', 'train_loss': '2.171', 'epoch': '0.2359', 'train/total_time_seconds': '281.5', 'train/time_per_step_avg': '0.232', 'train/epoch_time_elapsed': '353', 'train/estimated_remaining_minutes': '0'} +100%|██████████| 10/10 [00:13<00:00, 1.40s/it] +[transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 15.65it/s] +>>> FINISHED glu-waleed10-100L successfully diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/wandb-summary.json b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/wandb-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..c5af3da3b6b83e8fd10fc31aa3a6784ed0e7417b --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/files/wandb-summary.json @@ -0,0 +1 @@ +{"train/tensor_act_model_layers_25_residual_post_attn/std":0.62109375,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/range":0.2578125,"train/tensor_act_model_layers_17_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_16_mlp_down_proj/max_abs":16,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_36_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_83_mlp_down_proj_weight/std":0.0419921875,"train/tensor_act_model_layers_6_residual_post_attn/std":0.2412109375,"train/tensor_act_model_layers_78_mlp_down_proj/norm":30751.373046875,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_8_mlp_up_proj_weight/mean":-0.00010347366333007812,"train/tensor_act_model_layers_9_residual_pre_attn/max_abs":37.5,"train/tensor_act_model_layers_22_mlp/min":4,"train/tensor_act_model_layers_18_mlp_down_proj/std":0.37109375,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_28_residual_pre_attn/max_abs":114,"train/tensor_act_model_layers_97_mlp_gate_proj/max_abs":4.5,"train/tensor_act_model_layers_61_residual_pre_attn/max":246,"train/tensor_param_model_layers_33_mlp_down_proj_weight/norm":6.3125,"train/tensor_act_model_layers_37_mlp_up_proj/range":7.546875,"train/tensor_act_model_layers_91_mlp/norm":8192,"train/tensor_act_model_layers_27_residual_post_mlp/norm":228930.234375,"train/tensor_act_model_layers_15_residual_post_mlp/max":65.5,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_62_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_46_mlp_up_proj/max":3.828125,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_26_mlp_up_proj/min":-3.625,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_39_mlp_up_proj/max_abs":3.59375,"train/tensor_param_model_layers_29_mlp_up_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_17_residual_pre_attn/mean":68,"train/tensor_act_model_layers_91_mlp_up_proj/max_abs":4.0625,"train/tensor_param_model_layers_77_mlp_down_proj_weight/mean":0.0004291534423828125,"train/tensor_act_model_layers_2_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_75_mlp_down_proj_weight/mean":0.0004634857177734375,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/max":0.12451171875,"train/tensor_act_model_layers_66_mlp_gate_proj/min":-3.34375,"train/tensor_act_model_layers_15_residual_post_mlp/max_abs":65.5,"train/tensor_act_model_layers_85_mlp_down_proj/std":0.51171875,"train/tensor_act_model_layers_66_residual_post_attn/norm":540270.375,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_17_mlp_up_proj/std":1.8125,"train/tensor_act_model_layers_63_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_50_mlp_down_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_58_mlp_gate_proj/max":3.5,"train/tensor_act_model_layers_93_mlp_up_proj/norm":5456.74072265625,"train/tensor_act_model_layers_60_residual_post_mlp/mean":244,"train/tensor_act_model_layers_0_mlp_up_proj/min":-5.5,"train/tensor_act_model_layers_12_residual_pre_attn/min":46.5,"train/tensor_param_model_layers_74_mlp_down_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_26_residual_post_attn/max_abs":106,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_18_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_residual_post_attn/range":6,"train/tensor_act_model_layers_77_mlp_down_proj/norm":30736.501953125,"train/tensor_act_model_layers_89_mlp_gate_proj/range":7.734375,"train/tensor_act_model_layers_33_residual_pre_attn/range":4,"train/tensor_param_model_layers_33_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/mean":-4.26173210144043e-06,"train/tensor_act_model_layers_65_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_28_residual_pre_attn/max":114,"train/tensor_act_model_layers_37_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_79_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_22_residual_post_attn/mean":88,"train/tensor_act_model_layers_14_mlp/max":4,"train/tensor_act_model_layers_60_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp_gate_proj/std":1.859375,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_84_residual_pre_attn/mean":336,"train/tensor_act_model_layers_9_mlp_down_proj/range":2.75,"train/tensor_act_model_layers_10_residual_pre_attn/min":38.5,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_77_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/std":0.033935546875,"train/tensor_act_model_layers_83_mlp_down_proj/range":2.4375,"train/tensor_act_model_layers_12_mlp_down_proj/std":0.30078125,"train/tensor_act_model_layers_82_mlp_gate_proj/max_abs":4.40625,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_30_mlp_down_proj_weight/min":-0.119140625,"train/tensor_act_model_layers_72_mlp_down_proj/max":17.125,"train/tensor_act_model_layers_18_residual_pre_attn/max":73.5,"train/tensor_act_model_layers_91_mlp_down_proj/std":0.546875,"train/tensor_act_model_layers_3_residual_post_attn/min":11.375,"train/tensor_act_model_layers_61_mlp_gate_proj/std":1.8515625,"train/tensor_act_model_layers_25_residual_pre_attn/norm":204395.15625,"train/tensor_act_model_layers_23_mlp_down_proj/max":16.75,"train/tensor_act_model_layers_83_mlp_gate_proj/std":1.7734375,"train/tensor_act_model_layers_23_residual_post_attn/max":94,"train/tensor_act_model_layers_6_mlp_gate_proj/range":6.78125,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_55_mlp_down_proj_weight/range":0.25146484375,"train/tensor_act_model_layers_95_mlp_up_proj/max":4.40625,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/range":0.259765625,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_65_mlp_up_proj/norm":5393.90673828125,"train/tensor_act_model_layers_44_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_mlp_gate_proj/std":1.84375,"train/tensor_param_model_layers_55_mlp_up_proj_weight/min":-0.12353515625,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_51_residual_post_attn/mean":204,"train/tensor_param_model_layers_41_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_87_mlp_down_proj/range":2,"train/tensor_act_model_layers_1_residual_post_attn/norm":8193.548828125,"train/tensor_act_model_layers_23_mlp/range":0,"train/tensor_act_model_layers_49_residual_pre_attn/max_abs":198,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_28_mlp_down_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_5_mlp/mean":4,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_52_mlp_up_proj/norm":5469.14208984375,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_66_residual_post_mlp/min":264,"train/tensor_act_model_layers_19_mlp_up_proj/std":1.8046875,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_93_mlp_gate_proj/mean":-0.19140625,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/min":-0.146484375,"train/tensor_act_model_layers_65_residual_post_mlp/max_abs":266,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_38_mlp/max":4,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_89_mlp_up_proj/range":8.359375,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_52_residual_pre_attn/mean":208,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_42_mlp_down_proj_weight/norm":6.21875,"train/tensor_param_model_layers_70_mlp_up_proj_weight/max":0.12890625,"train/tensor_param_model_layers_72_mlp_up_proj_weight/std":0.037841796875,"train/tensor_act_model_layers_17_residual_pre_attn/min":66,"train/tensor_act_model_layers_79_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_32_mlp_up_proj_weight/max_abs":0.123046875,"train/tensor_param_model_layers_47_mlp_up_proj_weight/mean":0.0003490447998046875,"train/tensor_act_model_layers_23_residual_post_attn/min":90,"train/tensor_param_model_layers_32_mlp_up_proj_weight/max":0.123046875,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_78_mlp_down_proj_weight/std":0.038818359375,"train/tensor_act_model_layers_23_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/mean":-0.001190185546875,"train/tensor_act_model_layers_47_residual_post_attn/mean":188,"train/tensor_param_model_layers_86_mlp_up_proj_weight/norm":7.96875,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_63_mlp_up_proj/max":4.09375,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_88_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_51_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_87_residual_post_mlp/range":6,"train/tensor_param_model_layers_22_mlp_down_proj_weight/norm":6.21875,"train/tensor_param_model_layers_20_mlp_up_proj_weight/max":0.12890625,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_72_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/mean":-0.000286102294921875,"train/tensor_act_model_layers_86_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_14_mlp_down_proj/std":0.408203125,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/std":0.036376953125,"train/tensor_act_model_layers_46_residual_post_mlp/std":0.80859375,"train/tensor_act_model_layers_8_residual_post_attn/norm":65464.4375,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/std":0.037353515625,"train/tensor_act_model_layers_46_residual_post_attn/mean":184,"train/tensor_act_model_layers_20_residual_post_attn/min":78,"train/tensor_act_model_layers_71_mlp_gate_proj/min":-3.640625,"train/tensor_act_model_layers_65_residual_post_attn/max":262,"train/tensor_act_model_layers_80_mlp_gate_proj/norm":5252.36767578125,"train/tensor_act_model_layers_88_residual_post_attn/min":348,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_68_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_92_mlp_up_proj_weight/max_abs":0.1396484375,"train/tensor_param_model_layers_12_mlp_down_proj_weight/min":-0.11572265625,"train/tensor_act_model_layers_7_mlp_up_proj/max_abs":3.75,"train/tensor_act_model_layers_3_mlp_down_proj/norm":30677.93359375,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/std":0.03369140625,"train/tensor_param_model_layers_79_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_1_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_residual_post_attn/std":0.8359375,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_95_mlp_down_proj_weight/range":0.3076171875,"train/tensor_param_model_layers_71_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/max_abs":0.11767578125,"train/tensor_act_model_layers_61_residual_pre_attn/max_abs":246,"train/tensor_act_model_layers_68_residual_pre_attn/mean":272,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_36_mlp_up_proj_weight/range":0.25048828125,"train/tensor_param_model_layers_60_mlp_down_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_7_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_46_mlp_up_proj/norm":5405.67822265625,"train/tensor_act_model_layers_14_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_55_residual_post_mlp/max":226,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_73_mlp_down_proj_weight/max":0.1318359375,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_0_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_40_mlp_up_proj/range":6.859375,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_67_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_22_mlp/norm":8192,"train/tensor_param_model_layers_26_mlp_up_proj_weight/std":0.035400390625,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_residual_pre_attn/min":149,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_77_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_12_mlp_up_proj/max_abs":3.703125,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/norm":5.75,"train/tensor_act_model_layers_25_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp_up_proj/min":-3.203125,"train/tensor_act_model_layers_25_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/mean":0.0004215240478515625,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_25_residual_post_mlp/mean":104,"train/tensor_act_model_layers_51_mlp_down_proj/norm":30431.458984375,"train/tensor_act_model_layers_75_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_80_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_85_mlp_down_proj/norm":30738.5546875,"train/tensor_act_model_layers_90_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_59_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/range":0.2841796875,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_20_residual_post_attn/max":82,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/mean":3.078125,"train/tensor_act_model_layers_22_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_7_residual_pre_attn/min":27,"train/tensor_act_model_layers_31_mlp_gate_proj/norm":5285.00244140625,"train/tensor_act_model_layers_19_residual_post_mlp/min":78,"train/tensor_act_model_layers_17_residual_post_attn/mean":68,"train/tensor_act_model_layers_88_residual_post_attn/max":354,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_18_mlp/norm":8192,"train/tensor_act_model_layers_6_residual_pre_attn/norm":49132.38671875,"train/tensor_act_model_layers_82_mlp_gate_proj/mean":-0.01507568359375,"train/tensor_act_model_layers_75_residual_post_mlp/min":300,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_65_residual_post_attn/min":256,"train/tensor_act_model_layers_29_residual_post_attn/min":114,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_33_residual_pre_attn/mean":132,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/range":0.22119140625,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_54_mlp_down_proj/mean":14.9375,"train/tensor_param_model_layers_18_mlp_down_proj_weight/range":0.25048828125,"train/tensor_act_model_layers_54_mlp_up_proj/std":1.828125,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_74_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_3_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_8_mlp/min":4,"train/tensor_act_model_layers_35_mlp_up_proj/mean":-0.0712890625,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_25_mlp_down_proj/max":16.25,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/max":0.125,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_31_mlp/std":0,"train/tensor_act_model_layers_29_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_68_mlp_gate_proj/max_abs":3.9375,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/min":-0.10693359375,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_76_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_30_mlp_gate_proj/mean":-0.11279296875,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_90_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_66_mlp_up_proj/max_abs":3.734375,"train/tensor_act_model_layers_93_mlp/max_abs":4,"train/tensor_act_model_layers_34_mlp_down_proj/max":16.25,"train/tensor_act_model_layers_99_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_5_mlp_down_proj/range":1.875,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_9_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_mlp/min":4,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_35_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_36_residual_post_mlp/range":5,"train/tensor_act_model_layers_16_mlp_gate_proj/min":-3.71875,"train/tensor_act_model_layers_74_residual_post_attn/range":6,"train/tensor_act_model_layers_52_mlp/min":4,"train/tensor_act_model_layers_34_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_89_mlp_gate_proj/max":4.09375,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_84_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_54_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_1_mlp_up_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_89_mlp_up_proj_weight/max_abs":0.1494140625,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_83_mlp_down_proj_weight/range":0.298828125,"train/tensor_param_model_layers_15_mlp_up_proj_weight/mean":-0.00063323974609375,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_19_mlp_down_proj_weight/mean":0.00170135498046875,"train/tensor_act_model_layers_31_residual_post_mlp/max":130,"train/tensor_param_model_layers_95_mlp_up_proj_weight/min":-0.1513671875,"train/tensor_act_model_layers_45_residual_post_mlp/max":186,"train/tensor_act_model_layers_56_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_43_mlp_up_proj/max":3.78125,"train/tensor_act_model_layers_94_residual_post_mlp/min":376,"train/tensor_act_model_layers_86_mlp/mean":4,"train/tensor_param_model_layers_56_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_27_residual_post_attn/std":0.625,"train/tensor_act_model_layers_7_residual_post_mlp/min":30.875,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_17_mlp/min":4,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_35_mlp_down_proj/max":16.5,"train/tensor_act_model_layers_70_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_76_mlp_down_proj/max_abs":16.375,"train/tensor_act_model_layers_64_residual_post_attn/std":0.76953125,"train/tensor_act_model_layers_76_residual_post_mlp/max":310,"train/tensor_act_model_layers_25_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_12_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_residual_post_mlp/mean":28,"train/tensor_act_model_layers_29_residual_post_attn/max":118.5,"train/tensor_act_model_layers_25_residual_post_mlp/norm":212562.453125,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_56_mlp_down_proj/range":2.5,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_38_residual_post_attn/norm":310758.375,"train/tensor_act_model_layers_40_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/min":-0.11181640625,"train/tensor_act_model_layers_35_residual_post_mlp/max_abs":146,"train/tensor_act_model_layers_23_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_59_mlp/norm":8192,"train/tensor_act_model_layers_68_residual_post_attn/norm":556654.3125,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_71_residual_pre_attn/mean":284,"train/tensor_act_model_layers_59_residual_pre_attn/min":233,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_89_mlp_up_proj/min":-4.5,"train/tensor_param_model_layers_7_mlp_down_proj_weight/max":0.1142578125,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_97_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_23_mlp_down_proj_weight/norm":6.21875,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_82_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_81_mlp_down_proj/norm":30546.33203125,"train/tensor_act_model_layers_25_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_62_residual_post_attn/max_abs":250,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/min":-0.1337890625,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_40_mlp_gate_proj/max":3.65625,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_54_mlp_gate_proj/range":7.328125,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_40_mlp_up_proj_weight/mean":-0.00191497802734375,"train/tensor_act_model_layers_92_residual_post_attn/min":364,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_69_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_26_mlp/range":0,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_80_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_mlp_up_proj/max_abs":3.46875,"train/tensor_param_model_layers_85_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_92_residual_post_attn/max":370,"train/tensor_act_model_layers_91_mlp/std":0,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_14_mlp_up_proj_weight/norm":6.15625,"train/tensor_param_model_layers_60_mlp_up_proj_weight/norm":6.46875,"train/tensor_act_model_layers_22_residual_pre_attn/std":0.58203125,"train/tensor_act_model_layers_60_residual_post_mlp/std":0.8359375,"train/tensor_act_model_layers_96_mlp_up_proj/std":1.8671875,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_33_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_94_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_residual_pre_attn/max_abs":16.5,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_0_residual_post_attn/norm":188.75314331054688,"train/tensor_act_model_layers_67_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_62_residual_post_mlp/std":0.89453125,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_46_mlp/mean":4,"train/tensor_act_model_layers_13_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_0_mlp_up_proj_weight/norm":10.375,"train/tensor_act_model_layers_57_residual_pre_attn/min":225,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_80_mlp_up_proj_weight/range":0.2666015625,"train/tensor_act_model_layers_28_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_28_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_mlp_down_proj/min":14.0625,"train/tensor_act_model_layers_39_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_up_proj/norm":5259.265625,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_52_mlp_gate_proj/max":3.4375,"train/tensor_act_model_layers_71_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_24_mlp_up_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_27_residual_post_mlp/min":110,"train/tensor_act_model_layers_58_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_19_residual_post_mlp/max_abs":82,"train/tensor_act_model_layers_49_residual_post_attn/norm":400885.46875,"train/tensor_act_model_layers_90_residual_post_attn/range":6,"train/tensor_act_model_layers_60_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_12_residual_pre_attn/mean":48,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_0_mlp_down_proj/std":21,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_64_residual_post_mlp/mean":260,"train/tensor_act_model_layers_86_residual_post_attn/norm":704109.875,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_91_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_55_residual_pre_attn/max_abs":222,"train/tensor_act_model_layers_72_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_27_residual_post_mlp/mean":112,"train/tensor_act_model_layers_99_residual_post_attn/min":392,"train/tensor_act_model_layers_3_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_74_residual_post_attn/max":298,"train/tensor_act_model_layers_75_mlp_gate_proj/min":-3.90625,"train/tensor_param_model_layers_5_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_86_mlp/max_abs":4,"train/tensor_act_model_layers_94_residual_post_mlp/range":6,"train/tensor_param_model_layers_17_mlp_up_proj_weight/max_abs":0.12451171875,"train/tensor_act_model_layers_75_residual_post_mlp/range":6,"train/tensor_param_model_layers_75_mlp_up_proj_weight/std":0.038330078125,"train/tensor_act_model_layers_57_mlp/min":4,"train/tensor_param_model_layers_93_mlp_down_proj_weight/max":0.1357421875,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_85_residual_post_attn/range":6,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_79_residual_pre_attn/min":312,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_1_mlp_down_proj_weight/min":-0.12109375,"train/tensor_act_model_layers_79_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_28_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_49_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_98_residual_post_attn/range":6,"train/tensor_act_model_layers_70_mlp_up_proj/max":3.921875,"train/tensor_act_model_layers_39_mlp/range":0,"train/tensor_param_model_layers_45_mlp_up_proj_weight/mean":-0.000926971435546875,"train/tensor_param_model_layers_48_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_74_mlp_up_proj/std":1.8203125,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_14_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_8_residual_pre_attn/max_abs":33.5,"train/tensor_act_model_layers_74_residual_post_attn/min":292,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_88_mlp_down_proj/mean":15.0625,"train/tensor_act_model_layers_99_residual_post_attn/max":398,"train/tensor_act_model_layers_50_mlp/range":0,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_99_mlp_down_proj_weight/std":0.04833984375,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_62_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_70_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_50_residual_post_mlp/std":0.81640625,"train/tensor_act_model_layers_28_mlp/norm":8192,"train/tensor_act_model_layers_56_mlp_down_proj/std":0.435546875,"train/tensor_act_model_layers_14_mlp_gate_proj/max_abs":3.671875,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_95_mlp_down_proj/max":17.625,"train/tensor_act_model_layers_25_mlp/max":4,"train/tensor_act_model_layers_4_mlp_down_proj/range":2.3125,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_89_residual_post_mlp/min":356,"train/tensor_act_model_layers_60_residual_post_mlp/norm":499204.84375,"train/tensor_act_model_layers_20_mlp_up_proj/std":1.890625,"train/tensor_act_model_layers_73_mlp/norm":8192,"train/tensor_param_model_layers_59_mlp_up_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_2_mlp_down_proj/max_abs":17.625,"train/tensor_param_model_layers_73_mlp_up_proj_weight/norm":6.84375,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_54_mlp_down_proj/max_abs":16.375,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_7_mlp_down_proj/std":0.373046875,"train/tensor_act_model_layers_86_residual_post_mlp/min":344,"train/tensor_act_model_layers_54_mlp_gate_proj/std":1.828125,"train/tensor_act_model_layers_85_residual_pre_attn/min":336,"train/tensor_act_model_layers_4_mlp_gate_proj/mean":-0.08447265625,"train/tensor_act_model_layers_1_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_79_mlp_up_proj/norm":5321.001953125,"train/tensor_act_model_layers_64_mlp_up_proj/std":1.8515625,"train/tensor_act_model_layers_14_mlp/norm":8192,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_28_residual_post_attn/range":4.5,"train/tensor_act_model_layers_81_residual_post_attn/max":326,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_4_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_52_mlp_down_proj/max_abs":16.125,"train/tensor_param_model_layers_35_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp/max":4,"train/tensor_act_model_layers_39_residual_post_mlp/std":0.80859375,"train/tensor_param_model_layers_39_mlp_down_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_4_mlp_gate_proj/max_abs":3.703125,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_64_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_mlp_gate_proj/max":3.875,"train/tensor_act_model_layers_88_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_81_mlp_down_proj/range":2.25,"train/tensor_act_model_layers_55_mlp_up_proj/norm":5353.39990234375,"train/tensor_act_model_layers_61_mlp/range":0,"train/tensor_param_model_layers_31_mlp_up_proj_weight/min":-0.12060546875,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_97_mlp_down_proj_weight/max_abs":0.146484375,"train/tensor_param_model_layers_28_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_9_mlp_up_proj_weight/max_abs":0.1220703125,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_44_mlp_gate_proj/range":7.484375,"train/tensor_param_model_layers_93_mlp_up_proj_weight/max":0.1455078125,"train/tensor_act_model_layers_61_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_93_mlp_up_proj_weight/max_abs":0.1455078125,"train/tensor_act_model_layers_1_mlp/min":4,"train/tensor_act_model_layers_24_residual_post_attn/max_abs":98,"train/tensor_param_model_layers_64_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_30_mlp_down_proj/range":1.8125,"train/tensor_act_model_layers_27_residual_pre_attn/norm":220754.3125,"train/tensor_param_model_layers_90_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_44_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_mlp_up_proj/norm":5391.3427734375,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_71_mlp_down_proj_weight/range":0.2666015625,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_81_residual_post_attn/min":320,"train/tensor_act_model_layers_64_residual_pre_attn/norm":523449.15625,"train/tensor_act_model_layers_74_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/max":0.150390625,"train/tensor_act_model_layers_62_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_residual_pre_attn/min":252,"train/tensor_act_model_layers_76_residual_post_attn/norm":622190.125,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/std":0.03466796875,"train/tensor_param_model_layers_32_mlp_up_proj_weight/norm":6.34375,"train/tensor_param_model_layers_60_mlp_up_proj_weight/std":0.03564453125,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/range":0.2998046875,"train/tensor_act_model_layers_93_residual_post_attn/max_abs":374,"train/tensor_act_model_layers_58_mlp_up_proj/range":7.484375,"train/tensor_act_model_layers_48_mlp_up_proj/max":3.953125,"train/tensor_act_model_layers_27_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_57_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_80_residual_pre_attn/max":322,"train/tensor_act_model_layers_13_residual_post_mlp/min":54.5,"train/tensor_param_model_layers_31_mlp_up_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/min":-0.12890625,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_37_mlp_up_proj_weight/norm":6.375,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_96_mlp_up_proj_weight/std":0.04833984375,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_11_mlp_up_proj/max":3.453125,"train/tensor_act_model_layers_74_mlp_up_proj/mean":0.0034027099609375,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_60_residual_pre_attn/norm":491012.90625,"train/tensor_param_model_layers_54_mlp_down_proj_weight/norm":6.3125,"train/tensor_act_model_layers_68_residual_post_mlp/min":272,"train/tensor_param_model_layers_86_mlp_up_proj_weight/std":0.0439453125,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_30_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_78_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_52_mlp_up_proj/range":7.40625,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_72_mlp/norm":8192,"train/tensor_param_model_layers_30_mlp_down_proj_weight/norm":6.25,"train/tensor_act_model_layers_51_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_36_mlp_down_proj/max":16.625,"train/tensor_param_model_layers_25_mlp_down_proj_weight/norm":6.25,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_46_residual_pre_attn/max_abs":186,"train/tensor_act_model_layers_44_mlp_gate_proj/min":-3.90625,"train/tensor_act_model_layers_47_mlp_gate_proj/mean":0.1494140625,"train/tensor_act_model_layers_32_mlp_gate_proj/norm":5385.76513671875,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp_up_proj/max_abs":3.65625,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_73_mlp_gate_proj/norm":5306.48974609375,"train/tensor_act_model_layers_79_residual_post_attn/range":6,"train/tensor_act_model_layers_6_mlp_gate_proj/norm":5326.57666015625,"train/tensor_act_model_layers_90_mlp_up_proj/max_abs":4.46875,"train/tensor_act_model_layers_53_residual_post_attn/max":214,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_53_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_7_residual_post_mlp/max":33.5,"train/tensor_act_model_layers_98_mlp/range":0,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_51_residual_post_attn/min":201,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/norm":8.125,"train/tensor_act_model_layers_63_residual_pre_attn/range":6,"train/tensor_param_model_layers_40_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_53_residual_post_attn/min":209,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_70_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_52_residual_pre_attn/range":5,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_1_residual_post_mlp/max_abs":8.4375,"train/tensor_act_model_layers_51_residual_post_attn/max":206,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_37_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_14_residual_pre_attn/min":54.5,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_residual_pre_attn/min":229,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_36_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_90_mlp_down_proj/std":0.55859375,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_54_mlp_up_proj_weight/mean":0.00151824951171875,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/min":-0.13671875,"train/tensor_param_model_layers_62_mlp_down_proj_weight/mean":0.00052642822265625,"train/tensor_act_model_layers_20_mlp_down_proj/std":0.45703125,"train/tensor_act_model_layers_65_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_96_mlp_gate_proj/range":8.78125,"train/tensor_act_model_layers_22_mlp_gate_proj/max_abs":4.125,"train/tensor_act_model_layers_92_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_28_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/min":-0.1357421875,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/range":0.25,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_37_mlp/max_abs":4,"train/tensor_act_model_layers_63_residual_post_attn/max_abs":254,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_51_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_31_mlp/norm":8192,"train/tensor_act_model_layers_64_residual_post_mlp/norm":532078.4375,"train/tensor_param_model_layers_47_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_46_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp_down_proj/mean":14.9375,"train/tensor_param_model_layers_96_mlp_down_proj_weight/range":0.28515625,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_43_residual_pre_attn/std":0.80859375,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_26_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_20_residual_post_mlp/norm":171675.4375,"train/tensor_act_model_layers_15_mlp_up_proj/min":-3.3125,"train/tensor_act_model_layers_89_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_97_mlp_up_proj_weight/min":-0.150390625,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_89_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/norm":6.46875,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_9_residual_post_attn/range":3,"train/tensor_act_model_layers_88_residual_post_mlp/norm":728685.8125,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_71_mlp/max":4,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_83_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_88_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_residual_pre_attn/range":5,"train/tensor_param_model_layers_46_mlp_up_proj_weight/range":0.255859375,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/max_abs":0.1328125,"train/tensor_act_model_layers_74_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_18_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_94_mlp_up_proj/range":8.875,"train/tensor_act_model_layers_32_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_97_mlp_up_proj/norm":5348.9130859375,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_82_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_97_mlp_down_proj_weight/range":0.2890625,"train/tensor_act_model_layers_73_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_41_residual_post_mlp/max_abs":170,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_93_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_residual_post_mlp/std":0.81640625,"train/tensor_param_model_layers_33_mlp_down_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_28_residual_post_attn/mean":111.5,"train/tensor_act_model_layers_94_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_14_residual_pre_attn/range":3,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_99_mlp_down_proj_weight/range":0.306640625,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/max":0.12451171875,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_5_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_92_residual_pre_attn/norm":753261.6875,"train/tensor_act_model_layers_29_mlp_gate_proj/min":-3.28125,"train/tensor_act_model_layers_19_residual_post_mlp/max":82,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_16_mlp_down_proj_weight/max":0.1240234375,"train/tensor_param_model_layers_19_mlp_up_proj_weight/min":-0.1171875,"train/tensor_act_model_layers_22_residual_pre_attn/mean":88,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_25_mlp_down_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_87_mlp_up_proj/std":1.8046875,"train/tensor_act_model_layers_75_residual_post_mlp/max":306,"train/tensor_act_model_layers_99_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_86_mlp_down_proj_weight/min":-0.1396484375,"train/tensor_act_model_layers_31_mlp_gate_proj/min":-3.75,"train/tensor_param_model_layers_76_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_71_mlp_up_proj_weight/max_abs":0.1357421875,"train/tensor_param_model_layers_47_mlp_up_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_82_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_65_residual_post_mlp/norm":540270.375,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/range":0.298828125,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_57_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_16_residual_post_mlp/std":0.55859375,"train/tensor_act_model_layers_32_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_60_mlp/min":4,"train/tensor_act_model_layers_6_mlp/max_abs":4,"train/tensor_act_model_layers_38_mlp_gate_proj/mean":0.003448486328125,"train/tensor_act_model_layers_12_mlp_down_proj/max_abs":15.6875,"train/tensor_act_model_layers_95_residual_post_attn/max_abs":382,"train/tensor_act_model_layers_57_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_mlp/min":4,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_gate_proj/min":-3.390625,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_90_residual_post_attn/mean":360,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_27_mlp_down_proj/max":15.875,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_98_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_mlp/min":4,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/norm":6.78125,"train/tensor_act_model_layers_11_residual_pre_attn/min":42.5,"train/tensor_param_model_layers_86_mlp_down_proj_weight/mean":0.000637054443359375,"train/tensor_act_model_layers_32_mlp_up_proj/max_abs":3.828125,"train/tensor_param_model_layers_30_mlp_up_proj_weight/max":0.126953125,"train/tensor_act_model_layers_49_mlp/mean":4,"train/tensor_act_model_layers_94_mlp_gate_proj/range":9,"train/tensor_act_model_layers_86_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_17_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_mlp/norm":8192,"train/tensor_act_model_layers_80_mlp_down_proj/mean":14.9375,"train/tensor_param_model_layers_98_mlp_up_proj_weight/range":0.3115234375,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/norm":6.25,"train/tensor_act_model_layers_41_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_53_mlp_down_proj_weight/max_abs":0.12353515625,"train/tensor_param_model_layers_81_mlp_up_proj_weight/min":-0.130859375,"train/tensor_param_model_layers_41_mlp_up_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_13_residual_post_mlp/max":57.5,"train/tensor_act_model_layers_74_residual_pre_attn/norm":605806.125,"train/tensor_param_model_layers_42_mlp_down_proj_weight/range":0.2490234375,"train/tensor_param_model_layers_88_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_64_residual_post_mlp/max_abs":262,"train/tensor_param_model_layers_2_mlp_down_proj_weight/min":-0.0986328125,"train/tensor_act_model_layers_51_residual_post_mlp/norm":425461.25,"train/tensor_act_model_layers_92_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_68_residual_post_mlp/max":278,"train/tensor_act_model_layers_46_mlp_gate_proj/mean":0.0120849609375,"train/tensor_act_model_layers_98_mlp_up_proj/max_abs":4.96875,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/mean":0,"eval/runtime":16.1593,"train/tensor_act_model_layers_89_mlp_down_proj/norm":30883.82421875,"train/tensor_act_model_layers_58_mlp/range":0,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_68_mlp_gate_proj/std":1.8671875,"train/tensor_param_model_layers_56_mlp_down_proj_weight/max":0.1259765625,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_92_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_71_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_mlp/norm":8192,"train/tensor_act_model_layers_85_residual_post_mlp/range":6,"train/tensor_act_model_layers_86_mlp_gate_proj/min":-3.859375,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_47_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_80_mlp/min":4,"train/tensor_act_model_layers_34_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_14_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_7_mlp_up_proj_weight/std":0.033203125,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_62_mlp/mean":4,"train/tensor_act_model_layers_37_residual_post_attn/mean":148,"train/tensor_act_model_layers_78_mlp/std":0,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_0_mlp_gate_proj/min":-5.84375,"train/tensor_act_model_layers_69_mlp_down_proj/range":1.75,"train/tensor_act_model_layers_92_mlp_gate_proj/mean":0.043212890625,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_67_mlp_gate_proj/norm":5351.10400390625,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/mean":0.0002880096435546875,"train/tensor_act_model_layers_88_mlp/range":0,"train/tensor_act_model_layers_53_residual_post_mlp/norm":441845.125,"train/tensor_act_model_layers_74_mlp/max":4,"train/tensor_act_model_layers_99_residual_pre_attn/norm":810605.625,"train/tensor_act_model_layers_31_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_81_residual_post_mlp/mean":328,"train/tensor_param_model_layers_31_mlp_down_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_74_mlp_up_proj_weight/max":0.125,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_6_mlp_up_proj_weight/std":0.032958984375,"train/tensor_act_model_layers_89_residual_post_attn/mean":356,"train/tensor_act_model_layers_46_mlp_up_proj/range":7.453125,"train/tensor_act_model_layers_18_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_mlp_gate_proj/max":3.609375,"train/tensor_param_model_layers_68_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_85_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_10_mlp_gate_proj/mean":0.0400390625,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_27_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_54_mlp/max_abs":4,"train/tensor_act_model_layers_84_mlp_gate_proj/max":4.21875,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_90_mlp/std":0,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_51_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_4_mlp_up_proj_weight/norm":5.71875,"train/tensor_act_model_layers_39_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_3_mlp_up_proj_weight/range":0.20751953125,"train/tensor_param_model_layers_41_mlp_up_proj_weight/norm":6.28125,"train/tensor_act_model_layers_36_mlp_gate_proj/range":7.125,"train/tensor_act_model_layers_50_mlp_up_proj/std":1.8515625,"train/tensor_act_model_layers_85_mlp_gate_proj/range":8.40625,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_37_mlp_gate_proj/max_abs":3.875,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_97_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_residual_pre_attn/max_abs":330,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_81_residual_pre_attn/norm":663150,"train/tensor_act_model_layers_64_residual_post_attn/min":252,"train/tensor_act_model_layers_12_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/std":0.047119140625,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_9_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_15_residual_post_mlp/min":62.5,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_27_residual_post_attn/min":106,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_99_residual_post_mlp/mean":400,"train/tensor_act_model_layers_68_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_48_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_96_mlp/min":4,"train/tensor_act_model_layers_28_mlp_gate_proj/max":3.28125,"train/tensor_act_model_layers_31_residual_post_attn/mean":123.5,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_43_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_3_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_7_mlp_gate_proj/min":-3.59375,"train/tensor_act_model_layers_46_mlp/min":4,"train/tensor_param_model_layers_55_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_79_mlp_up_proj/std":1.8359375,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_mlp/std":0,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/max":0.1435546875,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/max":0.125,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_41_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_34_mlp/norm":8192,"train/tensor_act_model_layers_92_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_74_residual_post_mlp/mean":300,"train/tensor_act_model_layers_58_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_27_residual_post_attn/max":110,"train/tensor_act_model_layers_75_mlp/norm":8192,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_21_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_98_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_22_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_93_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_68_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_residual_post_attn/max":258,"train/tensor_param_model_layers_60_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_18_mlp_up_proj_weight/range":0.2578125,"train/tensor_act_model_layers_95_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_76_mlp/mean":4,"train/tensor_act_model_layers_82_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_69_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_29_residual_post_attn/std":0.67578125,"train/tensor_act_model_layers_88_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_59_mlp_up_proj/mean":-0.1201171875,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_27_mlp_down_proj_weight/norm":6.34375,"train/tensor_param_model_layers_99_mlp_up_proj_weight/mean":0.0001964569091796875,"train/tensor_act_model_layers_45_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_62_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_68_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_41_mlp_gate_proj/min":-3.703125,"train/tensor_act_model_layers_90_mlp_gate_proj/mean":0.205078125,"train/tensor_act_model_layers_91_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/std":0.0419921875,"train/tensor_act_model_layers_23_mlp_down_proj/norm":30491.583984375,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_9_residual_post_mlp/min":38.5,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_65_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_77_mlp_down_proj_weight/max":0.1318359375,"train/tensor_act_model_layers_25_mlp_gate_proj/norm":5337.7802734375,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/max_abs":0.1376953125,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/std":0.037841796875,"train/tensor_param_model_layers_13_mlp_up_proj_weight/std":0.0341796875,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_23_mlp_up_proj/max":3.703125,"train/tensor_param_model_layers_98_mlp_up_proj_weight/std":0.048095703125,"train/tensor_act_model_layers_24_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_98_mlp/max":4,"train/tensor_act_model_layers_23_residual_pre_attn/range":4,"train/tensor_param_model_layers_90_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_60_residual_post_attn/max":242,"train/tensor_act_model_layers_17_mlp_down_proj/max":16.25,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_mlp_gate_proj/max_abs":4.28125,"train/tensor_act_model_layers_97_mlp_up_proj/max":4.34375,"train/tensor_act_model_layers_60_residual_post_attn/max_abs":242,"train/tensor_act_model_layers_65_residual_post_attn/range":6,"train/tensor_act_model_layers_79_mlp_down_proj/range":3.1875,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/std":0.04833984375,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_7_mlp_gate_proj/norm":5479.712890625,"train/tensor_act_model_layers_35_residual_pre_attn/std":0.7109375,"train/tensor_act_model_layers_44_residual_pre_attn/min":173,"train/tensor_act_model_layers_29_residual_pre_attn/max_abs":118.5,"train/tensor_act_model_layers_93_mlp/range":0,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_32_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_40_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_residual_post_mlp/max":114,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_25_residual_post_mlp/range":4,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_73_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_14_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_10_mlp_up_proj_weight/norm":6.15625,"train/tensor_act_model_layers_97_residual_post_attn/max_abs":390,"train/tensor_act_model_layers_77_mlp_up_proj/norm":5364.07861328125,"train/tensor_act_model_layers_84_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_99_residual_post_mlp/norm":818797.6875,"train/tensor_act_model_layers_10_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_53_residual_pre_attn/norm":433653.1875,"train/tensor_param_model_layers_86_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_73_mlp_gate_proj/std":1.828125,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/range":0.251953125,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_51_residual_pre_attn/norm":417269.3125,"train/tensor_act_model_layers_74_residual_post_mlp/norm":613998.125,"train/tensor_act_model_layers_71_residual_post_mlp/range":6,"train/tensor_act_model_layers_77_mlp_up_proj/std":1.8359375,"train/tensor_act_model_layers_3_mlp_gate_proj/range":6.953125,"train/tensor_param_model_layers_24_mlp_up_proj_weight/norm":6.375,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_98_residual_post_mlp/range":6,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_40_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_9_residual_pre_attn/min":34.5,"train/tensor_act_model_layers_89_residual_post_mlp/max":362,"train/tensor_act_model_layers_80_mlp_down_proj/max_abs":16.875,"train/tensor_act_model_layers_79_mlp_down_proj/std":0.51953125,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_84_mlp_up_proj/range":8.03125,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_36_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_55_mlp/max":4,"train/tensor_act_model_layers_4_residual_post_attn/mean":16,"train/tensor_act_model_layers_73_residual_pre_attn/min":288,"train/tensor_act_model_layers_7_mlp_down_proj/max":16,"train/tensor_act_model_layers_74_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_17_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_60_residual_post_attn/min":237,"train/tensor_act_model_layers_52_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_21_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_20_residual_post_attn/std":0.58203125,"train/tensor_act_model_layers_17_mlp_up_proj/mean":-0.0615234375,"train/tensor_act_model_layers_46_mlp_gate_proj/max":3.265625,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_60_mlp_up_proj/range":7.21875,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_92_mlp_gate_proj/range":8.40625,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_12_mlp/min":4,"train/tensor_act_model_layers_88_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_93_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_31_mlp_up_proj/max":3.453125,"train/tensor_act_model_layers_24_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_89_mlp/norm":8192,"train/tensor_act_model_layers_95_mlp_up_proj/mean":0.1279296875,"train/tensor_act_model_layers_17_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_51_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_8_mlp_up_proj_weight/min":-0.123046875,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_54_mlp_down_proj_weight/std":0.034912109375,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_residual_post_attn/std":0.58203125,"train/tensor_act_model_layers_24_mlp/mean":4,"train/tensor_act_model_layers_54_mlp_gate_proj/norm":5339.44921875,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/mean":-0.0003528594970703125,"train/tensor_act_model_layers_49_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_7_mlp_up_proj/max":3.75,"train/tensor_act_model_layers_22_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_18_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_3_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_47_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_28_mlp_up_proj/min":-3.515625,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/max_abs":0.15234375,"train/tensor_act_model_layers_46_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_86_residual_post_mlp/max":350,"train/tensor_act_model_layers_46_mlp/max":4,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_49_mlp_up_proj/min":-4.1875,"train/tensor_param_model_layers_39_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_69_residual_pre_attn/max":278,"train/tensor_act_model_layers_28_mlp_up_proj/range":6.953125,"train/tensor_param_model_layers_20_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_39_mlp_up_proj/mean":-0.1923828125,"train/tensor_act_model_layers_66_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_51_residual_post_mlp/mean":208,"train/tensor_act_model_layers_47_mlp_gate_proj/range":7.734375,"train/tensor_act_model_layers_13_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_47_mlp_down_proj_weight/max_abs":0.12451171875,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_84_residual_post_attn/max_abs":338,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_84_mlp_up_proj/max_abs":4.28125,"train/tensor_act_model_layers_82_mlp/max":4,"train/tensor_param_model_layers_96_mlp_down_proj_weight/mean":-6.532669067382812e-05,"train/tensor_act_model_layers_53_residual_post_mlp/mean":216,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_42_residual_post_mlp/std":0.80859375,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_9_mlp_up_proj/std":1.8515625,"train/tensor_param_model_layers_96_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_20_mlp_gate_proj/mean":-0.17578125,"train/tensor_param_model_layers_62_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_91_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp/range":0,"train/tensor_act_model_layers_84_mlp_gate_proj/norm":5165.4765625,"train/tensor_act_model_layers_52_residual_pre_attn/min":205,"train/tensor_act_model_layers_61_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_2_mlp_up_proj_weight/std":0.03173828125,"train/tensor_param_model_layers_52_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_57_mlp_up_proj/mean":-0.0208740234375,"train/tensor_act_model_layers_19_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/std":0.045166015625,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_95_residual_post_attn/mean":380,"train/tensor_param_model_layers_2_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/max_abs":0.1357421875,"train/tensor_act_model_layers_26_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_40_mlp_down_proj/max_abs":16,"train/tensor_act_model_layers_71_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_63_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_44_mlp_up_proj_weight/max":0.12060546875,"train/tensor_act_model_layers_22_mlp_gate_proj/std":1.875,"train/tensor_act_model_layers_32_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_35_mlp_up_proj_weight/norm":6.375,"train/tensor_act_model_layers_81_residual_post_mlp/norm":671341.9375,"train/tensor_act_model_layers_63_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_0_mlp_down_proj/max":112,"train/tensor_param_model_layers_77_mlp_up_proj_weight/range":0.265625,"train/tensor_param_model_layers_45_mlp_down_proj_weight/min":-0.123046875,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_90_mlp_up_proj_weight/min":-0.142578125,"train/tensor_act_model_layers_61_residual_pre_attn/mean":244,"train/tensor_param_model_layers_53_mlp_up_proj_weight/max_abs":0.1279296875,"train/tensor_param_model_layers_12_mlp_down_proj_weight/mean":0.00012683868408203125,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_20_residual_post_mlp/mean":84,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_88_residual_post_mlp/mean":356,"train/tensor_act_model_layers_5_mlp/min":4,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_25_residual_post_attn/max":102,"train/tensor_act_model_layers_29_residual_post_mlp/norm":245186.5625,"train/tensor_act_model_layers_23_residual_pre_attn/norm":188059.078125,"train/tensor_act_model_layers_8_residual_post_mlp/std":0.373046875,"train/tensor_act_model_layers_75_mlp_down_proj/range":2.1875,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_67_residual_post_mlp/max_abs":274,"train/tensor_act_model_layers_8_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_56_residual_pre_attn/mean":224,"train/tensor_act_model_layers_36_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_64_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_64_residual_post_attn/mean":256,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_70_residual_post_attn/range":6,"train/tensor_param_model_layers_55_mlp_down_proj_weight/norm":6.375,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_20_residual_post_attn/norm":163483.640625,"train/tensor_act_model_layers_8_mlp_gate_proj/mean":-0.035888671875,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_67_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_78_mlp/norm":8192,"train/tensor_act_model_layers_72_mlp_gate_proj/mean":-0.1162109375,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_72_mlp_down_proj_weight/min":-0.12890625,"train/tensor_act_model_layers_51_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_mlp_down_proj/max_abs":16.125,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/range":0.25,"train/tensor_act_model_layers_70_mlp_up_proj/mean":0.07763671875,"train/tensor_act_model_layers_76_residual_post_mlp/min":304,"train/tensor_act_model_layers_81_mlp_up_proj/range":7.625,"train/tensor_param_model_layers_26_mlp_down_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_71_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_90_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_30_mlp_up_proj/mean":-0.0625,"train/tensor_param_model_layers_32_mlp_down_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_63_residual_pre_attn/std":0.89453125,"train/tensor_act_model_layers_13_mlp_down_proj/range":3.8125,"train/tensor_act_model_layers_53_mlp_down_proj/range":2,"train/tensor_act_model_layers_64_mlp_down_proj/max":16.875,"train/tensor_act_model_layers_47_mlp/max":4,"train/tensor_act_model_layers_10_mlp_up_proj/mean":-0.130859375,"train/tensor_act_model_layers_16_mlp_up_proj/min":-3.78125,"train/tensor_act_model_layers_32_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_32_mlp_up_proj/max":3.3125,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/frac_near_user_limit":0.984649658203125,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_25_residual_post_attn/min":98,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_32_residual_pre_attn/min":126,"train/tensor_act_model_layers_92_residual_pre_attn/max":370,"train/tensor_act_model_layers_61_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_8_mlp_up_proj_weight/range":0.240234375,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_94_residual_post_mlp/max":382,"train/tensor_act_model_layers_21_residual_post_attn/norm":171675.4375,"train/tensor_act_model_layers_45_residual_post_mlp/min":181,"train/tensor_act_model_layers_31_residual_post_mlp/min":126,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_11_mlp_gate_proj/mean":0.0203857421875,"train/tensor_param_model_layers_63_mlp_down_proj_weight/norm":6.40625,"train/tensor_act_model_layers_23_mlp/mean":4,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_28_mlp/mean":4,"train/tensor_param_model_layers_89_mlp_up_proj_weight/max":0.140625,"train/tensor_act_model_layers_67_mlp_down_proj/max_abs":16.375,"train/tensor_act_model_layers_69_mlp/max_abs":4,"train/tensor_act_model_layers_65_residual_post_mlp/mean":264,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_65_residual_pre_attn/norm":532078.4375,"train/tensor_act_model_layers_5_mlp/std":0,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_8_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_71_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_8_mlp_up_proj/std":1.828125,"train/tensor_param_model_layers_61_mlp_down_proj_weight/mean":-0.0003604888916015625,"train/tensor_act_model_layers_24_residual_post_attn/mean":96,"train/tensor_param_model_layers_29_mlp_down_proj_weight/mean":0.0012969970703125,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_59_mlp_up_proj_weight/norm":6.4375,"train/tensor_act_model_layers_2_mlp_down_proj/range":3.5,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_9_residual_post_mlp/max":41.5,"train/tensor_act_model_layers_94_residual_pre_attn/max_abs":378,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_39_mlp_down_proj_weight/mean":-0.0003643035888671875,"train/tensor_act_model_layers_9_residual_pre_attn/range":3,"train/tensor_act_model_layers_85_residual_post_attn/max_abs":342,"train/tensor_act_model_layers_92_residual_post_mlp/norm":761453.75,"train/tensor_act_model_layers_64_mlp/mean":4,"train/tensor_act_model_layers_95_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_67_mlp_down_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_27_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_94_mlp_gate_proj/max":4.65625,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/range":0.24169921875,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_67_mlp_down_proj/std":0.416015625,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_82_mlp_up_proj/max_abs":3.96875,"train/tensor_act_model_layers_78_residual_post_attn/norm":638574.0625,"train/tensor_param_model_layers_77_mlp_up_proj_weight/mean":-0.00177001953125,"train/tensor_param_model_layers_28_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_51_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_98_mlp_down_proj/norm":30488.97265625,"train/tensor_act_model_layers_9_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_52_mlp_down_proj/range":1.8125,"train/tensor_act_model_layers_53_residual_post_attn/std":0.81640625,"train/tensor_act_model_layers_67_residual_post_attn/mean":268,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/max_abs":0.1357421875,"train/tensor_param_model_layers_56_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_36_residual_post_mlp/std":0.80859375,"train/tensor_act_model_layers_55_mlp/std":0,"train/tensor_act_model_layers_25_mlp/max_abs":4,"train/tensor_act_model_layers_51_residual_post_attn/std":0.81640625,"train/tensor_act_model_layers_69_residual_post_attn/norm":564846.3125,"train/tensor_act_model_layers_88_mlp_gate_proj/norm":5271.6025390625,"train/tensor_act_model_layers_4_residual_pre_attn/std":0.1611328125,"train/tensor_param_model_layers_72_mlp_down_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_48_mlp_gate_proj/norm":5351.36083984375,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_78_residual_pre_attn/mean":312,"train/tensor_act_model_layers_6_mlp_up_proj/max":3.4375,"train/tensor_act_model_layers_14_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_1_residual_post_mlp/std":0.11669921875,"train/tensor_act_model_layers_79_residual_pre_attn/mean":316,"train/tensor_act_model_layers_23_residual_post_mlp/mean":96,"train/tensor_param_model_layers_48_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_88_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_98_mlp/std":0,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/std":0.03662109375,"train/tensor_act_model_layers_20_residual_pre_attn/max":82,"train/tensor_act_model_layers_2_mlp/max_abs":4,"train/tensor_param_model_layers_24_mlp_down_proj_weight/min":-0.125,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/mean":-0.000949859619140625,"train/tensor_act_model_layers_48_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_27_mlp_down_proj_weight/range":0.24658203125,"train/tensor_act_model_layers_14_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/norm":6.15625,"train/tensor_act_model_layers_88_residual_pre_attn/norm":720493.8125,"train/tensor_act_model_layers_70_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_20_residual_pre_attn/norm":163485.40625,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/mean":-0.00058746337890625,"train/tensor_param_model_layers_35_mlp_down_proj_weight/norm":6.34375,"train/tensor_act_model_layers_80_mlp/max":4,"train/tensor_param_model_layers_23_mlp_up_proj_weight/std":0.03515625,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_1_mlp_up_proj_weight/min":-0.1123046875,"train/tensor_act_model_layers_39_mlp/norm":8192,"train/tensor_act_model_layers_13_residual_post_attn/mean":52,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_51_mlp_gate_proj/min":-3.75,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/std":0.048095703125,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_91_mlp_gate_proj/mean":-0.0201416015625,"train/tensor_act_model_layers_19_residual_pre_attn/range":4,"train/tensor_act_model_layers_18_mlp_gate_proj/mean":-0.15234375,"train/tensor_act_model_layers_81_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_27_residual_pre_attn/range":4,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_4_mlp_up_proj/max":3.546875,"train/tensor_act_model_layers_1_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/norm":6.3125,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_53_mlp_gate_proj/norm":5372.16357421875,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_70_mlp/min":4,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_51_mlp_up_proj/max_abs":3.796875,"train/tensor_act_model_layers_56_mlp/min":4,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_12_mlp_up_proj/max":3.453125,"train/tensor_act_model_layers_9_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_92_residual_post_mlp/mean":372,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_87_mlp_down_proj/std":0.458984375,"train/tensor_act_model_layers_6_residual_post_mlp/norm":57287.984375,"train/tensor_act_model_layers_21_mlp_gate_proj/std":1.8515625,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_5_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/max":0.11376953125,"train/tensor_act_model_layers_27_mlp_gate_proj/max_abs":3.515625,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/norm":6.375,"train/tensor_act_model_layers_60_mlp_up_proj/max_abs":3.734375,"train/tensor_act_model_layers_0_mlp/std":0.002655029296875,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_64_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_81_residual_post_attn/norm":663150,"train/tensor_act_model_layers_99_mlp_gate_proj/std":1.8984375,"train/tensor_act_model_layers_22_mlp_gate_proj/mean":-0.03466796875,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_66_mlp_down_proj/norm":30557.263671875,"train/tensor_act_model_layers_26_mlp_down_proj/std":0.435546875,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_99_mlp_down_proj/norm":30607.171875,"train/tensor_act_model_layers_96_mlp_gate_proj/min":-4.59375,"train/tensor_act_model_layers_0_mlp_gate_proj/max_abs":6.25,"train/tensor_act_model_layers_81_mlp_gate_proj/max":4.15625,"train/tensor_act_model_layers_39_mlp_down_proj/norm":30479.361328125,"train/tensor_act_model_layers_17_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_53_residual_post_attn/norm":433653.1875,"train/tensor_param_model_layers_3_mlp_up_proj_weight/max":0.10107421875,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_93_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_34_mlp_up_proj/min":-3.453125,"train/tensor_act_model_layers_33_residual_pre_attn/std":0.71484375,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_43_mlp_gate_proj/min":-3.859375,"train/tensor_act_model_layers_66_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_54_residual_post_mlp/range":5,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_42_residual_pre_attn/range":5,"train/tensor_act_model_layers_82_mlp_down_proj/std":0.5,"train/tensor_act_model_layers_69_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_66_residual_post_mlp/max":270,"train/tensor_act_model_layers_73_mlp_up_proj/norm":5389.455078125,"train/tensor_act_model_layers_27_mlp_up_proj/max":3.40625,"train/tensor_param_model_layers_50_mlp_up_proj_weight/range":0.2470703125,"train/tensor_act_model_layers_23_residual_post_mlp/norm":196250.90625,"train/tensor_act_model_layers_63_mlp_gate_proj/norm":5295.73828125,"train/tensor_act_model_layers_34_residual_pre_attn/min":134,"train/tensor_act_model_layers_24_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_18_mlp_up_proj/norm":5271.0947265625,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/min":-0.125,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_39_residual_post_attn/range":5,"train/tensor_act_model_layers_29_residual_pre_attn/norm":237010.703125,"train/tensor_act_model_layers_10_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_25_mlp_gate_proj/range":7.1875,"train/tensor_param_model_layers_63_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_20_mlp_down_proj/range":2.375,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_66_mlp_down_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_55_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_mlp/norm":8192,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_3_residual_pre_attn/max":12.4375,"train/tensor_act_model_layers_97_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_29_residual_post_mlp/mean":119.5,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_55_residual_post_mlp/min":221,"train/tensor_act_model_layers_99_mlp_gate_proj/norm":5512.70751953125,"train/tensor_act_model_layers_90_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_23_residual_post_attn/max_abs":94,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_66_mlp_gate_proj/norm":5235.94677734375,"train/tensor_act_model_layers_81_residual_pre_attn/range":6,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_70_mlp_down_proj/range":2.125,"train/tensor_act_model_layers_28_mlp_down_proj/min":14.1875,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_41_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_63_mlp_down_proj/norm":30598.126953125,"train/tensor_param_model_layers_81_mlp_up_proj_weight/std":0.040771484375,"train/tensor_act_model_layers_21_residual_post_attn/std":0.58203125,"train/tensor_act_model_layers_59_mlp_gate_proj/mean":0.134765625,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_77_mlp_gate_proj/range":7.6875,"train/tensor_param_model_layers_23_mlp_down_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_63_mlp_down_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_77_mlp_down_proj/mean":15,"train/tensor_act_model_layers_73_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_57_residual_pre_attn/mean":228,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_64_mlp_down_proj/range":2.625,"train/tensor_act_model_layers_42_residual_post_attn/norm":343525.96875,"train/tensor_act_model_layers_82_residual_post_mlp/range":6,"train/tensor_act_model_layers_21_mlp/min":4,"train/tensor_param_model_layers_33_mlp_down_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_76_mlp_down_proj_weight/range":0.2626953125,"train/tensor_act_model_layers_62_mlp_down_proj/max_abs":16.5,"train/tensor_act_model_layers_40_residual_pre_attn/norm":327142.15625,"train/tensor_param_model_layers_91_mlp_up_proj_weight/max":0.154296875,"train/tensor_param_model_layers_45_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_95_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_26_residual_post_attn/mean":104,"train/tensor_param_model_layers_70_mlp_up_proj_weight/norm":6.84375,"train/tensor_act_model_layers_27_residual_post_attn/norm":220738.359375,"train/tensor_act_model_layers_53_mlp_down_proj/norm":30581.79296875,"train/tensor_act_model_layers_94_residual_post_attn/range":6,"train/tensor_act_model_layers_62_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_53_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_19_residual_post_attn/norm":155293.640625,"train/tensor_act_model_layers_58_residual_pre_attn/max_abs":234,"train/tensor_act_model_layers_28_mlp_down_proj/max":16.375,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_73_residual_pre_attn/norm":597614.1875,"train/tensor_act_model_layers_2_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_param_model_layers_4_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp_gate_proj/max_abs":4.15625,"train/tensor_act_model_layers_3_mlp_down_proj/std":0.55859375,"train/tensor_act_model_layers_78_mlp_down_proj/mean":15,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/mean":0.00064849853515625,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_29_residual_pre_attn/max":118.5,"train/tensor_act_model_layers_11_residual_post_mlp/std":0.396484375,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/min":-0.1298828125,"train/tensor_param_model_layers_94_mlp_down_proj_weight/norm":8.4375,"train/tensor_act_model_layers_80_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_5_residual_post_mlp/std":0.23046875,"train/tensor_act_model_layers_16_residual_post_attn/mean":63.75,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_16_mlp_down_proj_weight/mean":-0.000789642333984375,"train/tensor_act_model_layers_47_mlp_up_proj/max":3.421875,"train/tensor_act_model_layers_65_mlp_down_proj/max_abs":16.25,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_59_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_residual_post_mlp/max_abs":49.5,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_45_mlp_up_proj/max":4.15625,"train/tensor_param_model_layers_52_mlp_up_proj_weight/range":0.2548828125,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_16_residual_pre_attn/min":62.5,"train/tensor_param_model_layers_28_mlp_down_proj_weight/std":0.03466796875,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/range":0.2568359375,"train/tensor_act_model_layers_76_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/norm":7.53125,"train/tensor_param_model_layers_37_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_5_mlp_down_proj/max_abs":16.125,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_92_mlp_up_proj_weight/norm":8.3125,"train/tensor_act_model_layers_92_mlp/min":4,"train/tensor_act_model_layers_32_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_21_residual_pre_attn/norm":171675.4375,"train/tensor_act_model_layers_92_mlp/max_abs":4,"train/tensor_param_model_layers_97_mlp_up_proj_weight/std":0.04736328125,"train/tensor_act_model_layers_27_mlp_gate_proj/mean":0.189453125,"train/tensor_act_model_layers_81_mlp_down_proj/max":16.5,"train/tensor_act_model_layers_50_residual_post_attn/min":197,"train/tensor_act_model_layers_23_mlp_gate_proj/norm":5469.81787109375,"train/tensor_act_model_layers_26_mlp_up_proj/mean":0.049072265625,"train/tensor_param_model_layers_52_mlp_down_proj_weight/max":0.126953125,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_36_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp_up_proj/range":7.28125,"train/tensor_act_model_layers_77_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_98_residual_pre_attn/max":394,"train/tensor_act_model_layers_0_mlp_down_proj/range":103.8125,"train/tensor_param_model_layers_68_mlp_down_proj_weight/mean":0.0003662109375,"train/tensor_act_model_layers_73_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_53_residual_pre_attn/min":209,"train/tensor_act_model_layers_26_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/mean":-0.000911712646484375,"train/tensor_act_model_layers_79_residual_pre_attn/max_abs":318,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_17_mlp_down_proj_weight/range":0.2392578125,"train/tensor_param_model_layers_59_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_40_residual_post_attn/std":0.80859375,"train/tensor_param_model_layers_61_mlp_down_proj_weight/std":0.035400390625,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/max":0.1318359375,"train/tensor_param_model_layers_42_mlp_down_proj_weight/std":0.034423828125,"train/tensor_param_model_layers_82_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_residual_post_attn/max":202,"train/tensor_act_model_layers_93_mlp/mean":4,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_81_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_92_mlp_up_proj_weight/range":0.2763671875,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/min":-0.16796875,"train/tensor_act_model_layers_81_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_48_mlp_gate_proj/min":-3.59375,"train/tensor_act_model_layers_83_residual_post_mlp/range":6,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp/range":0,"train/tensor_act_model_layers_68_mlp_up_proj/max":4.09375,"train/tensor_act_model_layers_44_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_19_mlp_up_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_80_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_2_mlp_up_proj_weight/max":0.11962890625,"train/tensor_act_model_layers_65_mlp/mean":4,"train/tensor_act_model_layers_68_mlp/norm":8192,"train/tensor_act_model_layers_9_residual_post_mlp/mean":40,"train/tensor_act_model_layers_0_residual_post_attn/mean":0.00174713134765625,"train/tensor_act_model_layers_86_mlp/max":4,"train/tensor_act_model_layers_40_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_79_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_94_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_21_mlp_up_proj/std":1.859375,"train/tensor_act_model_layers_34_residual_post_mlp/std":0.7109375,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_38_mlp_up_proj_weight/norm":6.3125,"train/tensor_param_model_layers_96_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_67_mlp_down_proj/norm":30510.8984375,"train/tensor_act_model_layers_96_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_38_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_69_residual_post_attn/range":6,"train/tensor_act_model_layers_86_mlp_up_proj/max":4.0625,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/max_abs":0.134765625,"train/tensor_act_model_layers_18_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_80_mlp_gate_proj/min":-4.21875,"train/tensor_act_model_layers_19_mlp/mean":4,"train/tensor_act_model_layers_70_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_residual_pre_attn/mean":252,"train/tensor_act_model_layers_85_mlp_down_proj/mean":15,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/range":0.2919921875,"train/tensor_act_model_layers_51_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_80_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_16_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/max":0.11767578125,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/max":0.12255859375,"train/tensor_act_model_layers_24_mlp/max_abs":4,"train/tensor_param_model_layers_14_mlp_down_proj_weight/mean":-0.000591278076171875,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_72_mlp_up_proj/min":-3.984375,"train/tensor_act_model_layers_62_mlp_down_proj/range":2.1875,"train/tensor_param_model_layers_94_mlp_up_proj_weight/max_abs":0.1435546875,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_73_residual_post_attn/mean":292,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_88_mlp_down_proj/range":3.125,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_12_residual_post_attn/mean":48,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/range":0.251953125,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_50_mlp_up_proj/max_abs":3.90625,"train/tensor_act_model_layers_48_residual_post_attn/mean":192,"train/tensor_act_model_layers_79_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_65_mlp_down_proj_weight/range":0.24755859375,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_21_residual_post_mlp/mean":88,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_15_mlp/range":0,"train/tensor_param_model_layers_84_mlp_down_proj_weight/range":0.2763671875,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_84_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_mlp/max_abs":4,"train/tensor_act_model_layers_56_residual_pre_attn/std":0.8203125,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_15_mlp_down_proj/range":1.8125,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_89_mlp_gate_proj/norm":5219.42138671875,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_50_mlp_up_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_30_residual_pre_attn/max_abs":122.5,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_mlp/max_abs":4,"train/tensor_act_model_layers_0_residual_post_mlp/max_abs":4.375,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_59_residual_post_attn/std":0.8359375,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_55_mlp_down_proj/std":0.341796875,"train/tensor_param_model_layers_97_mlp_down_proj_weight/norm":8.5625,"train/tensor_act_model_layers_62_mlp/std":0,"train/tensor_act_model_layers_47_mlp_gate_proj/std":1.8828125,"train/tensor_act_model_layers_2_residual_post_attn/frac_near_user_limit":0,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/max":0.1337890625,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_mlp/norm":8192,"train/tensor_param_model_layers_1_mlp_up_proj_weight/norm":6.09375,"train/tensor_param_model_layers_71_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_50_mlp_down_proj/std":0.369140625,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_84_mlp_down_proj/norm":30624.697265625,"train/tensor_act_model_layers_17_mlp_up_proj/max_abs":3.890625,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_96_residual_post_mlp/range":6,"train/tensor_act_model_layers_60_mlp/norm":8192,"train/tensor_param_model_layers_84_mlp_down_proj_weight/max_abs":0.142578125,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_8_residual_post_attn/mean":32,"train/tensor_param_model_layers_15_mlp_down_proj_weight/mean":-0.0002593994140625,"train/tensor_act_model_layers_64_mlp_gate_proj/max":3.71875,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/norm":6.21875,"train/tensor_act_model_layers_49_mlp_gate_proj/mean":0.11669921875,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/std":0.04833984375,"train/tensor_act_model_layers_9_mlp_gate_proj/norm":5338.8798828125,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/mean":-0.00015544891357421875,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_57_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_35_mlp_gate_proj/max_abs":3.640625,"train/tensor_act_model_layers_88_residual_pre_attn/max_abs":354,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_22_mlp_down_proj_weight/min":-0.123046875,"train/tensor_act_model_layers_39_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_39_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_8_mlp_down_proj/range":1.875,"train/tensor_act_model_layers_65_residual_post_mlp/min":260,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_83_mlp/max":4,"train/tensor_act_model_layers_30_residual_post_mlp/max":126.5,"train/tensor_act_model_layers_36_residual_pre_attn/min":141,"train/tensor_act_model_layers_26_residual_pre_attn/max_abs":106,"train/tensor_act_model_layers_16_mlp_gate_proj/range":7.328125,"train/tensor_act_model_layers_41_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_mlp_down_proj/norm":30624.732421875,"train/tensor_param_model_layers_23_mlp_up_proj_weight/min":-0.125,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_70_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_25_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_95_mlp/mean":4,"train/tensor_act_model_layers_85_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_21_residual_post_mlp/norm":179867.25,"train/tensor_act_model_layers_74_mlp_down_proj/std":0.46484375,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/min":-0.126953125,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_65_residual_post_attn/mean":260,"train/tensor_act_model_layers_83_mlp_up_proj/max":4.03125,"train/tensor_act_model_layers_72_residual_post_attn/max_abs":290,"train/tensor_param_model_layers_85_mlp_down_proj_weight/max":0.1318359375,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/min":-0.1357421875,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_95_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_40_residual_post_mlp/min":161,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/norm":7.96875,"train/tensor_act_model_layers_27_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_residual_pre_attn/max_abs":8.4375,"train/tensor_act_model_layers_40_mlp/min":4,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_51_mlp_down_proj_weight/min":-0.12890625,"train/tensor_act_model_layers_36_mlp_gate_proj/max_abs":3.84375,"train/tensor_act_model_layers_0_residual_post_attn/max":0.373046875,"train/tensor_act_model_layers_51_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_78_residual_post_mlp/min":312,"train/tensor_act_model_layers_16_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_76_mlp_gate_proj/min":-3.8125,"train/tensor_act_model_layers_43_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp/std":0,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_82_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_83_mlp/norm":8192,"train/tensor_param_model_layers_94_mlp_up_proj_weight/mean":-0.0015869140625,"train/tensor_act_model_layers_91_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_57_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/max":0.119140625,"train/tensor_act_model_layers_32_residual_post_attn/max_abs":130,"train/tensor_act_model_layers_43_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_0_residual_pre_attn/mean":0.0017547607421875,"train/tensor_param_model_layers_68_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_85_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_32_residual_post_mlp/std":0.71484375,"train/tensor_act_model_layers_94_mlp_down_proj/min":14,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_71_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/norm":6.46875,"train/tensor_act_model_layers_73_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_25_mlp_gate_proj/max_abs":3.6875,"train/tensor_act_model_layers_1_residual_post_attn/mean":4,"train/tensor_param_model_layers_57_mlp_up_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_47_residual_post_mlp/min":189,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/range":0.25927734375,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_46_mlp_down_proj_weight/max":0.125,"train/tensor_param_model_layers_14_mlp_down_proj_weight/std":0.033203125,"train/tensor_act_model_layers_32_residual_post_attn/range":4.5,"train/tensor_param_model_layers_76_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/mean":-0.00183868408203125,"train/tensor_act_model_layers_39_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_44_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp/std":0,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_30_residual_post_attn/range":4.5,"train/tensor_param_model_layers_90_mlp_up_proj_weight/std":0.044921875,"train/tensor_act_model_layers_88_mlp/mean":4,"train/tensor_act_model_layers_23_residual_pre_attn/max_abs":94,"train/tensor_act_model_layers_89_residual_post_attn/max_abs":358,"train/tensor_param_model_layers_55_mlp_up_proj_weight/norm":6.4375,"train/tensor_param_model_layers_6_mlp_up_proj_weight/max":0.11279296875,"train/tensor_param_model_layers_6_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_64_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_23_residual_post_attn/norm":188059.078125,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/mean":-0.00086212158203125,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_51_residual_post_mlp/max_abs":210,"train/tensor_act_model_layers_94_mlp_down_proj/max":18.625,"train/tensor_act_model_layers_74_mlp/max_abs":4,"train/tensor_act_model_layers_71_mlp_gate_proj/mean":-0.036865234375,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/max":0.1171875,"train/tensor_param_model_layers_94_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_25_mlp_down_proj/norm":30421.5625,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_81_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_3_mlp_up_proj/max":3.5,"train/tensor_act_model_layers_61_mlp_up_proj/max":3.40625,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_7_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_78_residual_post_mlp/max_abs":318,"train/tensor_act_model_layers_33_mlp_gate_proj/min":-4.125,"train/tensor_param_model_layers_88_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_30_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_95_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_9_residual_pre_attn/norm":73676.0625,"train/tensor_param_model_layers_1_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_60_mlp_up_proj/std":1.8515625,"train/tensor_act_model_layers_29_residual_post_attn/max_abs":118.5,"train/tensor_param_model_layers_94_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_41_residual_post_attn/mean":164,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_34_mlp_up_proj/mean":-0.0172119140625,"train/tensor_act_model_layers_65_mlp/max":4,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_13_mlp/range":0,"train/tensor_act_model_layers_2_residual_post_mlp/range":1.0625,"train/tensor_act_model_layers_59_mlp_up_proj/std":1.8203125,"train/tensor_act_model_layers_48_mlp_down_proj/norm":30463.28125,"train/tensor_param_model_layers_32_mlp_down_proj_weight/mean":-0.000576019287109375,"train/tensor_act_model_layers_13_mlp_up_proj/mean":-0.09228515625,"train/tensor_param_model_layers_4_mlp_down_proj_weight/std":0.0303955078125,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/max_abs":408,"train/tensor_act_model_layers_21_residual_post_mlp/min":86,"train/tensor_act_model_layers_46_mlp_down_proj/max":16.125,"train/tensor_param_model_layers_53_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_39_mlp_up_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_20_mlp/std":0,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_99_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_87_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_88_residual_post_attn/max_abs":354,"train/tensor_act_model_layers_75_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_69_residual_pre_attn/range":6,"train/tensor_act_model_layers_82_residual_pre_attn/mean":328,"train/tensor_param_model_layers_27_mlp_up_proj_weight/mean":-0.000774383544921875,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_98_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_72_residual_post_attn/range":6,"train/tensor_act_model_layers_88_mlp_up_proj/max":3.796875,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_82_mlp_up_proj/mean":0.004913330078125,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_67_mlp_down_proj_weight/norm":6.5625,"train/tensor_act_model_layers_58_mlp/min":4,"train/tensor_act_model_layers_56_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_residual_post_mlp/std":0.81640625,"train/tensor_act_model_layers_49_mlp/range":0,"train/tensor_act_model_layers_81_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_60_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_mlp/norm":491012.90625,"train/tensor_act_model_layers_13_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp_gate_proj/mean":0.0135498046875,"train/tensor_act_model_layers_39_residual_post_attn/max":158,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_52_mlp/std":0,"train/tensor_act_model_layers_42_residual_pre_attn/min":165,"train/tensor_act_model_layers_63_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_88_mlp_down_proj/norm":30813.1640625,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/norm":6.5625,"train/tensor_act_model_layers_98_mlp_gate_proj/std":1.859375,"train/tensor_act_model_layers_58_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_mlp_up_proj/max_abs":3.96875,"train/tensor_act_model_layers_92_residual_post_attn/mean":368,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_43_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_50_mlp/std":0,"train/tensor_act_model_layers_39_residual_post_attn/min":153,"train/tensor_param_model_layers_65_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_35_residual_post_attn/mean":140,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/max":0.1591796875,"train/tensor_param_model_layers_33_mlp_down_proj_weight/min":-0.126953125,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/max":0.1337890625,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/norm":6.9375,"train/tensor_act_model_layers_87_mlp/range":0,"train/tensor_act_model_layers_21_mlp/norm":8192,"train/tensor_act_model_layers_95_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/norm":7.25,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_17_mlp_gate_proj/max":3.75,"train/tensor_param_model_layers_52_mlp_up_proj_weight/min":-0.125,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_51_mlp/std":0,"train/tensor_param_model_layers_25_mlp_down_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_98_residual_pre_attn/mean":392,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_98_mlp_gate_proj/norm":5423.919921875,"train/tensor_act_model_layers_27_mlp_up_proj/range":7.9375,"train/tensor_act_model_layers_78_residual_pre_attn/range":6,"train/tensor_act_model_layers_38_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_93_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_35_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_down_proj_weight/max_abs":0.1201171875,"train/tensor_act_model_layers_21_mlp_down_proj/range":2.375,"train/tensor_act_model_layers_46_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_81_mlp_up_proj_weight/mean":0.00078582763671875,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_31_residual_post_attn/range":4.5,"train/tensor_param_model_layers_15_mlp_down_proj_weight/max":0.125,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/mean":-0.0004558563232421875,"train/tensor_param_model_layers_59_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_37_mlp/mean":4,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_21_mlp_up_proj_weight/min":-0.12451171875,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_73_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_15_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_33_mlp_up_proj_weight/max":0.12255859375,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/min":-0.140625,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/norm":8.5625,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_post_mlp/max_abs":222,"train/tensor_act_model_layers_73_residual_post_mlp/norm":605806.125,"train/tensor_act_model_layers_50_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_40_residual_post_mlp/norm":335334.0625,"train/tensor_act_model_layers_73_mlp_down_proj/range":1.8125,"train/tensor_act_model_layers_89_residual_post_attn/min":352,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_98_mlp_up_proj_weight/min":-0.1630859375,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_12_mlp/norm":8192,"train/tensor_param_model_layers_13_mlp_up_proj_weight/min":-0.12060546875,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_9_residual_post_attn/std":0.373046875,"train/tensor_act_model_layers_21_residual_pre_attn/range":4,"train/tensor_act_model_layers_21_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/norm":6.625,"train/tensor_act_model_layers_2_mlp_up_proj/std":1.8671875,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_86_residual_pre_attn/mean":344,"train/tensor_act_model_layers_8_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_46_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_20_mlp_down_proj_weight/mean":0.000186920166015625,"train/tensor_act_model_layers_65_mlp_gate_proj/min":-3.75,"train/tensor_act_model_layers_29_mlp_up_proj/max":3.25,"train/tensor_act_model_layers_51_residual_pre_attn/std":0.81640625,"train/tensor_act_model_layers_98_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/range":770,"train/tensor_act_model_layers_48_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_residual_post_mlp/max":98,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_19_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_66_mlp_down_proj_weight/min":-0.12890625,"train/tensor_act_model_layers_30_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/std":0.03759765625,"train/tensor_act_model_layers_44_residual_post_mlp/std":0.80859375,"train/tensor_act_model_layers_53_mlp/range":0,"train/tensor_act_model_layers_89_residual_post_attn/max":358,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_20_residual_pre_attn/range":4,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_86_mlp_up_proj_weight/range":0.2841796875,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_88_mlp_up_proj_weight/range":0.3037109375,"train/tensor_act_model_layers_70_mlp/mean":4,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_3_mlp_up_proj_weight/std":0.03173828125,"train/tensor_act_model_layers_92_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_76_mlp_down_proj/range":2.1875,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_99_residual_post_mlp/max":402,"train/tensor_act_model_layers_70_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_82_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_76_mlp_down_proj_weight/std":0.03857421875,"train/tensor_act_model_layers_1_mlp_up_proj/norm":5376.38818359375,"train/tensor_act_model_layers_46_mlp_gate_proj/max_abs":3.59375,"train/tensor_act_model_layers_84_mlp_down_proj/max":16.25,"train/tensor_param_model_layers_5_mlp_up_proj_weight/mean":0.000736236572265625,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/mean":-0.0003910064697265625,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_11_mlp_up_proj/max_abs":3.46875,"train/tensor_param_model_layers_43_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_30_mlp_gate_proj/max_abs":3.78125,"train/tensor_act_model_layers_60_mlp_down_proj/max":16.625,"train/tensor_act_model_layers_40_mlp/norm":8192,"train/tensor_act_model_layers_93_residual_pre_attn/min":368,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/min":-0.1220703125,"train/tensor_param_model_layers_55_mlp_up_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_14_residual_post_attn/norm":114426.2109375,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_27_mlp_gate_proj/range":7,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/range":0.2626953125,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_63_mlp/std":0,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/max":0.16796875,"train/tensor_act_model_layers_45_mlp_down_proj/max_abs":16.625,"train/tensor_act_model_layers_19_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/max_abs":0.1259765625,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_mlp_up_proj/mean":0.2197265625,"train/tensor_act_model_layers_49_residual_post_mlp/min":197,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_residual_post_mlp/max_abs":150,"train/tensor_act_model_layers_73_residual_post_mlp/mean":296,"train/tensor_act_model_layers_40_residual_post_mlp/mean":164,"train/tensor_act_model_layers_96_residual_post_mlp/min":384,"train/tensor_param_model_layers_87_mlp_down_proj_weight/min":-0.130859375,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_3_residual_pre_attn/max_abs":12.4375,"train/tensor_act_model_layers_20_mlp/range":0,"train/tensor_param_model_layers_93_mlp_down_proj_weight/norm":8.4375,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/min":-0.1474609375,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/range":0.255859375,"train/tensor_act_model_layers_45_mlp/min":4,"train/tensor_act_model_layers_62_residual_pre_attn/mean":248,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_up_proj_weight/mean":-0.00013637542724609375,"train/tensor_act_model_layers_39_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_69_mlp_gate_proj/max_abs":4,"train/tensor_param_model_layers_10_mlp_down_proj_weight/std":0.033447265625,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_69_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_22_mlp_up_proj_weight/std":0.03466796875,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_84_mlp_up_proj/max":3.75,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_84_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_60_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_75_mlp_up_proj_weight/mean":7.361173629760742e-06,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/max_abs":0.1552734375,"train/tensor_act_model_layers_24_mlp_down_proj/range":1.5625,"train/tensor_act_model_layers_98_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_78_mlp_up_proj/max_abs":4.03125,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/mean":0.0003757476806640625,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_93_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_down_proj_weight/min":-0.125,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_58_residual_post_attn/norm":474629.03125,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_11_mlp_down_proj_weight/norm":6,"train/tensor_act_model_layers_14_mlp_up_proj/min":-3.671875,"train/tensor_act_model_layers_7_mlp_gate_proj/max_abs":3.59375,"train/tensor_act_model_layers_27_mlp_gate_proj/max":3.484375,"train/tensor_act_model_layers_84_mlp_down_proj/min":14.1875,"train/tensor_param_model_layers_2_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_17_mlp_up_proj/range":7.59375,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_96_mlp/max_abs":4,"train/tensor_act_model_layers_59_residual_pre_attn/norm":482821,"train/tensor_param_model_layers_4_mlp_down_proj_weight/mean":0.00024127960205078125,"train/tensor_act_model_layers_17_mlp_gate_proj/mean":0.1591796875,"train/tensor_act_model_layers_39_mlp_gate_proj/norm":5284.04541015625,"train/tensor_param_model_layers_69_mlp_down_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_84_mlp_down_proj_weight/min":-0.142578125,"train/tensor_act_model_layers_64_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp/mean":4,"train/tensor_act_model_layers_59_residual_post_mlp/mean":240,"train/tensor_param_model_layers_52_mlp_down_proj_weight/norm":6.28125,"train/tensor_act_model_layers_53_residual_pre_attn/range":5,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_52_residual_post_attn/norm":425461.25,"train/tensor_param_model_layers_55_mlp_up_proj_weight/std":0.03564453125,"train/tensor_act_model_layers_63_residual_post_attn/norm":515541.3125,"train/tensor_act_model_layers_46_residual_pre_attn/min":181,"train/tensor_param_model_layers_62_mlp_up_proj_weight/max":0.1240234375,"train/tensor_act_model_layers_17_residual_post_mlp/std":0.5625,"train/tensor_act_model_layers_63_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_99_residual_pre_attn/max_abs":398,"train/tensor_param_model_layers_96_mlp_down_proj_weight/min":-0.1416015625,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_31_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/max_abs":0.1220703125,"train/tensor_act_model_layers_48_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_67_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_6_mlp_down_proj/max_abs":16.375,"train/tensor_param_model_layers_89_mlp_down_proj_weight/range":0.28125,"train/tensor_act_model_layers_39_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_23_mlp_up_proj/norm":5388.8330078125,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_18_mlp_down_proj/range":1.625,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_35_mlp_up_proj/max":3.65625,"train/tensor_act_model_layers_49_mlp_down_proj/std":0.41015625,"train/tensor_act_model_layers_41_mlp/norm":8192,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_99_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_44_mlp_down_proj_weight/std":0.03466796875,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_53_mlp_down_proj/std":0.4140625,"train/tensor_act_model_layers_16_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_45_mlp_gate_proj/std":1.859375,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_30_mlp_gate_proj/min":-3.71875,"train/tensor_act_model_layers_18_mlp_up_proj/max":3.28125,"train/tensor_act_model_layers_39_mlp_gate_proj/std":1.8125,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_55_mlp/max_abs":4,"train/tensor_act_model_layers_26_residual_pre_attn/std":0.62109375,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_mlp_down_proj/max_abs":16.25,"train/tensor_act_model_layers_61_residual_post_attn/range":5,"train/tensor_act_model_layers_87_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_72_mlp_up_proj_weight/min":-0.130859375,"train/tensor_act_model_layers_1_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_38_mlp_up_proj/max":3.703125,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_91_mlp_down_proj_weight/norm":8.125,"train/tensor_act_model_layers_85_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_75_mlp_up_proj/min":-3.875,"train/tensor_param_model_layers_79_mlp_down_proj_weight/min":-0.142578125,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/max_abs":0.1474609375,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_72_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_32_mlp_down_proj/norm":30434.6015625,"train/tensor_act_model_layers_10_residual_post_mlp/std":0.39453125,"train/tensor_act_model_layers_37_mlp/max":4,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_73_mlp_down_proj/norm":30515.3046875,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_99_mlp_up_proj_weight/max":0.1513671875,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_67_mlp_up_proj/min":-3.578125,"train/tensor_act_model_layers_3_mlp_down_proj/max":18.375,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_88_mlp_gate_proj/max_abs":4.03125,"train/tensor_act_model_layers_84_mlp/norm":8192,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_20_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_13_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_49_residual_post_attn/max_abs":198,"train/tensor_act_model_layers_49_residual_post_mlp/max":202,"train/tensor_param_model_layers_23_mlp_up_proj_weight/range":0.25390625,"train/tensor_act_model_layers_95_mlp_up_proj/range":8.65625,"train/tensor_act_model_layers_40_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_58_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_12_residual_post_mlp/std":0.4296875,"train/tensor_param_model_layers_18_mlp_down_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_92_mlp_down_proj_weight/max":0.134765625,"train/tensor_act_model_layers_91_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_10_residual_post_mlp/range":3,"train/tensor_act_model_layers_23_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_mlp_gate_proj/max_abs":4.5,"train/tensor_act_model_layers_80_mlp_gate_proj/range":8.125,"train/tensor_act_model_layers_78_mlp_down_proj/std":0.4609375,"train/tensor_param_model_layers_80_mlp_down_proj_weight/norm":7.15625,"train/tensor_act_model_layers_96_residual_post_mlp/max":390,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_24_mlp_down_proj_weight/norm":6.25,"train/tensor_param_model_layers_95_mlp_up_proj_weight/norm":8.5625,"train/tensor_act_model_layers_31_mlp_down_proj/norm":30420.634765625,"train/tensor_param_model_layers_50_mlp_up_proj_weight/norm":6.3125,"train/tensor_param_model_layers_21_mlp_down_proj_weight/max":0.1201171875,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_89_residual_post_mlp/mean":360,"train/tensor_act_model_layers_39_residual_post_attn/max_abs":158,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_90_residual_post_mlp/max_abs":366,"train/tensor_act_model_layers_14_mlp_gate_proj/mean":-0.26171875,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_47_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_99_mlp_up_proj/max_abs":4.6875,"train/tensor_param_model_layers_27_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_75_mlp_down_proj_weight/min":-0.1337890625,"train/tensor_param_model_layers_51_mlp_up_proj_weight/min":-0.11767578125,"train/tensor_act_model_layers_85_mlp_gate_proj/norm":5244.75439453125,"train/tensor_act_model_layers_25_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_18_residual_post_attn/mean":72,"train/tensor_act_model_layers_68_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_39_residual_pre_attn/norm":318950.25,"train/tensor_act_model_layers_94_mlp/mean":4,"train/tensor_act_model_layers_51_mlp_gate_proj/norm":5419.55517578125,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_33_mlp_gate_proj/range":7.71875,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_83_residual_post_mlp/max_abs":338,"train/tensor_act_model_layers_95_mlp/norm":8192,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_91_mlp/max":4,"train/tensor_act_model_layers_81_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_25_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_82_mlp_down_proj_weight/min":-0.1328125,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_80_mlp_down_proj/norm":30647.25,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_53_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_44_mlp_up_proj/min":-4.15625,"train/tensor_act_model_layers_11_residual_post_attn/range":3,"train/tensor_param_model_layers_17_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_34_mlp/std":0,"train/tensor_param_model_layers_89_mlp_down_proj_weight/mean":0.00133514404296875,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/max_abs":0.11962890625,"train/tensor_param_model_layers_27_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_gate_proj/max":3.28125,"train/tensor_act_model_layers_12_residual_post_mlp/max_abs":53.5,"train/tensor_param_model_layers_76_mlp_up_proj_weight/range":0.26171875,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_50_mlp_gate_proj/max_abs":3.765625,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_71_mlp_up_proj/std":1.84375,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_89_mlp_down_proj/mean":15.0625,"train/tensor_act_model_layers_35_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_9_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_29_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_53_mlp_up_proj_weight/min":-0.12255859375,"train/tensor_act_model_layers_7_residual_post_attn/max":29.375,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_58_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_22_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_1_residual_post_attn/max_abs":4.4375,"train/tensor_act_model_layers_35_residual_post_attn/range":5,"train/tensor_param_model_layers_46_mlp_up_proj_weight/max":0.130859375,"train/tensor_act_model_layers_4_residual_pre_attn/max":16.5,"train/tensor_act_model_layers_27_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_65_mlp_up_proj/max_abs":3.6875,"train/tensor_act_model_layers_81_mlp_up_proj/std":1.7890625,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_58_residual_post_mlp/std":0.8359375,"train/tensor_param_model_layers_98_mlp_down_proj_weight/min":-0.1513671875,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/range":0.2841796875,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp/std":0,"train/tensor_act_model_layers_14_residual_post_mlp/std":0.494140625,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_8_mlp_down_proj_weight/max_abs":0.123046875,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_49_mlp_up_proj_weight/max_abs":0.12255859375,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_residual_pre_attn/mean":28,"train/tensor_param_model_layers_86_mlp_up_proj_weight/min":-0.13671875,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_56_mlp_up_proj_weight/range":0.2666015625,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/mean":0.00019168853759765625,"train/tensor_act_model_layers_36_mlp/mean":4,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_20_mlp/max_abs":4,"train/tensor_act_model_layers_74_residual_post_mlp/range":6,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_96_mlp_up_proj_weight/min":-0.1435546875,"train/tensor_act_model_layers_34_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_56_mlp_up_proj_weight/max_abs":0.134765625,"train/tensor_act_model_layers_91_residual_post_mlp/max_abs":370,"train/tensor_act_model_layers_89_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/std":0.03564453125,"train/tensor_act_model_layers_23_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_16_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_39_residual_post_mlp/mean":160,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_13_mlp_gate_proj/mean":0.0625,"train/tensor_act_model_layers_2_mlp_gate_proj/std":1.8359375,"train/tensor_param_model_layers_37_mlp_down_proj_weight/max_abs":0.1259765625,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/range":0.2763671875,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_89_residual_pre_attn/norm":728685.8125,"train/tensor_param_model_layers_81_mlp_down_proj_weight/norm":7.34375,"train/tensor_act_model_layers_76_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_55_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_mlp/range":0,"train/tensor_param_model_layers_49_mlp_up_proj_weight/mean":0.001068115234375,"train/tensor_act_model_layers_35_mlp_down_proj/max_abs":16.5,"train/tensor_act_model_layers_84_residual_pre_attn/max":338,"train/tensor_act_model_layers_58_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_29_mlp_gate_proj/mean":0.0751953125,"train/tensor_act_model_layers_92_mlp_up_proj/std":1.8359375,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_80_residual_pre_attn/max_abs":322,"train/tensor_act_model_layers_77_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_42_residual_post_mlp/max_abs":174,"train/tensor_act_model_layers_4_residual_post_attn/norm":32747.095703125,"train/tensor_param_model_layers_40_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_60_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_attn/min":233,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_48_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_95_mlp_gate_proj/max":4.5625,"train/tensor_act_model_layers_15_residual_pre_attn/mean":59.75,"train/tensor_act_model_layers_41_residual_pre_attn/max":166,"train/tensor_act_model_layers_52_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_31_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_86_mlp_gate_proj/max_abs":3.875,"train/tensor_param_model_layers_15_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_83_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_35_mlp/min":4,"train/tensor_param_model_layers_10_mlp_down_proj_weight/max_abs":0.12451171875,"train/tensor_act_model_layers_72_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_43_mlp_up_proj_weight/norm":6.25,"train/tensor_param_model_layers_17_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_69_mlp_up_proj/norm":5384.93994140625,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_2_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_64_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/max_abs":0.130859375,"train/tensor_act_model_layers_5_mlp/range":0,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_70_residual_post_attn/max_abs":282,"train/tensor_param_model_layers_16_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/norm":5.9375,"train/tensor_act_model_layers_2_mlp_up_proj/norm":5398.28955078125,"train/tensor_act_model_layers_21_residual_post_mlp/max":90,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_33_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_11_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/max":0.126953125,"train/tensor_param_model_layers_12_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_38_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/max_abs":0.11962890625,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_25_mlp_up_proj/max":3.34375,"train/tensor_act_model_layers_28_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_64_mlp_down_proj_weight/std":0.035400390625,"train/tensor_param_model_layers_47_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp_up_proj/mean":0.0306396484375,"train/tensor_act_model_layers_59_mlp_down_proj/std":0.404296875,"train/tensor_act_model_layers_42_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_mlp_gate_proj/std":1.8203125,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_60_mlp_up_proj_weight/min":-0.1259765625,"train/tensor_act_model_layers_81_mlp_gate_proj/norm":5251.146484375,"train/tensor_act_model_layers_55_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_mlp_gate_proj/mean":0.035400390625,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_50_mlp_down_proj_weight/range":0.25,"train/tensor_act_model_layers_24_residual_pre_attn/max":98,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_50_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_4_mlp_down_proj_weight/norm":5.5,"train/tensor_act_model_layers_68_residual_pre_attn/range":6,"train/tensor_act_model_layers_89_mlp/min":4,"train/tensor_param_model_layers_35_mlp_down_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_62_mlp_up_proj/std":1.859375,"train/tensor_param_model_layers_65_mlp_down_proj_weight/min":-0.1279296875,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_74_mlp/min":4,"train/tensor_act_model_layers_39_residual_post_mlp/norm":327142.15625,"train/tensor_param_model_layers_82_mlp_down_proj_weight/mean":-8.869171142578125e-05,"train/tensor_param_model_layers_6_mlp_up_proj_weight/norm":5.9375,"train/tensor_act_model_layers_70_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp_up_proj/mean":-0.0260009765625,"train/tensor_param_model_layers_72_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_47_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_81_residual_post_attn/range":6,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_44_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_39_residual_pre_attn/min":153,"train/tensor_param_model_layers_64_mlp_up_proj_weight/std":0.036376953125,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_16_residual_pre_attn/mean":63.75,"train/tensor_param_model_layers_12_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_attn/max":238,"train/tensor_act_model_layers_36_mlp_up_proj/range":7.140625,"train/tensor_act_model_layers_76_mlp/range":0,"train/tensor_act_model_layers_66_mlp_up_proj/mean":-0.1640625,"train/tensor_act_model_layers_85_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_76_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_52_mlp_down_proj/max":16.125,"train/tensor_param_model_layers_90_mlp_up_proj_weight/range":0.279296875,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_residual_post_attn/mean":176,"train/tensor_param_model_layers_31_mlp_down_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_11_mlp_up_proj_weight/norm":6.1875,"train/tensor_act_model_layers_97_residual_pre_attn/max":390,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_35_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_2_mlp_up_proj_weight/max_abs":0.11962890625,"train/tensor_act_model_layers_30_mlp_up_proj/std":1.8359375,"train/tensor_param_model_layers_75_mlp_up_proj_weight/min":-0.130859375,"train/tensor_act_model_layers_52_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_53_mlp/max_abs":4,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_80_mlp_up_proj/std":1.8203125,"train/tensor_param_model_layers_40_mlp_up_proj_weight/range":0.25,"train/tensor_act_model_layers_50_residual_pre_attn/min":197,"train/tensor_act_model_layers_7_mlp/mean":4,"train/tensor_act_model_layers_76_mlp_up_proj/min":-3.671875,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/max_abs":0.12451171875,"train/tensor_act_model_layers_51_mlp_up_proj/max":3.625,"train/tensor_act_model_layers_24_residual_pre_attn/mean":96,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_11_mlp_down_proj_weight/max":0.111328125,"train/tensor_act_model_layers_37_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/norm":6.46875,"train/tensor_act_model_layers_5_mlp_up_proj/norm":5226.37060546875,"train/tensor_act_model_layers_63_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_mlp_gate_proj/min":-3.53125,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/mean":-0.001495361328125,"train/tensor_act_model_layers_15_residual_post_mlp/range":3,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_21_mlp_up_proj/norm":5388.234375,"train/tensor_act_model_layers_38_residual_pre_attn/range":5,"train/tensor_param_model_layers_5_mlp_down_proj_weight/range":0.22705078125,"train/tensor_act_model_layers_85_mlp_gate_proj/max_abs":4.21875,"train/tensor_act_model_layers_73_residual_post_attn/min":288,"train/tensor_act_model_layers_38_mlp_down_proj/std":0.35546875,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/mean":-0.00113677978515625,"train/tensor_act_model_layers_30_residual_post_attn/mean":119.5,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/max":0.1318359375,"train/tensor_act_model_layers_40_residual_post_attn/min":157,"train/tensor_act_model_layers_65_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_89_residual_post_mlp/norm":736877.8125,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_73_residual_post_attn/max":294,"train/tensor_param_model_layers_59_mlp_up_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_3_residual_pre_attn/std":0.1357421875,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_79_residual_post_mlp/std":0.7578125,"train_loss":2.1711830139160155,"train/tensor_act_model_layers_40_residual_post_attn/max":162,"train/tensor_act_model_layers_59_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_37_mlp_down_proj/max":16.375,"train/tensor_act_model_layers_53_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_27_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_31_mlp_up_proj/max_abs":3.734375,"train/tensor_param_model_layers_73_mlp_up_proj_weight/max":0.1279296875,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_24_residual_post_mlp/max_abs":102,"train/tensor_act_model_layers_50_residual_post_attn/std":0.81640625,"train/tensor_act_model_layers_57_residual_post_mlp/max_abs":234,"train/tensor_act_model_layers_99_residual_post_mlp/min":396,"train/tensor_act_model_layers_13_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_53_mlp_up_proj/min":-3.75,"train/tensor_act_model_layers_13_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_88_residual_post_mlp/range":6,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_48_mlp_up_proj_weight/range":0.2431640625,"train/tensor_param_model_layers_78_mlp_down_proj_weight/norm":7.03125,"train/tensor_param_model_layers_89_mlp_up_proj_weight/norm":8.125,"train/tensor_act_model_layers_77_mlp_down_proj/max_abs":17.25,"train/tensor_act_model_layers_24_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/min":-0.125,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_23_residual_post_mlp/min":94,"train/tensor_act_model_layers_51_mlp_up_proj/norm":5308.59619140625,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_33_mlp/mean":4,"train/tensor_param_model_layers_51_mlp_down_proj_weight/mean":-0.00138092041015625,"train/tensor_act_model_layers_85_mlp/std":0,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_41_mlp_down_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_54_residual_pre_attn/max":218,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/range":0.22314453125,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/mean":-0.000331878662109375,"train/tensor_param_model_layers_3_mlp_up_proj_weight/max_abs":0.1064453125,"train/tensor_act_model_layers_54_mlp/std":0,"train/tensor_act_model_layers_8_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_57_residual_post_mlp/std":0.8359375,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/std":0.03173828125,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_mlp_gate_proj/mean":0.1650390625,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/max_abs":0.1318359375,"train/tensor_act_model_layers_10_mlp/range":0,"train/tensor_param_model_layers_49_mlp_down_proj_weight/max":0.12353515625,"train/tensor_act_model_layers_30_residual_post_mlp/min":122,"train/tensor_act_model_layers_90_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp/norm":8192,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_65_residual_post_mlp/max":266,"train/tensor_act_model_layers_62_residual_pre_attn/min":245,"train/tensor_act_model_layers_79_mlp/range":0,"train/tensor_act_model_layers_79_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_9_residual_post_mlp/norm":81851.6796875,"train/tensor_act_model_layers_12_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/mean":-0.0011138916015625,"train/tensor_act_model_layers_21_residual_post_attn/max":86,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_57_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_residual_pre_attn/min":237,"train/tensor_act_model_layers_48_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp_down_proj/max_abs":15.8125,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_76_residual_pre_attn/range":6,"train/tensor_param_model_layers_6_mlp_down_proj_weight/std":0.0322265625,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_35_mlp_gate_proj/max":3.546875,"train/tensor_act_model_layers_90_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_89_mlp_down_proj/max":17.125,"train/tensor_param_model_layers_29_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_37_residual_pre_attn/max":150,"train/tensor_param_model_layers_31_mlp_up_proj_weight/norm":6.375,"train/tensor_act_model_layers_9_mlp_up_proj/range":7.265625,"train/tensor_param_model_layers_19_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_86_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_22_residual_post_mlp/max_abs":94,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_89_mlp_down_proj/min":14.125,"train/tensor_act_model_layers_55_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_87_residual_pre_attn/mean":348,"train/tensor_act_model_layers_64_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_gate_proj/max":3.5,"train/tensor_act_model_layers_50_residual_pre_attn/norm":409077.375,"train/tensor_act_model_layers_1_residual_post_attn/max":4.4375,"train/tensor_act_model_layers_68_mlp_up_proj/mean":-0.09912109375,"train/tensor_act_model_layers_40_residual_post_mlp/max":166,"train/tensor_param_model_layers_12_mlp_up_proj_weight/mean":0.0007781982421875,"train/tensor_act_model_layers_79_mlp/max":4,"train/tensor_param_model_layers_97_mlp_up_proj_weight/range":0.30078125,"train/tensor_act_model_layers_31_mlp_up_proj/norm":5324.001953125,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_60_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_18_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_25_mlp_gate_proj/max":3.5,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/max_abs":0.1240234375,"train/tensor_param_model_layers_60_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_78_mlp/range":0,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/min":-0.12890625,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_53_mlp_down_proj_weight/max":0.12060546875,"train/tensor_param_model_layers_30_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_20_mlp_up_proj_weight/mean":0.000492095947265625,"train/tensor_act_model_layers_62_residual_post_mlp/range":6,"train/tensor_act_model_layers_72_mlp_up_proj/range":7.5625,"train/tensor_act_model_layers_91_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_48_mlp_gate_proj/range":7.5,"train/tensor_act_model_layers_47_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_residual_pre_attn/max_abs":194,"train/tensor_act_model_layers_59_residual_post_mlp/std":0.8359375,"train/tensor_act_model_layers_1_residual_pre_attn/mean":4,"train/tensor_act_model_layers_21_residual_post_attn/min":82,"train/tensor_act_model_layers_6_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_23_mlp_up_proj/range":7.078125,"train/tensor_param_model_layers_44_mlp_up_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_61_mlp_up_proj/norm":5430.2646484375,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_43_residual_post_mlp/max_abs":178,"train/tensor_act_model_layers_81_residual_pre_attn/max":326,"train/tensor_act_model_layers_10_mlp_gate_proj/std":1.8359375,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/range":0.2451171875,"train/tensor_param_model_layers_61_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp_gate_proj/range":7.171875,"train/tensor_act_model_layers_90_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/mean":-0.000377655029296875,"train/tensor_act_model_layers_80_residual_pre_attn/mean":320,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_47_residual_post_mlp/max":194,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/mean":0.00170135498046875,"train/tensor_act_model_layers_45_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_83_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_78_residual_post_mlp/max":318,"train/tensor_act_model_layers_63_mlp_gate_proj/max_abs":4.03125,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_40_mlp_down_proj/min":14.3125,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_50_residual_post_mlp/mean":204,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_19_mlp_gate_proj/range":7.03125,"train/tensor_act_model_layers_82_mlp_up_proj/std":1.8671875,"train/tensor_act_model_layers_6_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_97_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_41_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_5_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_5_mlp_down_proj_weight/min":-0.10693359375,"train/tensor_act_model_layers_93_mlp_gate_proj/min":-4.03125,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_21_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_66_mlp_down_proj_weight/norm":6.59375,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_86_residual_pre_attn/range":6,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_9_mlp_gate_proj/max":3.625,"train/tensor_act_model_layers_41_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_49_residual_pre_attn/range":5,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_14_mlp_up_proj/norm":5383.50146484375,"train/tensor_act_model_layers_38_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_28_mlp_down_proj/max_abs":16.375,"train/tensor_act_model_layers_52_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_residual_pre_attn/mean":232,"train/tensor_param_model_layers_27_mlp_down_proj_weight/max":0.1220703125,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/norm":6.375,"train/tensor_act_model_layers_88_mlp_gate_proj/std":1.8203125,"train/tensor_act_model_layers_40_mlp_down_proj/max":16,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_42_mlp_up_proj/max_abs":3.765625,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_25_residual_pre_attn/std":0.61328125,"train/tensor_param_model_layers_68_mlp_up_proj_weight/norm":6.75,"train/tensor_act_model_layers_81_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_54_mlp/min":4,"train/tensor_act_model_layers_35_mlp/std":0,"train/tensor_act_model_layers_11_mlp_gate_proj/std":1.8671875,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_56_residual_post_mlp/range":5,"train/tensor_param_model_layers_95_mlp_up_proj_weight/std":0.04736328125,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_89_mlp_down_proj_weight/max":0.138671875,"train/tensor_act_model_layers_5_mlp_down_proj/norm":30561.240234375,"train/tensor_param_model_layers_93_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_89_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_0_residual_post_attn/range":0.71484375,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_76_mlp_up_proj/mean":0.03564453125,"train/tensor_act_model_layers_98_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_93_mlp_down_proj/range":2.5625,"train/tensor_act_model_layers_1_mlp_up_proj/max":4.09375,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/min":-0.1240234375,"train/tensor_param_model_layers_94_mlp_down_proj_weight/max":0.1513671875,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/norm":6.3125,"train/tensor_act_model_layers_52_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_2_mlp_up_proj_weight/range":0.22265625,"train/tensor_act_model_layers_55_mlp_up_proj/min":-3.953125,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_87_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/norm":6.3125,"train/tensor_param_model_layers_88_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_80_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_99_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_38_mlp_down_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_60_mlp_down_proj/max_abs":16.625,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_3_residual_pre_attn/range":1.0625,"train/tensor_act_model_layers_66_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_23_mlp_down_proj_weight/min":-0.12255859375,"train/tensor_param_model_layers_98_mlp_down_proj_weight/norm":8.5625,"train/tensor_act_model_layers_37_mlp_up_proj/max":3.546875,"train/tensor_act_model_layers_39_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_69_mlp_up_proj/max":3.953125,"train/tensor_act_model_layers_40_mlp_up_proj/std":1.8046875,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_58_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_mlp/max_abs":4,"train/tensor_act_model_layers_10_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_10_residual_post_mlp/max_abs":45.5,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_15_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_gate_proj/max_abs":3.5,"train/tensor_param_model_layers_18_mlp_down_proj_weight/std":0.03466796875,"train/tensor_param_model_layers_17_mlp_up_proj_weight/range":0.24072265625,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_77_mlp_gate_proj/norm":5392.25146484375,"train/tensor_act_model_layers_43_mlp_gate_proj/mean":-0.07421875,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_67_mlp_down_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_42_mlp_gate_proj/std":1.8671875,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_97_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_71_mlp_down_proj/range":1.75,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_mlp/range":0,"train/tensor_act_model_layers_90_residual_pre_attn/mean":360,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/norm":8.3125,"train/tensor_act_model_layers_33_residual_post_attn/norm":269781.90625,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_54_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_mlp_up_proj/range":7.6875,"train/tensor_act_model_layers_57_mlp/max":4,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_68_residual_pre_attn/max_abs":274,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_66_residual_pre_attn/range":6,"train/tensor_act_model_layers_21_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_23_mlp/std":0,"train/tensor_act_model_layers_50_residual_post_mlp/norm":417269.3125,"train/tensor_param_model_layers_72_mlp_up_proj_weight/mean":-7.104873657226562e-05,"train/tensor_param_model_layers_2_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_mlp_gate_proj/range":8.03125,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_33_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp_gate_proj/max_abs":4.59375,"train/tensor_act_model_layers_68_mlp/range":0,"train/tensor_param_model_layers_83_mlp_up_proj_weight/norm":7.65625,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_60_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_up_proj/range":6.796875,"train/tensor_param_model_layers_63_mlp_down_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_38_residual_post_mlp/max_abs":158,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_73_mlp_down_proj_weight/norm":6.78125,"train/tensor_act_model_layers_54_mlp_up_proj/norm":5317.9443359375,"train/tensor_act_model_layers_89_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_35_mlp_gate_proj/range":7.1875,"train/tensor_param_model_layers_19_mlp_down_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_57_mlp_gate_proj/range":7.625,"train/tensor_act_model_layers_38_residual_post_mlp/std":0.80859375,"train/tensor_act_model_layers_52_mlp/norm":8192,"train/tensor_act_model_layers_69_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_23_mlp_gate_proj/max":4.03125,"train/tensor_param_model_layers_26_mlp_up_proj_weight/min":-0.125,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_46_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_78_mlp_gate_proj/norm":5272.63818359375,"train/tensor_act_model_layers_24_mlp_gate_proj/max_abs":3.84375,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_72_mlp_gate_proj/max":4,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/min":-0.1416015625,"train/tensor_act_model_layers_25_mlp_up_proj/range":6.84375,"train/tensor_param_model_layers_80_mlp_down_proj_weight/max":0.12890625,"train/tensor_act_model_layers_15_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_13_mlp_down_proj_weight/norm":6.125,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_33_residual_post_attn/max_abs":134,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_78_residual_post_attn/range":6,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/norm":6.4375,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_77_mlp_gate_proj/min":-3.890625,"train/tensor_param_model_layers_74_mlp_up_proj_weight/norm":6.90625,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_71_mlp_up_proj_weight/mean":-0.00118255615234375,"train/tensor_act_model_layers_59_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_1_residual_post_attn/frac_near_user_limit":0,"train/tensor_act_model_layers_90_residual_pre_attn/max":362,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_6_mlp_down_proj_weight/max":0.12060546875,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_62_mlp_up_proj/range":7.65625,"train/tensor_param_model_layers_44_mlp_down_proj_weight/range":0.2587890625,"train/tensor_act_model_layers_32_residual_pre_attn/range":4,"train/tensor_act_model_layers_57_mlp_gate_proj/max_abs":3.875,"train/tensor_act_model_layers_9_mlp_gate_proj/max_abs":3.71875,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/mean":-0.001373291015625,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_28_mlp_gate_proj/mean":0.2177734375,"train/tensor_param_model_layers_86_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_96_mlp_up_proj_weight/norm":8.75,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_13_residual_post_attn/std":0.46875,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/max":0.11376953125,"train/tensor_act_model_layers_66_mlp_gate_proj/range":6.875,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/mean":0.000560760498046875,"train/tensor_param_model_layers_97_mlp_down_proj_weight/min":-0.142578125,"train/tensor_act_model_layers_80_mlp/norm":8192,"train/tensor_act_model_layers_44_mlp_up_proj/mean":0.18359375,"train/tensor_act_model_layers_5_residual_pre_attn/norm":40949.03515625,"train/tensor_act_model_layers_92_mlp_up_proj/range":8.625,"train/tensor_act_model_layers_6_mlp/min":4,"train/tensor_act_model_layers_31_mlp_down_proj/std":0.35546875,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_65_mlp/max_abs":4,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_45_residual_post_mlp/std":0.80859375,"train/tensor_param_model_layers_52_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_47_mlp_up_proj/norm":5280.3115234375,"train/tensor_act_model_layers_2_residual_post_mlp/min":11.375,"train/tensor_act_model_layers_18_residual_pre_attn/norm":147125.609375,"train/tensor_param_model_layers_71_mlp_down_proj_weight/mean":0.00026702880859375,"train/tensor_act_model_layers_31_residual_post_mlp/std":0.6484375,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_78_residual_pre_attn/min":308,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_80_mlp_up_proj_weight/norm":7.21875,"train/tensor_act_model_layers_14_residual_post_attn/max_abs":57.5,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_29_mlp/range":0,"train/tensor_act_model_layers_42_residual_post_mlp/norm":351717.84375,"train/tensor_act_model_layers_76_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_65_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_29_mlp_down_proj/std":0.388671875,"train/tensor_param_model_layers_31_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_47_residual_pre_attn/min":185,"train/tensor_act_model_layers_19_residual_post_attn/std":0.58203125,"train/tensor_act_model_layers_37_mlp/std":0,"train/tensor_act_model_layers_38_mlp/min":4,"train/tensor_act_model_layers_51_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_36_residual_pre_attn/max_abs":146,"train/tensor_act_model_layers_67_residual_post_attn/max":270,"train/tensor_act_model_layers_44_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_90_mlp/max":4,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_69_mlp_down_proj_weight/norm":6.6875,"train/tensor_act_model_layers_33_mlp_gate_proj/norm":5319.4990234375,"train/tensor_act_model_layers_86_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_63_mlp_up_proj_weight/mean":-0.00013065338134765625,"train/tensor_act_model_layers_38_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_13_mlp/max":4,"train/tensor_act_model_layers_29_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_11_residual_post_attn/min":42.5,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_26_mlp_down_proj_weight/mean":-0.0003185272216796875,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_64_mlp_up_proj/norm":5380.533203125,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/min":-0.1337890625,"train/tensor_act_model_layers_39_residual_post_mlp/max":162,"train/tensor_act_model_layers_22_mlp/mean":4,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_72_residual_post_attn/max":290,"train/tensor_param_model_layers_36_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_60_mlp_down_proj_weight/norm":6.40625,"train/tensor_param_model_layers_99_mlp_down_proj_weight/max":0.1650390625,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_36_residual_post_attn/mean":144,"train/tensor_act_model_layers_69_mlp_gate_proj/mean":-0.07470703125,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_19_mlp_down_proj/std":0.37890625,"train/tensor_act_model_layers_9_mlp/min":4,"train/tensor_act_model_layers_30_residual_post_mlp/max_abs":126.5,"train/tensor_act_model_layers_7_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_99_mlp_up_proj/range":9.21875,"train/tensor_param_model_layers_87_mlp_up_proj_weight/norm":7.75,"train/tensor_param_model_layers_73_mlp_up_proj_weight/min":-0.126953125,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_39_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_67_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_69_mlp_up_proj/range":7.75,"train/tensor_act_model_layers_13_residual_pre_attn/norm":106329.6875,"train/tensor_act_model_layers_91_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_25_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_11_mlp/std":0,"train/tensor_act_model_layers_55_residual_pre_attn/range":5,"train/tensor_act_model_layers_24_mlp_up_proj/norm":5431.421875,"train/tensor_param_model_layers_73_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_40_residual_post_attn/norm":327142.15625,"train/tensor_act_model_layers_21_mlp_up_proj/range":7.015625,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_10_mlp_down_proj_weight/mean":-0.0003871917724609375,"train/tensor_param_model_layers_97_mlp_up_proj_weight/norm":8.625,"train/tensor_act_model_layers_21_residual_pre_attn/min":82,"train/tensor_act_model_layers_0_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_mlp_up_proj/max_abs":4.15625,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_mlp_down_proj/range":1.875,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_47_mlp/norm":8192,"train/tensor_act_model_layers_18_residual_post_attn/std":0.58203125,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_56_mlp_down_proj_weight/mean":0.00012159347534179688,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_26_mlp_up_proj_weight/range":0.251953125,"train/tensor_act_model_layers_60_mlp/std":0,"train/tensor_act_model_layers_42_residual_post_mlp/min":169,"train/tensor_act_model_layers_72_residual_post_attn/min":284,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/mean":0.000774383544921875,"train/tensor_act_model_layers_86_mlp_down_proj/std":0.5625,"train/tensor_param_model_layers_75_mlp_up_proj_weight/max":0.1318359375,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/mean":-0.000270843505859375,"train/tensor_param_model_layers_93_mlp_down_proj_weight/std":0.046630859375,"train/tensor_act_model_layers_11_residual_post_attn/max":45.5,"train/tensor_act_model_layers_33_mlp/max_abs":4,"train/tensor_act_model_layers_97_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_residual_pre_attn/norm":155293.640625,"train/tensor_param_model_layers_80_mlp_down_proj_weight/range":0.271484375,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/max_abs":0.142578125,"train/tensor_act_model_layers_15_mlp_up_proj/max":3.625,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/mean":0.000545501708984375,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_28_residual_pre_attn/std":0.625,"train/tensor_param_model_layers_38_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_19_residual_post_mlp/mean":80,"train/tensor_act_model_layers_79_residual_post_mlp/max_abs":322,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_31_residual_post_mlp/norm":261421.984375,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/mean":0.00091552734375,"train/tensor_act_model_layers_74_mlp_gate_proj/mean":0.04052734375,"train/tensor_act_model_layers_24_mlp_up_proj/std":1.8671875,"train/tensor_act_model_layers_11_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_54_residual_post_mlp/norm":450021.09375,"train/tensor_param_model_layers_93_mlp_up_proj_weight/std":0.04736328125,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_10_mlp_up_proj_weight/range":0.25,"train/tensor_act_model_layers_27_mlp_up_proj/min":-4.53125,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_2_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_mlp/max_abs":4,"train/tensor_param_model_layers_83_mlp_up_proj_weight/range":0.267578125,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_70_residual_post_attn/min":276,"train/tensor_param_model_layers_93_mlp_up_proj_weight/range":0.2841796875,"train/tensor_act_model_layers_33_mlp/range":0,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_attn/norm":482821,"train/tensor_act_model_layers_34_mlp_up_proj/max":3.34375,"train/tensor_param_model_layers_6_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/range":465,"train/tensor_act_model_layers_18_mlp_gate_proj/max_abs":3.71875,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_72_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/max":0.125,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/std":0.04736328125,"train/tensor_act_model_layers_48_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_60_mlp_up_proj_weight/max":0.126953125,"train/tensor_act_model_layers_9_mlp_down_proj/std":0.447265625,"train/tensor_act_model_layers_62_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_70_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_32_mlp/std":0,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/mean":-0.000335693359375,"train/tensor_act_model_layers_13_residual_post_mlp/mean":56,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_36_mlp_down_proj_weight/mean":9.489059448242188e-05,"train/tensor_act_model_layers_3_mlp_gate_proj/norm":5351.54443359375,"train/tensor_act_model_layers_11_residual_pre_attn/max_abs":45.5,"train/tensor_act_model_layers_61_residual_pre_attn/std":0.8359375,"train/tensor_param_model_layers_83_mlp_down_proj_weight/max_abs":0.1591796875,"train/tensor_act_model_layers_80_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_70_residual_post_attn/norm":573038.25,"train/tensor_act_model_layers_60_residual_pre_attn/range":5,"train/tensor_act_model_layers_15_mlp_down_proj/norm":30466.552734375,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_5_residual_post_mlp/range":1.75,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_94_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_84_residual_pre_attn/range":6,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/min":-0.119140625,"train/tensor_act_model_layers_81_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_14_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_12_mlp_up_proj/min":-3.703125,"train/tensor_act_model_layers_41_mlp_down_proj/std":0.38671875,"train/tensor_param_model_layers_30_mlp_down_proj_weight/max_abs":0.12158203125,"train/tensor_act_model_layers_62_mlp/range":0,"train/tensor_param_model_layers_75_mlp_down_proj_weight/range":0.2626953125,"train/tensor_act_model_layers_12_mlp_gate_proj/mean":0.041015625,"train/tensor_act_model_layers_5_residual_post_attn/std":0.23046875,"train/tensor_act_model_layers_37_residual_post_attn/range":5,"train/tensor_act_model_layers_51_residual_post_attn/max_abs":206,"train/tensor_act_model_layers_50_residual_post_mlp/max":206,"train/tensor_param_model_layers_51_mlp_up_proj_weight/mean":0.0004863739013671875,"train/tensor_act_model_layers_43_residual_pre_attn/max":174,"train/tensor_act_model_layers_37_residual_post_mlp/norm":310758.375,"train/tensor_act_model_layers_25_mlp_up_proj/norm":5374.0615234375,"train/tensor_param_model_layers_94_mlp_up_proj_weight/range":0.28515625,"train/tensor_act_model_layers_50_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_70_residual_post_attn/max":282,"train/tensor_act_model_layers_64_mlp_down_proj/max_abs":16.875,"train/tensor_act_model_layers_0_mlp_down_proj/max_abs":112,"train/tensor_act_model_layers_31_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_32_mlp_down_proj_weight/max_abs":0.12890625,"train/tensor_param_model_layers_68_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_54_residual_post_mlp/mean":220,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_4_residual_post_mlp/std":0.193359375,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_40_mlp_down_proj_weight/norm":6.25,"train/tensor_act_model_layers_23_residual_post_mlp/max_abs":98,"train/tensor_act_model_layers_42_residual_post_attn/std":0.80859375,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_1_residual_pre_attn/std":0.09228515625,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/min":-0.11962890625,"train/tensor_act_model_layers_58_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_43_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_51_residual_pre_attn/max_abs":206,"train/tensor_param_model_layers_51_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_9_mlp_up_proj/min":-3.765625,"train/tensor_act_model_layers_66_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_70_residual_pre_attn/max":282,"train/tensor_act_model_layers_69_mlp_down_proj/max_abs":16,"train/tensor_act_model_layers_38_residual_post_attn/min":149,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_31_residual_post_mlp/mean":127.5,"train/tensor_act_model_layers_22_residual_post_mlp/range":4,"train/tensor_act_model_layers_40_residual_post_attn/range":5,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_19_residual_post_mlp/norm":163485.40625,"train/tensor_act_model_layers_68_mlp/min":4,"train/tensor_act_model_layers_78_residual_pre_attn/max_abs":314,"train/tensor_act_model_layers_69_mlp_gate_proj/max":3.515625,"train/tensor_param_model_layers_9_mlp_up_proj_weight/norm":6.15625,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/norm":7.28125,"train/tensor_act_model_layers_47_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_64_mlp_gate_proj/mean":0.1982421875,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_81_mlp_gate_proj/std":1.8046875,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_43_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_mlp_gate_proj/max":4.84375,"train/tensor_act_model_layers_31_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_51_residual_post_mlp/range":5,"train/tensor_act_model_layers_43_mlp/min":4,"train/tensor_act_model_layers_78_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_residual_post_mlp/std":0.515625,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_87_residual_post_attn/mean":348,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_93_residual_post_mlp/min":372,"train/tensor_act_model_layers_65_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_26_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_21_mlp_gate_proj/max":3.734375,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp_gate_proj/mean":0.08544921875,"train/tensor_act_model_layers_77_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_86_mlp_down_proj/mean":15.0625,"train/tensor_param_model_layers_86_mlp_up_proj_weight/max":0.1474609375,"train/tensor_act_model_layers_13_residual_post_mlp/norm":114445.7734375,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_96_mlp_up_proj_weight/max":0.15625,"train/tensor_param_model_layers_7_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_20_mlp/min":4,"train/tensor_param_model_layers_15_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_91_residual_pre_attn/max":366,"train/tensor_act_model_layers_47_mlp/range":0,"train/tensor_act_model_layers_91_mlp_gate_proj/range":8.71875,"train/tensor_param_model_layers_34_mlp_down_proj_weight/norm":6.25,"train/tensor_param_model_layers_40_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_mlp_gate_proj/range":7.1875,"train/tensor_act_model_layers_89_mlp_up_proj/max_abs":4.5,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_96_residual_pre_attn/min":380,"train/tensor_param_model_layers_7_mlp_down_proj_weight/max_abs":0.1220703125,"train/tensor_act_model_layers_38_residual_post_attn/max":154,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/min":-0.134765625,"train/tensor_act_model_layers_93_mlp/min":4,"train/tensor_act_model_layers_99_mlp_up_proj/std":1.8515625,"train/tensor_param_model_layers_53_mlp_up_proj_weight/max":0.1279296875,"train/tensor_param_model_layers_46_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_47_residual_post_mlp/range":5,"train/tensor_act_model_layers_30_mlp/min":4,"train/tensor_act_model_layers_71_residual_pre_attn/max":286,"train/tensor_act_model_layers_49_residual_pre_attn/min":193,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_79_mlp_gate_proj/min":-4,"train/tensor_param_model_layers_54_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_37_residual_post_mlp/mean":152,"train/tensor_act_model_layers_55_mlp_gate_proj/norm":5356.4541015625,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_residual_pre_attn/norm":302566.5,"train/tensor_act_model_layers_80_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_76_mlp/max":4,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_96_residual_post_mlp/frac_near_user_limit":1,"train/epoch":0.23592854735422986,"train/tensor_act_model_layers_47_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_41_mlp_gate_proj/range":7.15625,"train/tensor_act_model_layers_94_residual_post_attn/min":372,"train/tensor_param_model_layers_29_mlp_up_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/mean":-0.000286102294921875,"train/tensor_act_model_layers_76_residual_pre_attn/mean":304,"train/tensor_param_model_layers_56_mlp_down_proj_weight/range":0.255859375,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_46_residual_post_mlp/range":5,"train/tensor_act_model_layers_14_residual_post_attn/min":54.5,"train/tensor_param_model_layers_1_mlp_up_proj_weight/std":0.033447265625,"train/tensor_act_model_layers_54_mlp/norm":8192,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/std":0.033935546875,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_50_mlp_down_proj_weight/max":0.123046875,"train/tensor_act_model_layers_9_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_90_mlp_gate_proj/range":7.921875,"train/tensor_act_model_layers_15_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_65_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_0_mlp_down_proj/mean":34.75,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_residual_pre_attn/max_abs":24.875,"train/tensor_act_model_layers_59_mlp_gate_proj/max_abs":4.03125,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_2_residual_post_mlp/max":12.4375,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_0_mlp_gate_proj/mean":-0.00775146484375,"train/tensor_act_model_layers_47_mlp/std":0,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_32_mlp_up_proj/min":-3.828125,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/norm":7.65625,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_77_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_14_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/norm":6.25,"train/tensor_act_model_layers_16_mlp_up_proj/max":3.53125,"train/tensor_act_model_layers_91_residual_pre_attn/range":6,"train/tensor_act_model_layers_49_mlp_up_proj/norm":5506.42919921875,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_36_mlp_gate_proj/mean":-0.265625,"train/tensor_act_model_layers_8_residual_pre_attn/std":0.314453125,"train/tensor_act_model_layers_2_mlp/min":4,"train/tensor_param_model_layers_51_mlp_up_proj_weight/max":0.125,"train/tensor_param_model_layers_34_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_residual_pre_attn/range":6,"train/tensor_param_model_layers_62_mlp_down_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_47_mlp_up_proj_weight/max":0.125,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_10_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_80_residual_post_attn/max_abs":322,"train/tensor_param_model_layers_96_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_96_mlp_up_proj/max_abs":4.65625,"train/tensor_act_model_layers_31_residual_pre_attn/norm":253346.515625,"train/tensor_param_model_layers_83_mlp_down_proj_weight/max":0.1591796875,"train/tensor_act_model_layers_58_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_down_proj/norm":30357.533203125,"train/tensor_act_model_layers_14_residual_post_attn/max":57.5,"train/tensor_act_model_layers_17_residual_post_mlp/range":3.5,"train/tensor_act_model_layers_87_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_88_residual_post_mlp/max_abs":358,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_12_mlp_down_proj_weight/range":0.24072265625,"train/tensor_act_model_layers_47_mlp_down_proj/max":16.5,"train/tensor_act_model_layers_26_residual_post_mlp/max_abs":110,"train/tensor_act_model_layers_94_residual_post_attn/max":378,"train/tensor_param_model_layers_53_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_pre_attn/norm":441845.125,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_72_mlp_up_proj_weight/max":0.1259765625,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_70_mlp_up_proj_weight/std":0.037841796875,"train/tensor_act_model_layers_84_mlp_gate_proj/std":1.78125,"train/tensor_act_model_layers_0_mlp_gate_proj/range":12.09375,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_37_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_58_mlp_up_proj/mean":-0.0673828125,"train/tensor_act_model_layers_78_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_mlp_gate_proj/std":1.875,"train/tensor_act_model_layers_67_mlp_up_proj/norm":5234.94140625,"train/tensor_act_model_layers_19_mlp_down_proj/norm":30410.458984375,"train/tensor_act_model_layers_84_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp_up_proj/range":7.390625,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_1_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_10_mlp_up_proj/std":1.859375,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_9_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_99_mlp_up_proj_weight/min":-0.146484375,"train/tensor_param_model_layers_14_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_residual_post_mlp/min":213,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_51_mlp_down_proj/max_abs":15.9375,"train/tensor_act_model_layers_23_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_7_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_56_mlp/max":4,"train/tensor_act_model_layers_73_mlp_up_proj/range":7.421875,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/max":0.13671875,"train/tensor_act_model_layers_48_residual_post_attn/min":189,"train/tensor_act_model_layers_99_mlp/range":0,"train/tensor_act_model_layers_75_mlp_up_proj/norm":5266.39111328125,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_20_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_1_mlp_down_proj/mean":18.625,"train/tensor_param_model_layers_85_mlp_down_proj_weight/mean":-0.00048828125,"train/tensor_param_model_layers_78_mlp_down_proj_weight/max":0.1357421875,"train/tensor_act_model_layers_71_mlp/std":0,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_5_residual_post_mlp/mean":24,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_59_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_1_mlp_gate_proj/mean":0.0810546875,"train/tensor_act_model_layers_35_residual_post_attn/max":142,"train/tensor_act_model_layers_55_residual_post_mlp/std":0.8203125,"train/tensor_param_model_layers_40_mlp_down_proj_weight/range":0.248046875,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_85_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_32_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_84_mlp/min":4,"train/tensor_act_model_layers_38_mlp/std":0,"train/tensor_act_model_layers_49_mlp_up_proj/max":3.390625,"train/tensor_act_model_layers_35_residual_post_attn/min":137,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/mean":0.000621795654296875,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_15_mlp/max":4,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_66_mlp_down_proj/max_abs":16.875,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_12_mlp_gate_proj/range":7.265625,"train/tensor_act_model_layers_93_mlp_up_proj/std":1.875,"train/tensor_act_model_layers_12_residual_post_attn/max_abs":49.5,"train/tensor_act_model_layers_81_mlp/mean":4,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_52_mlp_up_proj_weight/norm":6.375,"train/tensor_act_model_layers_48_mlp/max_abs":4,"train/tensor_act_model_layers_16_residual_post_mlp/min":66,"train/tensor_act_model_layers_31_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_45_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_8_mlp_up_proj/min":-3.21875,"train/tensor_param_model_layers_67_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_28_mlp_up_proj/max":3.4375,"train/tensor_act_model_layers_3_mlp_up_proj/mean":-0.0771484375,"train/tensor_act_model_layers_68_mlp_gate_proj/max":3.78125,"train/tensor_act_model_layers_54_residual_pre_attn/max_abs":218,"train/tensor_act_model_layers_99_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_11_mlp_down_proj/max_abs":16.625,"train/tensor_act_model_layers_48_residual_post_attn/max":194,"train/tensor_act_model_layers_31_residual_post_attn/std":0.69140625,"train/tensor_act_model_layers_67_residual_pre_attn/max":270,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_27_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_up_proj/mean":-0.1435546875,"train/tensor_act_model_layers_42_residual_post_mlp/mean":172,"train/tensor_act_model_layers_2_residual_pre_attn/min":7.5,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_post_attn/std":0.8203125,"train/tensor_param_model_layers_62_mlp_up_proj_weight/min":-0.1318359375,"train/tensor_act_model_layers_46_residual_post_attn/range":5,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_42_mlp/min":4,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/mean":8.821487426757812e-05,"train_steps_per_second":9.889,"train/tensor_act_model_layers_46_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_49_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_50_residual_post_attn/max_abs":202,"train/tensor_act_model_layers_62_mlp_gate_proj/max":3.75,"train/tensor_param_model_layers_26_mlp_up_proj_weight/max":0.126953125,"train/tensor_act_model_layers_11_mlp/min":4,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/norm":10.375,"train/tensor_act_model_layers_42_residual_pre_attn/norm":343525.96875,"train/tensor_act_model_layers_31_mlp_up_proj/min":-3.734375,"train/tensor_act_model_layers_92_mlp/mean":4,"train/tensor_act_model_layers_15_residual_post_attn/max_abs":61.5,"train/tensor_act_model_layers_22_mlp_up_proj/range":7.359375,"train/tensor_act_model_layers_93_mlp/std":0,"train/tensor_act_model_layers_67_mlp_gate_proj/min":-3.625,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/max":0.12890625,"train/tensor_act_model_layers_28_residual_post_attn/max":114.5,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_57_mlp_up_proj_weight/norm":6.40625,"train/tensor_act_model_layers_68_mlp_down_proj/range":1.9375,"train/tensor_act_model_layers_92_mlp_gate_proj/max":4.21875,"train/tensor_param_model_layers_25_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_attn/range":3.125,"train/tensor_param_model_layers_22_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_8_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_83_residual_post_attn/max":334,"train/tensor_act_model_layers_80_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_22_residual_pre_attn/max":90,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_95_mlp/min":4,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_96_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_18_residual_post_mlp/norm":155293.640625,"train/tensor_param_model_layers_51_mlp_down_proj_weight/max_abs":0.12890625,"train/tensor_param_model_layers_98_mlp_up_proj_weight/norm":8.6875,"train/tensor_act_model_layers_30_mlp/max":4,"train/tensor_act_model_layers_8_mlp/norm":8192,"train/tensor_act_model_layers_5_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/max":0.1279296875,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_90_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_11_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_78_residual_post_attn/max_abs":314,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/norm":6.5625,"train/tensor_act_model_layers_28_mlp_gate_proj/std":1.84375,"train/tensor_act_model_layers_78_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_7_mlp_up_proj_weight/min":-0.1142578125,"train/tensor_act_model_layers_6_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_40_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_87_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_33_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_residual_post_attn/min":328,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_29_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_29_residual_pre_attn/range":4.5,"train/tensor_act_model_layers_20_residual_post_attn/range":4,"train/tensor_act_model_layers_35_mlp_gate_proj/mean":0.0294189453125,"train/tensor_act_model_layers_23_mlp_up_proj/min":-3.375,"train/tensor_act_model_layers_46_residual_post_mlp/max":190,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/max_abs":0.1357421875,"train/tensor_act_model_layers_34_mlp_down_proj/max_abs":16.25,"train/tensor_param_model_layers_60_mlp_down_proj_weight/max":0.125,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_97_mlp_up_proj/min":-4.09375,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_28_residual_post_attn/min":110,"train/tensor_act_model_layers_96_mlp_up_proj/norm":5416.5009765625,"train/tensor_param_model_layers_0_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_16_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_17_mlp_up_proj_weight/norm":6.25,"train/tensor_param_model_layers_6_mlp_up_proj_weight/min":-0.125,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_79_mlp_gate_proj/norm":5243.95068359375,"train/tensor_act_model_layers_80_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_73_mlp_down_proj_weight/std":0.037353515625,"train/tensor_act_model_layers_17_mlp_gate_proj/max_abs":3.75,"train/tensor_param_model_layers_67_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp/max_abs":4,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_85_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp/mean":4,"train/tensor_act_model_layers_24_mlp_gate_proj/max":3.765625,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_46_mlp_gate_proj/std":1.875,"train/tensor_act_model_layers_76_residual_post_attn/range":6,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_8_mlp_up_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_58_residual_post_mlp/max_abs":238,"train/tensor_act_model_layers_72_mlp/mean":4,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_98_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_87_mlp_gate_proj/min":-3.875,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_51_mlp_down_proj/std":0.349609375,"train/tensor_act_model_layers_96_residual_post_attn/max":386,"train/tensor_act_model_layers_41_residual_pre_attn/mean":164,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_64_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_18_residual_post_mlp/mean":76,"train/tensor_act_model_layers_40_residual_pre_attn/min":157,"train/tensor_act_model_layers_90_mlp_up_proj/norm":5370.29248046875,"train/tensor_act_model_layers_75_residual_post_attn/mean":300,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/norm":6.84375,"train/tensor_act_model_layers_94_residual_post_attn/mean":376,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_mlp_gate_proj/range":6.75,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_11_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_residual_pre_attn/mean":220,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_55_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_81_mlp/max_abs":4,"train/tensor_param_model_layers_30_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_98_mlp_down_proj/max":18.25,"train/tensor_param_model_layers_58_mlp_down_proj_weight/mean":0.00177764892578125,"train/tensor_act_model_layers_54_mlp_up_proj/max_abs":3.53125,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_73_mlp_gate_proj/max":3.671875,"train/tensor_act_model_layers_81_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_3_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_80_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_2_mlp_up_proj_weight/min":-0.10302734375,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/range":0.2939453125,"train/tensor_act_model_layers_99_residual_pre_attn/range":6,"train/tensor_act_model_layers_98_mlp_down_proj/min":14,"train/tensor_act_model_layers_45_mlp_gate_proj/max_abs":3.796875,"train/tensor_act_model_layers_38_mlp_down_proj/range":1.8125,"train/tensor_act_model_layers_95_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_60_residual_post_mlp/max":246,"train/tensor_act_model_layers_96_residual_post_attn/min":380,"train/tensor_act_model_layers_68_mlp_up_proj/range":7.640625,"train/tensor_act_model_layers_62_residual_post_mlp/max":254,"train/tensor_act_model_layers_8_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_87_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_17_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_18_residual_pre_attn/std":0.5625,"train/tensor_act_model_layers_65_residual_pre_attn/min":256,"train/tensor_param_model_layers_46_mlp_up_proj_weight/max_abs":0.130859375,"train/tensor_act_model_layers_91_residual_pre_attn/mean":364,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_26_mlp_gate_proj/min":-3.40625,"train/tensor_act_model_layers_60_mlp_down_proj/norm":30546.0546875,"train/tensor_act_model_layers_10_mlp/std":0,"train/tensor_act_model_layers_41_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_96_mlp_down_proj/std":0.6171875,"train/tensor_act_model_layers_45_mlp/std":0,"train/tensor_act_model_layers_52_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_param_model_layers_72_mlp_down_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_36_residual_post_mlp/min":145,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp_gate_proj/min":-4.09375,"train/tensor_act_model_layers_8_mlp_gate_proj/range":6.53125,"train/tensor_act_model_layers_1_mlp/norm":8192,"train/tensor_act_model_layers_54_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_62_mlp_down_proj/norm":30527.283203125,"train/tensor_act_model_layers_6_mlp_down_proj/norm":30516.80078125,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_43_mlp_down_proj/std":0.38671875,"train/tensor_act_model_layers_76_mlp/std":0,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_50_mlp_down_proj/max_abs":16,"train/tensor_act_model_layers_89_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_85_mlp/range":0,"train/tensor_param_model_layers_35_mlp_up_proj_weight/range":0.25390625,"train/tensor_act_model_layers_22_mlp_gate_proj/max":3.890625,"train/tensor_act_model_layers_67_residual_post_attn/min":264,"train/tensor_act_model_layers_25_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/range":0.2421875,"train/tensor_act_model_layers_38_residual_pre_attn/max_abs":154,"train/tensor_act_model_layers_24_residual_post_mlp/range":4,"train/tensor_act_model_layers_38_residual_post_mlp/mean":156,"train/tensor_param_model_layers_13_mlp_up_proj_weight/mean":-0.000720977783203125,"train/tensor_act_model_layers_65_mlp_gate_proj/max_abs":3.8125,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_31_residual_post_attn/min":122,"train/tensor_act_model_layers_80_residual_post_attn/range":6,"train/tensor_act_model_layers_11_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_48_mlp_up_proj/norm":5350.0791015625,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_post_attn/min":213,"train/tensor_act_model_layers_19_mlp/min":4,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/max":0.142578125,"train/tensor_act_model_layers_50_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/max_abs":0,"train/global_step":3500,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_27_residual_post_mlp/std":0.625,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/max_abs":0.1318359375,"train/tensor_act_model_layers_30_residual_pre_attn/mean":119.5,"train/tensor_act_model_layers_83_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_88_mlp_down_proj/min":14.125,"train/tensor_param_model_layers_44_mlp_up_proj_weight/std":0.03515625,"train/tensor_param_model_layers_4_mlp_down_proj_weight/max":0.1083984375,"train/tensor_act_model_layers_23_residual_pre_attn/min":90,"train/tensor_act_model_layers_13_mlp/std":0,"train/tensor_act_model_layers_35_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_41_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_39_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_72_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/norm":6.34375,"train/tensor_param_model_layers_47_mlp_down_proj_weight/norm":6.1875,"train/tensor_act_model_layers_92_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_46_mlp_down_proj_weight/norm":6.25,"train/tensor_act_model_layers_50_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_85_mlp_gate_proj/min":-4.1875,"train/tensor_act_model_layers_77_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_24_mlp_down_proj/norm":30368.955078125,"train/tensor_param_model_layers_35_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_residual_post_mlp/range":2,"train/tensor_act_model_layers_50_residual_post_mlp/min":201,"train/tensor_act_model_layers_36_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_21_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_mlp_up_proj/min":-4.25,"train/tensor_act_model_layers_3_mlp/max_abs":4,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_10_residual_pre_attn/max_abs":41.5,"train/tensor_act_model_layers_88_mlp_down_proj/max":17.25,"train/tensor_act_model_layers_67_residual_post_mlp/range":6,"train/tensor_act_model_layers_40_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/norm":6.3125,"train/tensor_param_model_layers_40_mlp_down_proj_weight/min":-0.123046875,"train/tensor_act_model_layers_38_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_92_mlp_up_proj/max_abs":4.5625,"train/tensor_param_model_layers_58_mlp_up_proj_weight/norm":6.40625,"train/tensor_act_model_layers_14_residual_pre_attn/norm":114445.7734375,"train/tensor_act_model_layers_89_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_61_mlp_down_proj/norm":30599.26953125,"train/tensor_act_model_layers_48_residual_post_attn/std":0.81640625,"train/tensor_act_model_layers_11_mlp_down_proj/max":16.625,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/norm":6.28125,"train/tensor_param_model_layers_58_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_68_residual_post_mlp/max_abs":278,"train/tensor_act_model_layers_31_residual_post_attn/max":126.5,"train/tensor_param_model_layers_39_mlp_up_proj_weight/max":0.11767578125,"train/tensor_act_model_layers_94_residual_pre_attn/norm":769645.75,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_54_residual_post_attn/max":218,"train/tensor_act_model_layers_10_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_0_mlp_up_proj_weight/range":0.25,"train/tensor_param_model_layers_20_mlp_down_proj_weight/max":0.123046875,"train/tensor_act_model_layers_58_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp_down_proj/std":0.4375,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/range":0.259765625,"train/tensor_act_model_layers_37_residual_post_attn/min":145,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_71_mlp_gate_proj/max_abs":3.640625,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/range":0.2841796875,"train/tensor_act_model_layers_16_mlp/min":4,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/norm":6.25,"train/tensor_act_model_layers_27_mlp_up_proj/mean":-0.09912109375,"train/tensor_param_model_layers_26_mlp_down_proj_weight/max":0.12451171875,"train/tensor_act_model_layers_48_mlp/max":4,"train/tensor_param_model_layers_90_mlp_down_proj_weight/max_abs":0.14453125,"train/tensor_act_model_layers_40_mlp_gate_proj/mean":-0.1103515625,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_residual_pre_attn/mean":12,"train/tensor_act_model_layers_15_mlp_down_proj/max":16,"train/tensor_param_model_layers_50_mlp_down_proj_weight/mean":0.0014190673828125,"train/tensor_act_model_layers_42_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_4_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_32_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_16_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_mlp_gate_proj/mean":-0.049560546875,"train/tensor_act_model_layers_81_mlp_down_proj/max_abs":16.5,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_33_residual_pre_attn/max":134,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/min":-0.125,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_89_mlp_up_proj_weight/std":0.044921875,"train/tensor_act_model_layers_82_mlp_gate_proj/range":8.5,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_80_residual_post_mlp/max_abs":326,"train/tensor_act_model_layers_70_mlp_down_proj/max_abs":16.375,"train/tensor_act_model_layers_46_mlp_up_proj/min":-3.625,"train/tensor_act_model_layers_12_mlp_gate_proj/min":-3.53125,"train/tensor_act_model_layers_3_residual_post_attn/max":12.5625,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_42_residual_pre_attn/max_abs":170,"train/tensor_act_model_layers_25_residual_post_attn/mean":100,"train/tensor_act_model_layers_52_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_26_mlp_up_proj/max":3.609375,"train/tensor_act_model_layers_63_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_31_mlp/mean":4,"train/tensor_act_model_layers_91_mlp_gate_proj/min":-4.5,"train/tensor_act_model_layers_35_residual_pre_attn/norm":286181.65625,"train/tensor_param_model_layers_17_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_64_mlp/norm":8192,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_96_residual_post_mlp/norm":794221.6875,"train/tensor_act_model_layers_30_residual_pre_attn/min":118,"train/tensor_act_model_layers_36_residual_post_mlp/max":150,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_98_residual_post_attn/mean":392,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_89_residual_post_attn/range":6,"train/tensor_act_model_layers_63_mlp_down_proj/max":16.375,"train/tensor_param_model_layers_78_mlp_up_proj_weight/mean":0.0009918212890625,"train/tensor_act_model_layers_99_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_mlp_up_proj/std":1.859375,"train/tensor_act_model_layers_28_mlp_up_proj/norm":5377.7197265625,"train/tensor_act_model_layers_71_residual_post_attn/mean":284,"train/tensor_act_model_layers_16_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_94_mlp/std":0,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/mean":0.00124359130859375,"train/tensor_act_model_layers_83_residual_post_mlp/mean":336,"train/tensor_act_model_layers_97_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_14_mlp_gate_proj/min":-3.34375,"train/tensor_act_model_layers_15_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_48_residual_pre_attn/norm":392677.4375,"train/tensor_act_model_layers_75_residual_post_attn/range":6,"train/tensor_act_model_layers_44_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_32_mlp/max":4,"train/tensor_act_model_layers_39_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_7_mlp/range":0,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/norm":6.09375,"train/tensor_act_model_layers_14_residual_post_attn/std":0.494140625,"train/tensor_act_model_layers_28_residual_post_mlp/mean":115.5,"train/tensor_act_model_layers_60_residual_post_mlp/min":241,"train/tensor_act_model_layers_73_mlp_gate_proj/max_abs":3.671875,"train/tensor_act_model_layers_5_residual_post_mlp/norm":49132.38671875,"train/tensor_act_model_layers_33_mlp_up_proj/mean":-0.15625,"train/tensor_act_model_layers_94_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_23_mlp_down_proj/range":2.5,"train/tensor_act_model_layers_52_mlp_gate_proj/std":1.8828125,"train/tensor_act_model_layers_62_residual_post_mlp/min":248,"train/tensor_param_model_layers_95_mlp_down_proj_weight/min":-0.1640625,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_83_residual_pre_attn/max_abs":334,"train/tensor_act_model_layers_42_residual_post_attn/max":170,"train/tensor_act_model_layers_39_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_9_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_58_mlp/max":4,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_74_residual_post_attn/max_abs":298,"train/tensor_act_model_layers_35_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_57_mlp_gate_proj/max":3.875,"train/tensor_act_model_layers_33_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_28_residual_pre_attn/norm":228930.234375,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/min":-0.12890625,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_82_residual_pre_attn/max":330,"train/tensor_param_model_layers_64_mlp_down_proj_weight/norm":6.40625,"train/tensor_act_model_layers_91_residual_pre_attn/max_abs":366,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/range":0.2744140625,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/range":0.2470703125,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_67_residual_post_mlp/norm":556654.3125,"train/tensor_act_model_layers_48_residual_post_mlp/mean":196,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_23_mlp_up_proj_weight/max":0.12890625,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/mean":-0.00010728836059570312,"train/tensor_act_model_layers_82_mlp_down_proj/max_abs":16.75,"train/tensor_act_model_layers_83_residual_pre_attn/norm":679533.9375,"train/tensor_act_model_layers_71_mlp_down_proj/std":0.39453125,"train/tensor_act_model_layers_41_mlp_gate_proj/max_abs":3.703125,"train/tensor_act_model_layers_73_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_residual_pre_attn/max_abs":29,"train/tensor_act_model_layers_29_mlp/max_abs":4,"train/tensor_act_model_layers_75_mlp_gate_proj/mean":-0.03173828125,"train/tensor_act_model_layers_13_mlp_down_proj/min":14.1875,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/max":0.11962890625,"train/tensor_param_model_layers_1_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/norm":6.375,"train/tensor_param_model_layers_86_mlp_down_proj_weight/max_abs":0.1396484375,"train/tensor_param_model_layers_67_mlp_up_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_12_residual_pre_attn/max_abs":49.5,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/max_abs":0.1357421875,"train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_9_mlp/mean":4,"train/tensor_act_model_layers_3_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_65_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_14_residual_post_attn/range":3,"train/tensor_act_model_layers_8_mlp_gate_proj/std":1.859375,"train/tensor_act_model_layers_35_residual_post_mlp/mean":144,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_0_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_49_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_74_mlp_down_proj_weight/norm":6.875,"train/tensor_param_model_layers_57_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_33_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_3_mlp_up_proj_weight/norm":5.75,"train/tensor_act_model_layers_37_mlp_gate_proj/std":1.859375,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_13_mlp_down_proj/max":18,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_53_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_98_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_residual_pre_attn/max":334,"train/tensor_act_model_layers_20_mlp/max":4,"train/tensor_param_model_layers_80_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/mean":0.00128936767578125,"train/tensor_act_model_layers_38_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_79_mlp/norm":8192,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_90_residual_pre_attn/max_abs":362,"train/tensor_act_model_layers_94_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_17_mlp/max_abs":4,"train/tensor_act_model_layers_74_mlp_down_proj/range":3.4375,"train/tensor_param_model_layers_7_mlp_up_proj_weight/range":0.2392578125,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_95_residual_post_mlp/range":6,"train/tensor_act_model_layers_35_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_residual_post_attn/min":165,"train/tensor_act_model_layers_54_mlp_gate_proj/max":3.859375,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_11_mlp_up_proj/min":-3.46875,"train/tensor_param_model_layers_64_mlp_up_proj_weight/mean":0.000804901123046875,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_41_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_4_residual_pre_attn/range":1.125,"train/tensor_act_model_layers_44_mlp_down_proj/norm":30420.595703125,"train/tensor_act_model_layers_11_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_54_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_3_mlp_up_proj/range":6.96875,"train/tensor_act_model_layers_48_mlp_up_proj/min":-3.8125,"train/tensor_act_model_layers_67_residual_post_mlp/mean":272,"train/tensor_param_model_layers_66_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_residual_post_mlp/norm":400885.46875,"train/tensor_act_model_layers_55_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/std":0.033935546875,"train/tensor_act_model_layers_25_mlp_up_proj/max_abs":3.5,"train/tensor_act_model_layers_26_residual_pre_attn/mean":104,"train/tensor_act_model_layers_57_residual_pre_attn/max_abs":230,"train/tensor_act_model_layers_90_residual_post_mlp/range":6,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_24_residual_pre_attn/max_abs":98,"train/tensor_act_model_layers_1_residual_pre_attn/frac_near_user_limit":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_40_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_21_mlp/range":0,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_36_mlp_up_proj_weight/max_abs":0.1298828125,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_79_residual_post_attn/norm":646766,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_92_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_83_mlp_up_proj/mean":-0.03955078125,"train/tensor_act_model_layers_63_residual_pre_attn/max":254,"train/tensor_act_model_layers_96_residual_pre_attn/norm":786029.6875,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_43_mlp/norm":8192,"train/tensor_act_model_layers_97_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp/mean":4,"train/tensor_act_model_layers_78_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_16_mlp_down_proj/std":0.353515625,"train/tensor_act_model_layers_53_residual_post_mlp/max":218,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_90_mlp_down_proj_weight/mean":-0.000797271728515625,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/max":0.1513671875,"train/tensor_act_model_layers_84_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/range":0.2490234375,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_83_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_37_residual_post_mlp/range":5,"train/tensor_act_model_layers_0_residual_post_mlp/max":4.375,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_residual_post_mlp/norm":294374.625,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_19_residual_post_mlp/std":0.58203125,"train/tensor_act_model_layers_42_mlp_down_proj/max_abs":16.625,"train/tensor_act_model_layers_85_residual_post_attn/mean":340,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_45_residual_post_attn/max_abs":182,"train/tensor_param_model_layers_21_mlp_up_proj_weight/mean":0.0001506805419921875,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/norm":13888,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_75_residual_post_mlp/std":0.7578125,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/mean":0.0002307891845703125,"train/tensor_act_model_layers_10_residual_post_attn/max_abs":41.5,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_47_mlp/min":4,"train/tensor_act_model_layers_3_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/max_abs":0.1142578125,"train/tensor_param_model_layers_41_mlp_down_proj_weight/mean":-0.0003795623779296875,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_16_residual_post_mlp/max":69.5,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp/norm":8192,"train/tensor_act_model_layers_44_residual_post_attn/range":5,"train/tensor_act_model_layers_89_residual_post_mlp/range":6,"train/tensor_act_model_layers_7_mlp_down_proj/norm":30510.240234375,"train/tensor_act_model_layers_70_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/mean":0.00106048583984375,"train/tensor_param_model_layers_64_mlp_down_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_34_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp/mean":4,"train/tensor_act_model_layers_14_mlp/min":4,"train/tensor_act_model_layers_39_mlp/std":0,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_11_mlp_up_proj_weight/max_abs":0.12353515625,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_9_mlp/std":0,"train/tensor_param_model_layers_84_mlp_down_proj_weight/norm":7.59375,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_3_mlp_gate_proj/max":3.421875,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_36_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_21_mlp/mean":4,"train/tensor_act_model_layers_73_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_58_mlp_gate_proj/std":1.8203125,"train/tensor_param_model_layers_16_mlp_down_proj_weight/std":0.033935546875,"train/tensor_act_model_layers_57_mlp_up_proj/range":7.578125,"train/tensor_act_model_layers_72_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_23_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_58_mlp_gate_proj/norm":5262.40380859375,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_25_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_34_residual_pre_attn/mean":136,"train/tensor_param_model_layers_80_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp_down_proj/max_abs":17.25,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_94_mlp_gate_proj/mean":-0.07177734375,"train/tensor_act_model_layers_34_residual_post_attn/norm":277989.78125,"train/tensor_param_model_layers_80_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_99_mlp_up_proj_weight/max_abs":0.1513671875,"train/tensor_act_model_layers_98_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_61_residual_post_attn/norm":499204.84375,"train/tensor_act_model_layers_18_residual_post_attn/min":70,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_31_residual_post_mlp/range":4,"train/tensor_act_model_layers_20_residual_pre_attn/std":0.58203125,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/range":0.318359375,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_90_mlp_up_proj/std":1.8515625,"train/tensor_act_model_layers_51_mlp_gate_proj/range":7.125,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/range":0.23193359375,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_98_mlp_gate_proj/max_abs":4.875,"train/tensor_act_model_layers_83_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_37_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_70_mlp_up_proj/min":-4.09375,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_40_mlp_down_proj/range":1.6875,"train/tensor_act_model_layers_46_residual_post_mlp/min":185,"train/tensor_act_model_layers_45_mlp_down_proj/norm":30455.037109375,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_2_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_45_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_residual_post_mlp/norm":687725.9375,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/range":0.25146484375,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_60_mlp_gate_proj/range":7.1875,"train/tensor_act_model_layers_96_residual_post_mlp/mean":388,"train/tensor_act_model_layers_81_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_18_residual_post_attn/max":74,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/min":-0.1494140625,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/norm":6.40625,"train/tensor_act_model_layers_86_mlp_gate_proj/norm":5288.98291015625,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_39_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/norm":6.15625,"train/tensor_act_model_layers_28_mlp/min":4,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_66_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_54_mlp_up_proj/range":7.046875,"train/tensor_param_model_layers_54_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/range":0.28515625,"train/tensor_act_model_layers_67_residual_pre_attn/norm":548462.375,"train/tensor_act_model_layers_50_residual_post_attn/norm":409077.375,"train/tensor_act_model_layers_81_mlp/range":0,"train/tensor_act_model_layers_3_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_82_mlp_up_proj_weight/max_abs":0.1376953125,"train/tensor_act_model_layers_11_residual_post_attn/std":0.396484375,"train/tensor_act_model_layers_28_residual_post_mlp/norm":237010.703125,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_30_mlp_up_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_residual_post_attn/norm":794221.6875,"train/tensor_act_model_layers_70_residual_post_mlp/mean":284,"train/tensor_act_model_layers_70_mlp_up_proj/range":8.015625,"train/tensor_act_model_layers_38_residual_post_mlp/norm":318950.25,"train/tensor_act_model_layers_19_residual_post_attn/min":74,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_75_mlp_down_proj/std":0.38671875,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/max_abs":0.13671875,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_39_mlp_up_proj_weight/norm":6.3125,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_72_residual_pre_attn/norm":589422.1875,"train/tensor_param_model_layers_69_mlp_up_proj_weight/max_abs":0.1279296875,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_36_mlp_up_proj/max_abs":3.609375,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_85_mlp_up_proj/max_abs":4.21875,"train/tensor_act_model_layers_13_residual_post_mlp/range":3,"train/tensor_act_model_layers_33_mlp_up_proj/std":1.8125,"train/tensor_act_model_layers_8_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_91_mlp_up_proj_weight/min":-0.13671875,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/min":-0.1259765625,"train/tensor_act_model_layers_42_mlp_down_proj/max":16.625,"train/tensor_act_model_layers_38_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_5_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_79_mlp_down_proj_weight/norm":7.15625,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/mean":0.0004863739013671875,"train/tensor_act_model_layers_40_mlp_up_proj/mean":-0.2451171875,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/mean":6.4849853515625e-05,"train/tensor_act_model_layers_0_mlp_down_proj/min":8.1875,"train/tensor_param_model_layers_25_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_mlp/max":4,"train/tensor_param_model_layers_31_mlp_down_proj_weight/max":0.125,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_13_residual_post_attn/max":53.5,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_28_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_18_mlp_down_proj_weight/norm":6.28125,"train/tensor_param_model_layers_23_mlp_up_proj_weight/max_abs":0.12890625,"train/tensor_param_model_layers_74_mlp_up_proj_weight/std":0.0380859375,"train/tensor_act_model_layers_8_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_43_residual_pre_attn/max_abs":174,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/norm":5.9375,"train/tensor_act_model_layers_13_residual_post_attn/min":50.5,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_22_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_77_mlp_down_proj_weight/range":0.2685546875,"train/tensor_param_model_layers_63_mlp_up_proj_weight/range":0.25,"train/tensor_param_model_layers_30_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_10_mlp/max":4,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_4_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_73_mlp/max_abs":4,"train/tensor_act_model_layers_97_mlp_gate_proj/max":4.5,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/norm":7.78125,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_4_mlp_up_proj_weight/range":0.228515625,"train/tensor_act_model_layers_56_mlp_gate_proj/min":-3.875,"train/tensor_act_model_layers_35_residual_pre_attn/max":142,"train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_mlp/range":0,"train/tensor_param_model_layers_83_mlp_down_proj_weight/mean":0.000438690185546875,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_42_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_93_residual_post_mlp/max":378,"train/tensor_act_model_layers_61_mlp_gate_proj/max":3.671875,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_84_mlp/mean":4,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_83_mlp_gate_proj/max":4.125,"train/tensor_act_model_layers_38_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_5_residual_pre_attn/max":20.625,"train/tensor_act_model_layers_69_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_79_mlp_down_proj/mean":15,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_52_mlp_up_proj/std":1.8828125,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_57_mlp_gate_proj/mean":-0.154296875,"train/tensor_act_model_layers_19_residual_post_attn/max":78,"train/tensor_act_model_layers_4_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/range":0.25,"train/tensor_param_model_layers_68_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_32_mlp_gate_proj/max":3.90625,"train/tensor_param_model_layers_41_mlp_up_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_93_mlp_up_proj_weight/norm":8.5625,"train/tensor_act_model_layers_89_mlp_up_proj/max":3.859375,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_54_residual_post_attn/mean":216,"train/tensor_act_model_layers_11_residual_pre_attn/norm":90055.6953125,"train/tensor_act_model_layers_32_mlp_gate_proj/range":7.484375,"train/tensor_param_model_layers_68_mlp_up_proj_weight/max_abs":0.1318359375,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/norm":7.59375,"train/tensor_act_model_layers_94_mlp_up_proj/max_abs":4.59375,"train/tensor_act_model_layers_25_mlp/mean":4,"train/tensor_act_model_layers_22_residual_pre_attn/max_abs":90,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/norm":6.40625,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_14_residual_post_mlp/norm":122617.9140625,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/std":0.038330078125,"train/tensor_act_model_layers_11_residual_post_mlp/mean":48,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_17_residual_post_attn/max_abs":69.5,"train/tensor_act_model_layers_0_residual_pre_attn/min":-0.275390625,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_42_mlp_up_proj/std":1.8515625,"train/tensor_act_model_layers_19_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_56_mlp_down_proj_weight/std":0.035400390625,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_2_mlp_up_proj_weight/norm":5.75,"train/tensor_act_model_layers_50_mlp_down_proj/norm":30439.94921875,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/mean":-0.0020294189453125,"train/tensor_act_model_layers_95_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_65_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_13_residual_post_mlp/std":0.46875,"train/tensor_act_model_layers_26_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_98_mlp_gate_proj/range":9.5,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_68_mlp/max":4,"train/tensor_act_model_layers_43_mlp_up_proj/min":-3.5,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/min":-0.11767578125,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_1_mlp_gate_proj/max_abs":4.03125,"train/tensor_act_model_layers_31_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_73_mlp/range":0,"train/tensor_act_model_layers_98_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_59_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_7_mlp/max_abs":4,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_15_mlp_down_proj_weight/range":0.24658203125,"train/tensor_act_model_layers_66_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_77_residual_post_attn/norm":630382.0625,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_64_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_68_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_91_residual_post_attn/range":6,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/min":-0.1376953125,"train/tensor_param_model_layers_54_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_45_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_45_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_7_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_39_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_48_residual_post_mlp/max_abs":198,"train/tensor_act_model_layers_83_residual_post_attn/max_abs":334,"train/tensor_act_model_layers_63_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_77_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_39_mlp_down_proj/max":16.5,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/range":0.25,"train/tensor_act_model_layers_45_mlp_down_proj/max":16.625,"train/tensor_act_model_layers_58_mlp_up_proj/std":1.8046875,"train/tensor_act_model_layers_69_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_94_residual_post_mlp/norm":777837.6875,"train/tensor_act_model_layers_59_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_57_residual_post_attn/norm":466437.09375,"train/tensor_param_model_layers_64_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_17_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_96_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/max":0.14453125,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_94_mlp/max":4,"train/tensor_act_model_layers_28_residual_post_mlp/range":4.5,"train/tensor_act_model_layers_76_residual_post_attn/max_abs":306,"train/tensor_act_model_layers_89_mlp_down_proj/max_abs":17.125,"train/tensor_act_model_layers_9_mlp_gate_proj/min":-3.71875,"train/tensor_act_model_layers_20_mlp_up_proj/mean":0.06298828125,"train/tensor_act_model_layers_96_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_residual_post_mlp/mean":292,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_93_residual_post_attn/norm":761453.75,"train/tensor_param_model_layers_19_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_3_residual_post_attn/norm":24576.236328125,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_74_residual_pre_attn/max":298,"train/tensor_act_model_layers_34_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_24_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_mlp/max_abs":4,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_9_residual_post_attn/mean":36,"train/tensor_act_model_layers_70_residual_pre_attn/norm":573038.25,"train/tensor_act_model_layers_9_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/range":0.244140625,"train/tensor_param_model_layers_91_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_72_mlp/min":4,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/max_abs":0.1416015625,"train/tensor_act_model_layers_44_mlp_down_proj/std":0.359375,"train/tensor_act_model_layers_39_residual_pre_attn/range":5,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_67_mlp/range":0,"train/tensor_act_model_layers_94_residual_post_mlp/mean":380,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_28_residual_post_attn/std":0.67578125,"train/tensor_act_model_layers_70_mlp_gate_proj/max":3.734375,"train/tensor_act_model_layers_14_residual_post_mlp/mean":59.75,"train/tensor_act_model_layers_11_residual_post_mlp/norm":98207.4609375,"train/tensor_act_model_layers_43_residual_post_attn/norm":351717.84375,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_36_mlp_down_proj_weight/max":0.1220703125,"train/tensor_act_model_layers_94_mlp/max_abs":4,"train/tensor_act_model_layers_90_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_47_mlp_up_proj/max_abs":3.5625,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_23_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_86_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_96_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_0_residual_post_mlp/min":3.546875,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_2_mlp/mean":4,"train/tensor_act_model_layers_39_residual_post_mlp/min":157,"train/tensor_act_model_layers_0_mlp_up_proj/std":1.6171875,"train/tensor_act_model_layers_64_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_mlp_up_proj/norm":5284.29443359375,"train/tensor_act_model_layers_64_residual_post_mlp/range":6,"train/tensor_param_model_layers_66_mlp_up_proj_weight/range":0.251953125,"train/tensor_act_model_layers_3_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_89_mlp_gate_proj/std":1.796875,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_down_proj_weight/max":0.12255859375,"train/tensor_act_model_layers_57_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_90_mlp_gate_proj/min":-3.9375,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/max":0.1416015625,"train/tensor_act_model_layers_94_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_63_mlp_down_proj/max_abs":16.375,"train/tensor_act_model_layers_41_mlp_up_proj/max":3.65625,"train/tensor_act_model_layers_36_mlp/range":0,"train/tensor_act_model_layers_83_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_97_mlp_up_proj/range":8.4375,"train/tensor_act_model_layers_9_mlp_down_proj/max":17,"train/tensor_param_model_layers_98_mlp_up_proj_weight/max":0.1484375,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/max":0.1376953125,"train/tensor_act_model_layers_0_mlp/max":4,"train/tensor_param_model_layers_77_mlp_down_proj_weight/std":0.03857421875,"train/tensor_act_model_layers_42_residual_post_mlp/max":174,"train/tensor_act_model_layers_53_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_5_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_18_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_26_mlp_up_proj_weight/norm":6.40625,"train/tensor_param_model_layers_13_mlp_up_proj_weight/max":0.12890625,"train/tensor_act_model_layers_96_mlp_down_proj/norm":30689.609375,"train/tensor_act_model_layers_20_mlp_gate_proj/min":-3.390625,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_84_mlp_gate_proj/range":8.21875,"train/tensor_act_model_layers_71_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_residual_post_mlp/max_abs":402,"train/tensor_param_model_layers_7_mlp_up_proj_weight/mean":0.00055694580078125,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/mean":-0.000881195068359375,"train/tensor_param_model_layers_39_mlp_up_proj_weight/max_abs":0.1240234375,"train/tensor_act_model_layers_11_residual_post_attn/norm":90015.7578125,"train/tensor_act_model_layers_28_mlp_down_proj/range":2.1875,"train/tensor_act_model_layers_32_mlp_up_proj/range":7.140625,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/max":0.1513671875,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/range":0.25048828125,"train/tensor_act_model_layers_41_mlp_down_proj/norm":30441.2890625,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_79_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_21_mlp_up_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_37_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_45_residual_pre_attn/range":5,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/max":0.123046875,"train/tensor_act_model_layers_38_residual_pre_attn/norm":310758.375,"train/tensor_param_model_layers_33_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_44_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_56_residual_post_attn/max_abs":226,"train/tensor_param_model_layers_33_mlp_up_proj_weight/min":-0.1328125,"train/tensor_act_model_layers_83_mlp_up_proj/range":8.1875,"train/tensor_param_model_layers_98_mlp_down_proj_weight/max_abs":0.1513671875,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/norm":6.375,"train/tensor_act_model_layers_65_mlp_gate_proj/norm":5415.71728515625,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_72_residual_post_mlp/norm":597614.1875,"train/tensor_act_model_layers_41_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_61_mlp_up_proj/range":7.28125,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_10_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_63_mlp_up_proj/min":-4.09375,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_99_residual_pre_attn/min":392,"train/tensor_act_model_layers_33_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/max":0.1201171875,"train/tensor_act_model_layers_47_mlp/mean":4,"train/tensor_act_model_layers_80_residual_post_attn/norm":654958,"train/tensor_act_model_layers_88_residual_pre_attn/max":354,"train/tensor_param_model_layers_33_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_37_residual_post_attn/max":150,"train/tensor_act_model_layers_83_mlp_gate_proj/mean":0.1484375,"train/tensor_param_model_layers_52_mlp_up_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_18_mlp_gate_proj/min":-3.71875,"train/tensor_act_model_layers_95_residual_pre_attn/max":382,"train/tensor_act_model_layers_70_residual_post_mlp/norm":581230.25,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/min":-0.1201171875,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_15_mlp/max_abs":4,"train/tensor_act_model_layers_54_mlp_down_proj/max":16.375,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_mlp/norm":73676.0625,"train/tensor_act_model_layers_49_mlp_gate_proj/min":-3.296875,"train/tensor_act_model_layers_35_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_56_residual_pre_attn/max":226,"train/tensor_act_model_layers_96_mlp_gate_proj/norm":5387.6982421875,"train/tensor_act_model_layers_0_mlp/min":3.875,"train/tensor_param_model_layers_11_mlp_down_proj_weight/range":0.224609375,"train/tensor_param_model_layers_89_mlp_up_proj_weight/min":-0.1494140625,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/range":0.2490234375,"train/tensor_act_model_layers_54_residual_post_attn/range":5,"train/tensor_act_model_layers_33_residual_post_attn/range":4,"train/tensor_act_model_layers_13_mlp_gate_proj/max":3.671875,"train/tensor_act_model_layers_42_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_90_mlp_down_proj_weight/min":-0.138671875,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_21_mlp/std":0,"train/tensor_act_model_layers_57_residual_post_mlp/min":229,"train/tensor_act_model_layers_16_mlp_up_proj/max_abs":3.78125,"train/tensor_act_model_layers_44_mlp_up_proj/max":3.65625,"train/tensor_param_model_layers_90_mlp_down_proj_weight/range":0.283203125,"train/tensor_act_model_layers_77_mlp_down_proj/std":0.51171875,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_65_mlp_down_proj/norm":30516.544921875,"train/tensor_act_model_layers_56_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_79_mlp_up_proj_weight/mean":-0.000469207763671875,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp_gate_proj/norm":5346.8935546875,"train/tensor_act_model_layers_45_residual_post_attn/norm":368101.6875,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_66_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_30_residual_post_mlp/std":0.68359375,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_9_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_97_mlp_gate_proj/mean":-0.04296875,"train/tensor_act_model_layers_62_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_71_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_residual_pre_attn/range":5,"train/tensor_act_model_layers_6_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_7_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_85_residual_pre_attn/norm":695917.875,"train/tensor_act_model_layers_43_residual_post_mlp/mean":176,"train/tensor_act_model_layers_52_mlp/max":4,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_4_residual_post_attn/std":0.19140625,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_mlp/min":237,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_39_residual_post_attn/norm":318950.25,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_34_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/std":0.034423828125,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/norm":6.875,"train/tensor_act_model_layers_96_mlp/std":0,"train/tensor_act_model_layers_24_mlp/std":0,"train/tensor_act_model_layers_53_mlp_gate_proj/min":-4,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/mean":0.000827789306640625,"train/tensor_act_model_layers_74_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_15_mlp/mean":4,"train/tensor_param_model_layers_90_mlp_up_proj_weight/max":0.13671875,"train/tensor_act_model_layers_81_mlp/max":4,"train/tensor_act_model_layers_68_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_46_residual_post_mlp/max_abs":190,"train/tensor_act_model_layers_30_mlp_down_proj/std":0.34765625,"train/tensor_act_model_layers_34_mlp_down_proj/range":2,"train/tensor_act_model_layers_56_residual_post_attn/range":5,"train/tensor_act_model_layers_43_mlp_up_proj/norm":5473.888671875,"train/tensor_param_model_layers_14_mlp_down_proj_weight/max":0.115234375,"train/tensor_act_model_layers_70_mlp_gate_proj/mean":0.032470703125,"train/tensor_param_model_layers_46_mlp_down_proj_weight/std":0.03466796875,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_14_mlp/std":0,"train/tensor_param_model_layers_76_mlp_up_proj_weight/max_abs":0.1328125,"train/tensor_act_model_layers_40_residual_post_attn/max_abs":162,"train/tensor_act_model_layers_57_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_54_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_71_mlp/mean":4,"train/tensor_act_model_layers_32_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp/max_abs":4,"train/tensor_param_model_layers_54_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/std":0.04052734375,"train/tensor_act_model_layers_96_residual_post_attn/mean":384,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_mlp_up_proj/max":4.03125,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/std":0.0390625,"train/tensor_act_model_layers_5_mlp_gate_proj/max_abs":3.25,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_mlp_up_proj/min":-3.625,"train/tensor_act_model_layers_2_residual_post_mlp/mean":12,"train/tensor_param_model_layers_44_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_5_residual_post_mlp/max_abs":24.875,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_29_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_59_mlp/std":0,"train/tensor_act_model_layers_70_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_72_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_49_mlp_down_proj/range":1.625,"train/tensor_act_model_layers_66_residual_pre_attn/mean":264,"train/tensor_act_model_layers_25_mlp_gate_proj/std":1.84375,"train/tensor_act_model_layers_73_residual_post_mlp/max":298,"train/tensor_act_model_layers_75_mlp_up_proj/max":4.40625,"train/tensor_param_model_layers_49_mlp_down_proj_weight/norm":6.25,"train/tensor_act_model_layers_89_residual_pre_attn/min":352,"train/tensor_act_model_layers_63_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_0_residual_pre_attn/range":0.546875,"train/tensor_param_model_layers_46_mlp_up_proj_weight/norm":6.34375,"train/tensor_param_model_layers_90_mlp_up_proj_weight/norm":8.125,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/mean":-0.000720977783203125,"train/tensor_act_model_layers_44_mlp/max_abs":4,"train/tensor_act_model_layers_93_residual_pre_attn/mean":372,"train/tensor_param_model_layers_85_mlp_down_proj_weight/max_abs":0.1435546875,"train/tensor_act_model_layers_20_mlp/mean":4,"train/tensor_act_model_layers_85_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_34_mlp_gate_proj/std":1.828125,"_step":3464,"train/tensor_act_model_layers_78_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp_gate_proj/range":7.25,"train/tensor_act_model_layers_98_mlp_up_proj/mean":0.111328125,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_18_mlp_up_proj/min":-3.71875,"train/tensor_act_model_layers_78_mlp_gate_proj/min":-3.609375,"train/tensor_act_model_layers_99_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_48_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_52_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp_up_proj/range":7.125,"train/tensor_param_model_layers_84_mlp_up_proj_weight/range":0.2744140625,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_0_mlp_down_proj_weight/norm":10.375,"train/tensor_act_model_layers_22_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp/range":0,"train/tensor_act_model_layers_35_mlp_gate_proj/std":1.8515625,"train/tensor_act_model_layers_23_residual_post_mlp/std":0.58203125,"train/tensor_act_model_layers_35_mlp_up_proj/min":-4.03125,"train/tensor_act_model_layers_71_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_27_residual_pre_attn/min":106,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_mlp_up_proj/mean":0.001708984375,"train/tensor_act_model_layers_28_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_40_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_72_mlp_down_proj_weight/norm":6.8125,"train/tensor_param_model_layers_22_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_56_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_72_mlp_down_proj/max_abs":17.125,"train/tensor_act_model_layers_69_residual_post_mlp/norm":573038.25,"train/tensor_param_model_layers_75_mlp_up_proj_weight/range":0.2626953125,"train/tensor_act_model_layers_71_mlp_down_proj/max_abs":16,"train/tensor_act_model_layers_70_mlp_down_proj/max":16.375,"train/tensor_act_model_layers_44_residual_post_mlp/max":182,"train/tensor_act_model_layers_61_residual_post_mlp/max_abs":250,"train/tensor_act_model_layers_79_residual_post_mlp/min":316,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_67_mlp_down_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_12_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_85_mlp_up_proj_weight/norm":7.6875,"train/tensor_act_model_layers_56_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp/max":4,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/mean":-0.001678466796875,"train/tensor_param_model_layers_4_mlp_down_proj_weight/range":0.22607421875,"train/tensor_param_model_layers_37_mlp_up_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_43_mlp/max_abs":4,"train/tensor_act_model_layers_96_mlp/max":4,"train/tensor_act_model_layers_65_residual_post_mlp/range":6,"train/tensor_act_model_layers_99_residual_post_mlp/std":0.7578125,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_97_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_45_residual_pre_attn/mean":180,"train/tensor_act_model_layers_4_residual_pre_attn/norm":32756.287109375,"train/tensor_param_model_layers_99_mlp_down_proj_weight/mean":-0.00011491775512695312,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_29_residual_pre_attn/std":0.67578125,"train/tensor_act_model_layers_11_mlp_gate_proj/max":3.421875,"train/tensor_act_model_layers_65_residual_pre_attn/max_abs":262,"train/tensor_act_model_layers_15_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_26_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_74_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/max":0.12890625,"train_samples_per_second":632.92,"train/tensor_act_model_layers_30_mlp/range":0,"train/tensor_param_model_layers_75_mlp_up_proj_weight/max_abs":0.1318359375,"train/tensor_act_model_layers_52_residual_post_mlp/max":214,"train/tensor_act_model_layers_88_mlp_gate_proj/max":4.03125,"train/tensor_param_model_layers_63_mlp_down_proj_weight/range":0.24853515625,"train/tensor_act_model_layers_91_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_32_residual_pre_attn/mean":127.5,"train/tensor_act_model_layers_10_residual_post_attn/min":38.5,"train/tensor_act_model_layers_39_mlp_up_proj/range":7.171875,"train/tensor_param_model_layers_36_mlp_up_proj_weight/mean":-0.00112152099609375,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_59_mlp_gate_proj/range":7.375,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_45_mlp_up_proj/range":7.984375,"train/tensor_act_model_layers_7_residual_post_attn/min":26.875,"train/tensor_act_model_layers_85_mlp_up_proj/norm":5354.154296875,"train/tensor_act_model_layers_50_residual_post_attn/range":5,"train/tensor_act_model_layers_79_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_64_residual_post_mlp/min":256,"train/tensor_act_model_layers_24_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_53_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/max_abs":0.1513671875,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_76_mlp_up_proj_weight/mean":0.000278472900390625,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_76_mlp_down_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_53_residual_post_attn/max_abs":214,"train/tensor_act_model_layers_43_residual_post_attn/std":0.80859375,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_85_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_15_mlp_down_proj_weight/std":0.03369140625,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_53_mlp_down_proj_weight/range":0.244140625,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/range":0.251953125,"train/tensor_act_model_layers_2_residual_post_attn/range":1.03125,"train/tensor_param_model_layers_33_mlp_up_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_59_mlp/mean":4,"train/tensor_act_model_layers_58_mlp_down_proj/max_abs":16.125,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_58_residual_post_attn/range":5,"train/tensor_act_model_layers_85_residual_post_mlp/min":340,"train/tensor_act_model_layers_21_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_86_residual_pre_attn/min":340,"train/tensor_param_model_layers_51_mlp_up_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_33_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/mean":0.00049591064453125,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/range":0.251953125,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_10_residual_post_attn/max":41.5,"train/tensor_act_model_layers_26_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp/min":4,"train/tensor_act_model_layers_69_mlp_up_proj/min":-3.796875,"train/tensor_act_model_layers_7_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_37_mlp_up_proj/min":-4,"train/tensor_act_model_layers_41_residual_post_mlp/range":5,"train/tensor_act_model_layers_63_mlp/min":4,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_46_residual_pre_attn/mean":184,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_5_mlp_down_proj_weight/mean":0.0013427734375,"train/tensor_act_model_layers_36_residual_pre_attn/mean":144,"train/tensor_param_model_layers_87_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_10_mlp_gate_proj/max":3.296875,"train/tensor_act_model_layers_74_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/max_abs":0.16796875,"train/tensor_act_model_layers_49_residual_post_mlp/max_abs":202,"train/tensor_act_model_layers_14_residual_post_mlp/min":58.5,"train/tensor_param_model_layers_22_mlp_up_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_14_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_32_mlp_down_proj/max":16.25,"train/tensor_act_model_layers_47_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_94_mlp_up_proj/std":1.828125,"train/tensor_param_model_layers_11_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_54_mlp_up_proj_weight/min":-0.12451171875,"train/tensor_act_model_layers_72_mlp_gate_proj/std":1.8203125,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_41_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_86_mlp_down_proj/norm":30874.01171875,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_mlp_up_proj/max":3.34375,"train/tensor_param_model_layers_44_mlp_down_proj_weight/mean":0.000652313232421875,"train/tensor_act_model_layers_4_residual_post_mlp/mean":20,"train/tensor_act_model_layers_83_mlp_down_proj/max":16.625,"train/tensor_param_model_layers_21_mlp_down_proj_weight/norm":6.28125,"train/tensor_act_model_layers_45_mlp_gate_proj/norm":5385.80517578125,"train/tensor_act_model_layers_53_residual_post_attn/range":5,"train/tensor_param_model_layers_10_mlp_down_proj_weight/max":0.12451171875,"train/tensor_act_model_layers_13_mlp/min":4,"train/tensor_act_model_layers_23_mlp_gate_proj/std":1.890625,"train/tensor_act_model_layers_61_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_13_mlp_up_proj/std":1.84375,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_4_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_49_residual_post_attn/min":193,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_33_mlp_up_proj_weight/range":0.25537109375,"train/tensor_act_model_layers_50_residual_pre_attn/range":5,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_97_mlp_down_proj/max":18,"train/tensor_act_model_layers_76_mlp_gate_proj/mean":-0.2353515625,"train/tensor_act_model_layers_96_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_51_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_69_residual_pre_attn/norm":564846.3125,"train/tensor_act_model_layers_56_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_45_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_5_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_69_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_residual_post_mlp/min":233,"train/tensor_act_model_layers_59_mlp_gate_proj/min":-3.34375,"train/tensor_act_model_layers_77_mlp_up_proj/max_abs":4.125,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_7_residual_post_mlp/range":2.625,"train/tensor_act_model_layers_60_mlp_gate_proj/max_abs":3.609375,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_78_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_82_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_68_residual_pre_attn/min":268,"train/tensor_act_model_layers_26_mlp_down_proj/max_abs":16.5,"train/tensor_act_model_layers_58_mlp_down_proj/norm":30468.48828125,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_73_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_99_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_80_mlp/mean":4,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_84_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_42_mlp_gate_proj/max":3.6875,"train/tensor_act_model_layers_13_residual_pre_attn/min":50.5,"train/tensor_act_model_layers_97_mlp_down_proj/min":14,"train/tensor_act_model_layers_45_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_10_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_8_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_49_residual_post_attn/max":198,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_83_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_87_residual_pre_attn/max":350,"train/tensor_act_model_layers_11_residual_post_mlp/max":49.5,"train/tensor_act_model_layers_61_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_47_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_25_mlp_up_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/std":0.04833984375,"train/tensor_act_model_layers_61_residual_pre_attn/range":5,"train/tensor_param_model_layers_44_mlp_up_proj_weight/mean":0.001434326171875,"train/tensor_act_model_layers_21_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/min":-0.1552734375,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/std":0.05712890625,"train/tensor_act_model_layers_8_residual_pre_attn/range":2.625,"train/tensor_act_model_layers_83_residual_post_attn/mean":332,"train/tensor_act_model_layers_34_residual_pre_attn/range":4,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/min":-0.11669921875,"train/tensor_act_model_layers_9_mlp_down_proj/norm":30601.953125,"train/tensor_param_model_layers_56_mlp_up_proj_weight/mean":0.0003643035888671875,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_74_mlp_up_proj_weight/min":-0.1318359375,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_10_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_70_mlp/range":0,"train/tensor_act_model_layers_11_mlp_up_proj/range":6.921875,"train/tensor_act_model_layers_2_residual_pre_attn/norm":16381.2919921875,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_38_mlp_gate_proj/min":-3.5625,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_92_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp_up_proj/mean":0.0267333984375,"train/tensor_act_model_layers_11_mlp_gate_proj/range":6.90625,"train/tensor_param_model_layers_48_mlp_down_proj_weight/min":-0.12890625,"train/tensor_param_model_layers_91_mlp_up_proj_weight/std":0.0458984375,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_44_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_31_mlp_down_proj_weight/range":0.25,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_34_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_12_residual_post_mlp/min":50.5,"train/tensor_act_model_layers_20_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_mlp_gate_proj/max":3.875,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_87_mlp_up_proj/mean":0.2138671875,"train/tensor_act_model_layers_44_residual_pre_attn/mean":176,"train/tensor_act_model_layers_27_mlp_up_proj/max_abs":4.53125,"train/tensor_act_model_layers_45_mlp_gate_proj/max":3.328125,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_35_residual_pre_attn/max_abs":142,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/max":0.1279296875,"train/tensor_act_model_layers_88_mlp_up_proj/range":7.984375,"train/tensor_act_model_layers_13_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_15_mlp_up_proj/norm":5354.9267578125,"train/tensor_act_model_layers_11_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_86_mlp_down_proj_weight/range":0.27734375,"train/tensor_act_model_layers_51_residual_pre_attn/max":206,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_81_mlp_up_proj_weight/max":0.1396484375,"train/tensor_act_model_layers_18_mlp/mean":4,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_1_residual_pre_attn/range":0.828125,"train/tensor_act_model_layers_69_residual_post_mlp/mean":280,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_17_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_residual_post_mlp/range":5,"train/tensor_act_model_layers_77_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_12_residual_post_mlp/range":3,"train/tensor_act_model_layers_44_mlp_gate_proj/norm":5433.77880859375,"train/tensor_act_model_layers_73_residual_post_mlp/range":6,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_47_residual_pre_attn/mean":188,"train/tensor_param_model_layers_61_mlp_down_proj_weight/range":0.24755859375,"train/tensor_param_model_layers_65_mlp_down_proj_weight/norm":6.375,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_10_mlp_up_proj/range":6.4375,"train/tensor_act_model_layers_91_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_84_mlp_up_proj/norm":5156.427734375,"train/tensor_param_model_layers_19_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_63_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_60_mlp_up_proj/mean":0.0166015625,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_75_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_18_mlp_up_proj_weight/max_abs":0.1328125,"train/tensor_act_model_layers_32_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_47_residual_post_attn/range":5,"train/tensor_param_model_layers_30_mlp_up_proj_weight/min":-0.125,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_22_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_79_residual_pre_attn/range":6,"train/tensor_param_model_layers_40_mlp_up_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_78_mlp_gate_proj/range":7.390625,"train/tensor_param_model_layers_1_mlp_down_proj_weight/norm":6.03125,"train/tensor_param_model_layers_3_mlp_down_proj_weight/norm":5.625,"train/tensor_act_model_layers_34_residual_post_mlp/max":142,"train/tensor_act_model_layers_88_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_81_mlp_up_proj_weight/range":0.2705078125,"train/tensor_act_model_layers_55_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_param_model_layers_61_mlp_down_proj_weight/max":0.123046875,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_92_mlp_up_proj/mean":-0.0172119140625,"train/tensor_act_model_layers_57_residual_pre_attn/range":5,"train/tensor_act_model_layers_43_mlp/range":0,"train/tensor_act_model_layers_66_residual_pre_attn/min":260,"train/tensor_act_model_layers_36_mlp_gate_proj/std":1.828125,"train/tensor_param_model_layers_42_mlp_down_proj_weight/max":0.125,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_18_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_3_residual_post_mlp/std":0.1611328125,"train/tensor_param_model_layers_84_mlp_up_proj_weight/norm":7.625,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/std":0.0390625,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_78_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_10_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_97_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_12_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_7_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_67_mlp/max_abs":4,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_3_mlp_down_proj_weight/range":0.20654296875,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp_down_proj/mean":15.125,"train/tensor_act_model_layers_63_mlp_up_proj/range":8.1875,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp_up_proj/max":3.4375,"train/tensor_param_model_layers_19_mlp_up_proj_weight/max":0.1220703125,"train/tensor_act_model_layers_2_mlp_gate_proj/mean":-0.12158203125,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_93_residual_pre_attn/range":6,"train/tensor_act_model_layers_95_mlp_down_proj/range":3.5625,"train/tensor_param_model_layers_37_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_17_residual_post_mlp/min":70,"train/tensor_act_model_layers_77_residual_pre_attn/norm":630382.0625,"train/tensor_act_model_layers_93_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_22_mlp_down_proj_weight/range":0.248046875,"train/tensor_param_model_layers_82_mlp_down_proj_weight/range":0.26171875,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_50_mlp_gate_proj/norm":5417.3828125,"train/tensor_act_model_layers_9_mlp_gate_proj/range":7.34375,"train/tensor_param_model_layers_28_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_67_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_13_mlp_down_proj_weight/min":-0.12353515625,"train/tensor_act_model_layers_48_mlp/norm":8192,"train/tensor_act_model_layers_10_residual_post_mlp/min":42.5,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_16_mlp_up_proj/mean":-0.107421875,"train/tensor_act_model_layers_87_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_residual_post_mlp/norm":359909.75,"train/tensor_act_model_layers_67_residual_post_attn/range":6,"train/tensor_act_model_layers_50_mlp_gate_proj/range":7.3125,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_96_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_mlp_down_proj/range":2.125,"train/tensor_param_model_layers_25_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_76_mlp_up_proj/max":3.765625,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_94_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_residual_pre_attn/mean":300,"train/tensor_param_model_layers_62_mlp_up_proj_weight/norm":6.5,"train/tensor_act_model_layers_77_residual_pre_attn/range":6,"train/tensor_act_model_layers_51_mlp_up_proj/min":-3.796875,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_52_mlp_gate_proj/max_abs":3.546875,"train/tensor_param_model_layers_9_mlp_down_proj_weight/min":-0.125,"train/tensor_param_model_layers_52_mlp_down_proj_weight/std":0.03466796875,"train/tensor_param_model_layers_93_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_mlp_up_proj/range":6.9375,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_46_mlp/std":0,"train/tensor_act_model_layers_95_residual_post_attn/range":6,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_65_mlp_up_proj_weight/norm":6.5,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_97_mlp_up_proj_weight/max":0.150390625,"train/tensor_act_model_layers_1_residual_post_attn/min":3.515625,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_69_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp/max":4,"train/tensor_act_model_layers_99_mlp_up_proj/norm":5362.15966796875,"train/tensor_act_model_layers_94_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_46_residual_post_attn/max_abs":186,"train/tensor_act_model_layers_38_residual_post_mlp/min":153,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_17_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_22_mlp_up_proj/norm":5302.3505859375,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/std":0.03271484375,"train/tensor_act_model_layers_74_mlp_down_proj/max_abs":17.625,"train/tensor_act_model_layers_16_mlp/range":0,"train/tensor_act_model_layers_93_mlp_down_proj/std":0.55078125,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_75_residual_pre_attn/min":296,"train/tensor_param_model_layers_99_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_82_residual_post_attn/mean":328,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_10_mlp_down_proj/range":1.875,"train/tensor_act_model_layers_19_residual_pre_attn/min":74,"train/tensor_act_model_layers_42_mlp_gate_proj/max_abs":3.890625,"train/tensor_act_model_layers_8_mlp_up_proj/mean":-0.0135498046875,"train/tensor_act_model_layers_23_mlp_gate_proj/max_abs":4.03125,"train/tensor_act_model_layers_29_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_17_mlp_down_proj/max_abs":16.25,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_25_mlp_up_proj/min":-3.5,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/std":0.04150390625,"train/tensor_act_model_layers_69_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_13_mlp_up_proj/range":7.46875,"train/tensor_act_model_layers_77_residual_post_mlp/mean":312,"eval/steps_per_second":0.619,"train/tensor_act_model_layers_98_mlp/min":4,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_43_mlp_down_proj/range":2,"train/tensor_act_model_layers_59_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_38_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_99_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_19_residual_pre_attn/max_abs":78,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/norm":6.375,"train/tensor_act_model_layers_25_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/range":0.25,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_post_mlp/range":4,"train/tensor_act_model_layers_2_residual_post_attn/std":0.134765625,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_85_residual_post_mlp/norm":704109.875,"train/tensor_act_model_layers_22_mlp_up_proj/std":1.828125,"train/tensor_act_model_layers_42_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_residual_pre_attn/range":6,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_27_residual_post_attn/max_abs":110,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_66_mlp_up_proj/std":1.8046875,"train/tensor_act_model_layers_79_residual_post_attn/max_abs":318,"train/tensor_act_model_layers_55_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_residual_post_mlp/max_abs":226,"train/tensor_act_model_layers_16_residual_post_mlp/range":3.5,"train/tensor_act_model_layers_5_mlp_gate_proj/std":1.7890625,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_70_mlp/max":4,"train/tensor_param_model_layers_81_mlp_up_proj_weight/max_abs":0.1396484375,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_90_residual_post_attn/max_abs":362,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp/mean":4,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/std":0.03515625,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_66_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_76_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_71_mlp/max_abs":4,"train/tensor_act_model_layers_9_mlp_up_proj/mean":-0.05859375,"train/tensor_act_model_layers_13_mlp/norm":8192,"total_flos":1.1291993309184e+16,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_9_mlp/max_abs":4,"train/tensor_act_model_layers_61_mlp_gate_proj/norm":5370.19580078125,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_94_residual_pre_attn/min":372,"train/tensor_param_model_layers_50_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_32_residual_post_mlp/max":134,"train/tensor_act_model_layers_15_residual_pre_attn/max_abs":61.5,"train/tensor_act_model_layers_24_mlp_gate_proj/norm":5333.0693359375,"train/tensor_act_model_layers_92_residual_post_mlp/range":6,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_69_mlp_down_proj_weight/max_abs":0.1318359375,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_23_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_85_mlp_down_proj_weight/std":0.042236328125,"train/tensor_act_model_layers_87_mlp_gate_proj/mean":0.043701171875,"train/tensor_act_model_layers_60_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_45_mlp_down_proj/range":2.375,"train/tensor_act_model_layers_77_residual_post_mlp/norm":638574.0625,"train/tensor_act_model_layers_42_mlp_up_proj/norm":5370.0849609375,"train/tensor_act_model_layers_43_residual_pre_attn/norm":351717.84375,"train/tensor_act_model_layers_63_mlp_gate_proj/max":4.03125,"train/tensor_act_model_layers_33_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_72_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_52_mlp/max_abs":4,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_85_residual_post_mlp/mean":344,"train/tensor_act_model_layers_21_mlp_gate_proj/mean":0.024658203125,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_1_residual_pre_attn/max":4.375,"train/tensor_act_model_layers_29_mlp_up_proj/mean":-0.0264892578125,"train/tensor_act_model_layers_88_mlp/min":4,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_80_residual_pre_attn/range":6,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_35_residual_post_mlp/range":5,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_74_mlp/mean":4,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/range":0.24755859375,"train/tensor_param_model_layers_57_mlp_down_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_15_mlp_gate_proj/max":3.546875,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/max":0.1474609375,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_28_residual_post_mlp/max_abs":118.5,"train/tensor_act_model_layers_95_mlp_gate_proj/std":1.828125,"train/tensor_param_model_layers_96_mlp_down_proj_weight/max_abs":0.1435546875,"train/tensor_act_model_layers_59_mlp/max":4,"train/tensor_param_model_layers_75_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_5_mlp_gate_proj/range":6.46875,"train/tensor_act_model_layers_51_mlp/max_abs":4,"train/tensor_act_model_layers_14_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_95_mlp_down_proj_weight/mean":0.00012874603271484375,"train/tensor_act_model_layers_17_mlp_gate_proj/range":7.296875,"train/tensor_act_model_layers_11_residual_pre_attn/range":3,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/min":-0.140625,"train/tensor_act_model_layers_8_residual_pre_attn/max":33.5,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_6_mlp/mean":4,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/max":0.1240234375,"train/tensor_act_model_layers_57_mlp_down_proj/min":14.3125,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_94_mlp_up_proj/mean":-0.203125,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/std":0.033935546875,"train/tensor_act_model_layers_64_residual_post_mlp/max":262,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_58_mlp_down_proj_weight/min":-0.1318359375,"train/tensor_act_model_layers_56_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_residual_post_attn/norm":810605.625,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_47_mlp_up_proj/range":6.984375,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_88_mlp_up_proj/mean":-0.034912109375,"train/tensor_param_model_layers_59_mlp_up_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_61_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_76_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_39_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_82_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_0_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_98_mlp_up_proj/std":1.84375,"train/tensor_param_model_layers_37_mlp_down_proj_weight/mean":-0.000896453857421875,"train/tensor_act_model_layers_24_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_21_residual_post_mlp/std":0.58203125,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_68_mlp_up_proj/min":-3.546875,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_25_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_51_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_0_mlp_up_proj/mean":-0.008544921875,"train/tensor_act_model_layers_10_residual_post_mlp/norm":90055.6953125,"train/tensor_act_model_layers_60_mlp_gate_proj/norm":5425.6279296875,"train/tensor_param_model_layers_41_mlp_down_proj_weight/max":0.1162109375,"train/tensor_act_model_layers_10_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_20_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_97_mlp_down_proj/range":4,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_56_mlp_up_proj/mean":0.04638671875,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/mean":0.0010528564453125,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_17_mlp/max":4,"train/tensor_act_model_layers_3_mlp_down_proj/min":14.1875,"train/tensor_param_model_layers_1_mlp_down_proj_weight/mean":-0.0016632080078125,"train/tensor_param_model_layers_3_mlp_down_proj_weight/mean":0.0028533935546875,"train/tensor_act_model_layers_33_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_23_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_83_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_14_mlp_down_proj/range":2.25,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_57_mlp_down_proj_weight/mean":0.0007781982421875,"train/tensor_param_model_layers_11_mlp_down_proj_weight/std":0.032958984375,"train/tensor_act_model_layers_91_mlp/mean":4,"train/tensor_act_model_layers_56_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_93_mlp_down_proj_weight/max_abs":0.14453125,"train/tensor_act_model_layers_54_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/max":0.125,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_85_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_56_residual_post_attn/mean":224,"train/tensor_act_model_layers_24_mlp/norm":8192,"train/tensor_param_model_layers_52_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_8_mlp_down_proj_weight/min":-0.11865234375,"train/tensor_act_model_layers_85_residual_post_mlp/max":346,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_40_mlp/range":0,"train/tensor_act_model_layers_45_mlp_up_proj/min":-3.828125,"train/tensor_act_model_layers_48_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_65_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_7_mlp_down_proj/max_abs":16,"train/tensor_param_model_layers_92_mlp_down_proj_weight/mean":-0.00128173828125,"train/tensor_act_model_layers_43_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_mlp_gate_proj/max_abs":4.21875,"train/tensor_act_model_layers_84_residual_post_attn/norm":687725.9375,"train/tensor_act_model_layers_15_mlp_gate_proj/norm":5404.5078125,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_24_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/norm":6.1875,"train/tensor_act_model_layers_60_residual_post_attn/norm":491012.90625,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/max_abs":0.130859375,"train/tensor_act_model_layers_47_mlp_up_proj/min":-3.5625,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_mlp_gate_proj/max_abs":3.75,"train/tensor_act_model_layers_47_residual_post_mlp/std":0.80859375,"train/tensor_act_model_layers_76_residual_pre_attn/min":300,"train/tensor_act_model_layers_70_residual_pre_attn/range":6,"train/tensor_act_model_layers_81_mlp_up_proj/mean":0.1005859375,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_61_mlp_down_proj/range":1.875,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_23_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp_up_proj/max_abs":3.625,"train/tensor_act_model_layers_17_mlp_gate_proj/std":1.7890625,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_84_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_58_mlp_down_proj/range":1.8125,"train/tensor_act_model_layers_56_mlp_gate_proj/mean":0.080078125,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_7_mlp_down_proj_weight/std":0.03271484375,"train/tensor_act_model_layers_78_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_19_mlp_up_proj/max_abs":3.6875,"train/tensor_act_model_layers_6_mlp_gate_proj/std":1.8359375,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_31_residual_pre_attn/min":122,"train/tensor_act_model_layers_5_residual_pre_attn/std":0.193359375,"train/tensor_param_model_layers_53_mlp_down_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_45_residual_pre_attn/min":177,"train/tensor_act_model_layers_73_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_95_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_18_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_1_mlp_up_proj_weight/max":0.125,"train/tensor_param_model_layers_97_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_32_residual_post_mlp/min":130,"train/tensor_act_model_layers_77_residual_post_attn/min":304,"train/tensor_act_model_layers_98_mlp_gate_proj/max":4.625,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_95_mlp_gate_proj/mean":-0.1396484375,"train/tensor_act_model_layers_73_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_77_residual_post_attn/max":310,"train/tensor_param_model_layers_14_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_28_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_49_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_73_mlp_up_proj/std":1.859375,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_8_residual_pre_attn/norm":65482.046875,"train/tensor_act_model_layers_82_residual_post_mlp/max_abs":334,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_44_residual_post_mlp/range":5,"train/tensor_act_model_layers_39_mlp/max_abs":4,"train/tensor_act_model_layers_7_mlp_up_proj/range":7.203125,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_89_mlp_up_proj_weight/range":0.2900390625,"train/tensor_param_model_layers_3_mlp_up_proj_weight/min":-0.1064453125,"train/tensor_param_model_layers_21_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_71_mlp_down_proj_weight/min":-0.1337890625,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_59_residual_pre_attn/range":5,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/max_abs":0.1376953125,"train/tensor_param_model_layers_45_mlp_down_proj_weight/max_abs":0.1240234375,"train/tensor_param_model_layers_29_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_24_mlp_down_proj/max":15.8125,"train/tensor_act_model_layers_42_mlp_gate_proj/range":7.578125,"train/tensor_act_model_layers_0_mlp_gate_proj/max":6.25,"train/tensor_act_model_layers_67_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_mlp_down_proj/std":0.369140625,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/std":0.0380859375,"train/tensor_act_model_layers_14_mlp_up_proj/max":3.109375,"train/tensor_act_model_layers_76_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_22_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_30_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_89_mlp/max":4,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_57_residual_post_mlp/max":234,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_55_residual_post_attn/norm":450021.09375,"train/tensor_act_model_layers_49_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_40_residual_pre_attn/range":5,"train/tensor_param_model_layers_62_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_65_residual_post_mlp/std":0.7578125,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/mean":0.000606536865234375,"train/tensor_act_model_layers_86_mlp_down_proj/range":3.625,"train/tensor_param_model_layers_44_mlp_down_proj_weight/max_abs":0.130859375,"train/tensor_param_model_layers_27_mlp_down_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_88_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_17_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_38_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_84_mlp_up_proj/min":-4.28125,"train/tensor_act_model_layers_26_residual_pre_attn/max":106,"train/tensor_param_model_layers_95_mlp_up_proj_weight/max":0.142578125,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_9_mlp_down_proj_weight/range":0.248046875,"train/tensor_act_model_layers_30_residual_post_mlp/range":4.5,"train/tensor_act_model_layers_15_residual_post_attn/mean":59.75,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_97_residual_pre_attn/mean":388,"train/tensor_param_model_layers_89_mlp_down_proj_weight/std":0.04443359375,"train/tensor_param_model_layers_91_mlp_up_proj_weight/mean":0.00171661376953125,"train/tensor_act_model_layers_97_mlp/range":0,"train/tensor_act_model_layers_83_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_80_mlp_down_proj/std":0.4453125,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/min":-0.1298828125,"train/tensor_param_model_layers_94_mlp_down_proj_weight/std":0.046630859375,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_88_residual_post_attn/mean":352,"train/tensor_param_model_layers_69_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_residual_post_attn/max_abs":386,"train/tensor_act_model_layers_12_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_37_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp_up_proj/norm":5413.34228515625,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_48_residual_pre_attn/max":194,"train/tensor_param_model_layers_22_mlp_up_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_40_residual_post_mlp/std":0.80859375,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_33_mlp_down_proj/norm":30439.21484375,"train/tensor_act_model_layers_2_mlp_gate_proj/range":6.9375,"train/tensor_act_model_layers_27_mlp/norm":8192,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_23_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_21_mlp/max_abs":4,"train/tensor_act_model_layers_42_residual_post_attn/range":5,"train/tensor_act_model_layers_55_mlp_gate_proj/min":-3.515625,"train/tensor_act_model_layers_91_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_22_mlp_down_proj/range":2.25,"train/tensor_act_model_layers_6_residual_pre_attn/min":23.125,"train/tensor_act_model_layers_70_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_mlp_gate_proj/max":3.4375,"train/tensor_act_model_layers_48_mlp_up_proj/max_abs":3.953125,"train/tensor_act_model_layers_64_residual_pre_attn/max_abs":258,"train/tensor_param_model_layers_62_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_38_mlp_down_proj_weight/norm":6.21875,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_88_mlp_down_proj_weight/min":-0.142578125,"train/tensor_param_model_layers_76_mlp_down_proj_weight/mean":0.001556396484375,"train/tensor_act_model_layers_54_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_95_mlp/range":0,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_59_residual_post_mlp/max":242,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_22_mlp_down_proj/max":16.5,"train/tensor_act_model_layers_0_mlp_up_proj/range":11.125,"train/tensor_act_model_layers_80_mlp_up_proj/mean":-0.00113677978515625,"train/tensor_act_model_layers_89_mlp/mean":4,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_32_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp/min":4,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_35_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_13_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_mlp_up_proj/mean":-0.03857421875,"train/tensor_act_model_layers_10_residual_post_attn/mean":40,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_62_mlp_up_proj_weight/max_abs":0.1318359375,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_77_mlp/std":0,"train/tensor_act_model_layers_27_mlp_gate_proj/std":1.8515625,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/range":0.25,"train/tensor_act_model_layers_72_residual_post_attn/norm":589422.1875,"train/tensor_param_model_layers_48_mlp_down_proj_weight/mean":0.00146484375,"train/tensor_act_model_layers_45_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_91_residual_post_attn/norm":745069.8125,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_49_residual_post_attn/std":0.81640625,"train/tensor_param_model_layers_0_mlp_down_proj_weight/std":0.057373046875,"train/tensor_act_model_layers_91_mlp/min":4,"train/tensor_act_model_layers_12_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_19_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_17_residual_post_mlp/max":73.5,"train/tensor_param_model_layers_3_mlp_down_proj_weight/min":-0.1025390625,"train/tensor_act_model_layers_7_mlp_up_proj/norm":5353.73193359375,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_18_residual_post_mlp/range":4,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_34_residual_post_mlp/min":138,"train/tensor_act_model_layers_12_mlp_up_proj/norm":5465.32470703125,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_50_mlp_gate_proj/min":-3.765625,"train/tensor_act_model_layers_34_residual_post_attn/range":4,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_55_mlp/norm":8192,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_29_mlp_up_proj/min":-3.46875,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/max_abs":250,"train/tensor_act_model_layers_68_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_68_mlp_down_proj/max":16.25,"train/tensor_act_model_layers_34_mlp_gate_proj/norm":5311.525390625,"train/tensor_act_model_layers_43_residual_post_attn/max":174,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/max":0.11962890625,"train/tensor_param_model_layers_80_mlp_down_proj_weight/std":0.03955078125,"train/tensor_act_model_layers_14_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_6_residual_post_mlp/min":27,"train/tensor_act_model_layers_37_residual_pre_attn/std":0.80859375,"train/tensor_param_model_layers_49_mlp_down_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_44_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_5_mlp_up_proj/max":3.4375,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_25_mlp_down_proj/max_abs":16.25,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/max_abs":0.14453125,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_68_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_30_residual_post_attn/max_abs":122.5,"train/tensor_act_model_layers_43_residual_post_attn/min":169,"train/tensor_act_model_layers_55_mlp_down_proj/norm":30419.8515625,"train/tensor_param_model_layers_18_mlp_down_proj_weight/max":0.126953125,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp_up_proj/mean":0.0311279296875,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_78_mlp_up_proj/mean":0.126953125,"train/tensor_act_model_layers_31_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_57_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_3_mlp_gate_proj/max_abs":3.53125,"train/tensor_act_model_layers_10_residual_post_attn/std":0.39453125,"train/tensor_act_model_layers_98_mlp_up_proj/range":9.53125,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_32_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_74_mlp_gate_proj/min":-3.90625,"train/tensor_param_model_layers_76_mlp_down_proj_weight/max_abs":0.1328125,"train/tensor_act_model_layers_38_mlp/max_abs":4,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_49_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_10_residual_post_mlp/max":45.5,"train/tensor_act_model_layers_9_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/norm":7.84375,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_94_mlp/min":4,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_53_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_96_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_19_mlp_up_proj_weight/mean":-0.001800537109375,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_25_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_32_mlp_up_proj_weight/min":-0.12255859375,"train/tensor_param_model_layers_44_mlp_down_proj_weight/max":0.130859375,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_45_mlp_down_proj/mean":14.875,"train/learning_rate":9e-05,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_69_mlp_up_proj_weight/norm":6.78125,"train/tensor_act_model_layers_62_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_66_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_mlp/min":4,"train/tensor_act_model_layers_33_mlp/norm":8192,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_44_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_20_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_20_mlp_up_proj_weight/min":-0.12109375,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_46_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp/norm":8192,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_mlp_up_proj/mean":0.058837890625,"train/tensor_act_model_layers_25_residual_pre_attn/max":102,"train/tensor_act_model_layers_56_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_43_mlp_gate_proj/max_abs":3.859375,"train/tensor_param_model_layers_99_mlp_up_proj_weight/std":0.048583984375,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/std":0.0380859375,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_45_mlp/norm":8192,"train/tensor_act_model_layers_91_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_0_residual_post_attn/min":-0.341796875,"train/tensor_act_model_layers_74_residual_post_attn/mean":296,"train/tensor_act_model_layers_38_residual_post_mlp/max":158,"train/tensor_act_model_layers_76_mlp_down_proj/std":0.474609375,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_92_mlp_down_proj/min":14.125,"train/tensor_act_model_layers_85_residual_post_attn/max":342,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_8_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_45_mlp_up_proj/norm":5457.92578125,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/max":0.125,"train/tensor_param_model_layers_70_mlp_up_proj_weight/min":-0.1337890625,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_33_mlp_down_proj/std":0.365234375,"train/tensor_act_model_layers_85_residual_post_attn/min":336,"train/tensor_param_model_layers_62_mlp_up_proj_weight/std":0.035888671875,"train/tensor_param_model_layers_43_mlp_down_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_73_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_49_mlp_down_proj_weight/range":0.25341796875,"train/tensor_param_model_layers_55_mlp_up_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_8_residual_post_mlp/mean":36,"train/tensor_act_model_layers_35_mlp_up_proj/max_abs":4.03125,"train/tensor_act_model_layers_55_residual_pre_attn/min":217,"train/tensor_act_model_layers_57_mlp_up_proj/max_abs":3.84375,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/norm":6.375,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/mean":-0.000560760498046875,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_mlp_up_proj/max_abs":3.984375,"train/tensor_act_model_layers_86_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_92_mlp_down_proj/max":17,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_29_residual_post_mlp/max_abs":122.5,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/range":0.2529296875,"train/tensor_param_model_layers_82_mlp_up_proj_weight/mean":4.076957702636719e-05,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_88_mlp_up_proj/min":-4.1875,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/min":-0.1318359375,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_68_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/range":0.294921875,"train/tensor_act_model_layers_81_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_22_mlp_up_proj_weight/norm":6.28125,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_62_mlp_down_proj/max":16.5,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_49_residual_post_mlp/norm":409077.375,"train/tensor_param_model_layers_29_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_91_mlp_down_proj_weight/max_abs":0.1494140625,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_65_mlp_gate_proj/range":7.5625,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_47_mlp_down_proj_weight/min":-0.12451171875,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_16_mlp/norm":8192,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_62_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_58_residual_post_mlp/max":238,"train/tensor_act_model_layers_78_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_72_residual_pre_attn/max":290,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_17_mlp_down_proj_weight/max":0.1220703125,"train/tensor_act_model_layers_14_residual_post_mlp/max":61.5,"train/tensor_act_model_layers_61_mlp_up_proj/min":-3.875,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/mean":8.106231689453125e-05,"train/tensor_act_model_layers_92_mlp/range":0,"train/tensor_act_model_layers_10_residual_post_mlp/mean":44,"train/tensor_param_model_layers_63_mlp_up_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_23_mlp_up_proj_weight/mean":0.0020751953125,"train/tensor_act_model_layers_79_mlp_down_proj/norm":30754.21875,"train/tensor_act_model_layers_69_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_24_residual_pre_attn/std":0.58203125,"train/tensor_param_model_layers_29_mlp_up_proj_weight/min":-0.125,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_21_mlp_down_proj/max":16.625,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_mlp_up_proj/norm":5316.5302734375,"train/tensor_act_model_layers_68_mlp_down_proj/max_abs":16.25,"train/tensor_act_model_layers_99_mlp_down_proj/min":13.9375,"train/tensor_act_model_layers_29_residual_post_attn/norm":236994.703125,"train/tensor_act_model_layers_98_mlp/max_abs":4,"train/tensor_act_model_layers_2_residual_post_mlp/norm":24574.501953125,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_21_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_11_mlp/mean":4,"train/tensor_act_model_layers_99_mlp_down_proj/max":17.75,"train/tensor_act_model_layers_11_residual_post_mlp/min":46.5,"train/tensor_act_model_layers_3_mlp_down_proj/max_abs":18.375,"train/tensor_param_model_layers_92_mlp_down_proj_weight/std":0.045166015625,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_88_mlp_up_proj_weight/norm":8,"train/tensor_act_model_layers_93_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_69_mlp_down_proj/max":16,"train/tensor_act_model_layers_56_mlp_up_proj/std":1.828125,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_94_residual_post_mlp/max_abs":382,"train/tensor_param_model_layers_21_mlp_down_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_91_mlp/range":0,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_3_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_66_mlp_up_proj_weight/norm":6.59375,"train/tensor_act_model_layers_46_mlp_down_proj/range":1.8125,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_2_mlp_up_proj/min":-4.03125,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_34_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_47_mlp_gate_proj/max_abs":4.1875,"train/tensor_act_model_layers_71_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_0_mlp_down_proj_weight/mean":-0.00179290771484375,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/mean":-0.0020751953125,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_55_mlp_down_proj_weight/min":-0.126953125,"train/tensor_param_model_layers_39_mlp_down_proj_weight/min":-0.12353515625,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_46_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_1_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_1_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_65_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_54_residual_pre_attn/std":0.81640625,"train/tensor_param_model_layers_20_mlp_down_proj_weight/range":0.2431640625,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/mean":0.0001621246337890625,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/std":0.03564453125,"train/tensor_act_model_layers_47_mlp_gate_proj/max":4.1875,"train/tensor_act_model_layers_49_residual_post_mlp/std":0.81640625,"train/tensor_act_model_layers_21_mlp_gate_proj/max_abs":3.734375,"train/tensor_act_model_layers_10_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_80_mlp_up_proj/max_abs":3.859375,"train/tensor_act_model_layers_65_mlp_up_proj/std":1.859375,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_80_mlp_down_proj_weight/max_abs":0.142578125,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_mlp_up_proj/min":-4.15625,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_91_mlp_up_proj/std":1.8203125,"train/tensor_param_model_layers_7_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_12_residual_post_mlp/max":53.5,"train/tensor_act_model_layers_96_residual_post_mlp/std":0.7578125,"train/tensor_param_model_layers_40_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_35_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_12_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_5_mlp_gate_proj/norm":5182.15283203125,"train/tensor_act_model_layers_99_mlp_gate_proj/range":9.40625,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/range":0.259765625,"train/tensor_act_model_layers_92_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_24_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_20_mlp_down_proj_weight/max_abs":0.123046875,"train/tensor_act_model_layers_51_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_8_mlp_up_proj_weight/max":0.1171875,"train/tensor_act_model_layers_58_mlp/mean":4,"train/tensor_act_model_layers_82_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_15_residual_pre_attn/min":58.5,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_19_mlp_gate_proj/norm":5359.4833984375,"train/tensor_param_model_layers_28_mlp_down_proj_weight/mean":0.00052642822265625,"train/tensor_param_model_layers_12_mlp_up_proj_weight/range":0.2431640625,"train/tensor_act_model_layers_88_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_80_mlp_gate_proj/max_abs":4.21875,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_12_mlp/range":0,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_65_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp/min":4,"train/tensor_param_model_layers_31_mlp_up_proj_weight/range":0.24560546875,"train/tensor_param_model_layers_46_mlp_up_proj_weight/std":0.034912109375,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_33_mlp_up_proj/max_abs":3.84375,"train/tensor_act_model_layers_68_mlp_gate_proj/mean":-0.044921875,"train/tensor_act_model_layers_10_residual_pre_attn/norm":81851.6796875,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_5_mlp_down_proj_weight/norm":5.90625,"train/tensor_act_model_layers_50_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_16_mlp_up_proj_weight/mean":-0.0008392333984375,"train/tensor_act_model_layers_95_mlp_down_proj/max_abs":17.625,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_67_residual_pre_attn/max_abs":270,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/max":0.125,"train/tensor_param_model_layers_42_mlp_up_proj_weight/norm":6.3125,"train/tensor_param_model_layers_19_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_89_mlp_up_proj/norm":5283.8037109375,"train/tensor_act_model_layers_82_mlp/norm":8192,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_49_mlp_gate_proj/max_abs":3.984375,"train/tensor_act_model_layers_26_mlp_gate_proj/max_abs":3.765625,"train/tensor_act_model_layers_94_mlp_down_proj/max_abs":18.625,"train/tensor_param_model_layers_91_mlp_down_proj_weight/min":-0.1494140625,"train/tensor_act_model_layers_70_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/std":77,"train/tensor_act_model_layers_65_mlp_down_proj/std":0.376953125,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_66_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_83_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/range":0.279296875,"train/tensor_param_model_layers_20_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_73_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_88_mlp_down_proj_weight/mean":-0.000537872314453125,"train/tensor_param_model_layers_11_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_mlp/norm":8192,"train/tensor_act_model_layers_44_residual_post_attn/max_abs":178,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_64_mlp_down_proj_weight/max":0.12890625,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_96_mlp_down_proj/mean":15,"train/tensor_act_model_layers_8_residual_post_attn/std":0.36328125,"train/tensor_act_model_layers_3_mlp_down_proj/range":4.1875,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/mean":-0.0012054443359375,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_29_mlp_down_proj_weight/min":-0.11962890625,"train/tensor_act_model_layers_61_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_44_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_29_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_residual_post_mlp/max":29,"train/tensor_act_model_layers_41_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_48_mlp/mean":4,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_93_mlp_up_proj_weight/min":-0.138671875,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_66_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_72_mlp_up_proj/max":3.578125,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_mlp_up_proj/norm":5364.14599609375,"train/tensor_param_model_layers_29_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_30_mlp_up_proj_weight/range":0.251953125,"train/tensor_act_model_layers_73_residual_post_mlp/min":292,"train/tensor_param_model_layers_64_mlp_up_proj_weight/max":0.1279296875,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_1_mlp_up_proj/max_abs":4.1875,"train/tensor_param_model_layers_18_mlp_up_proj_weight/norm":6.3125,"train/tensor_act_model_layers_71_mlp_up_proj/mean":-0.1513671875,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_73_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_20_mlp_up_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_69_residual_post_attn/max":278,"train/tensor_param_model_layers_58_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_73_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_residual_post_mlp/norm":40949.03515625,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_32_residual_post_attn/norm":261376.921875,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_28_mlp_up_proj_weight/mean":-3.769993782043457e-06,"train/tensor_act_model_layers_54_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/range":0.2646484375,"train/tensor_act_model_layers_69_residual_post_attn/min":272,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_5_residual_post_attn/range":1.75,"train/tensor_param_model_layers_48_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_79_residual_post_mlp/max":322,"train/tensor_act_model_layers_31_mlp_gate_proj/range":7.4375,"train/tensor_act_model_layers_62_mlp_gate_proj/norm":5410.8525390625,"train/tensor_param_model_layers_31_mlp_up_proj_weight/max":0.125,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_72_mlp_up_proj_weight/range":0.2568359375,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp_gate_proj/std":1.8203125,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_27_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_1_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_78_mlp_up_proj/std":1.8125,"train/tensor_act_model_layers_9_residual_post_attn/norm":73660.0703125,"train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit":0,"train/tensor_act_model_layers_49_residual_pre_attn/norm":400885.46875,"train/tensor_act_model_layers_37_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_95_residual_pre_attn/mean":380,"train/tensor_act_model_layers_3_residual_post_attn/mean":12,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_10_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/std":0.036376953125,"train/tensor_act_model_layers_62_residual_post_attn/mean":248,"train/tensor_act_model_layers_69_residual_post_attn/max_abs":278,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_58_mlp_down_proj_weight/max_abs":0.1318359375,"train/tensor_act_model_layers_32_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_residual_post_mlp/min":177,"train/tensor_act_model_layers_86_mlp_up_proj/min":-3.78125,"train/tensor_act_model_layers_41_residual_pre_attn/max_abs":166,"train/tensor_act_model_layers_9_mlp/range":0,"train/tensor_act_model_layers_31_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_52_residual_post_mlp/min":209,"train/tensor_act_model_layers_71_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_75_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_21_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_73_mlp_up_proj_weight/std":0.037841796875,"train/tensor_param_model_layers_83_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_37_mlp_up_proj/norm":5487,"train/tensor_act_model_layers_49_residual_post_mlp/mean":200,"train/tensor_act_model_layers_77_residual_pre_attn/max":310,"train/tensor_act_model_layers_13_mlp_down_proj/max_abs":18,"train/tensor_act_model_layers_97_residual_post_mlp/min":388,"train/tensor_act_model_layers_2_mlp_gate_proj/min":-3.515625,"train/tensor_param_model_layers_10_mlp_up_proj_weight/mean":-0.0010223388671875,"train/tensor_act_model_layers_69_mlp_gate_proj/norm":5366.57958984375,"train/tensor_param_model_layers_61_mlp_down_proj_weight/norm":6.40625,"train/tensor_act_model_layers_30_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_77_mlp_gate_proj/std":1.859375,"train/tensor_param_model_layers_29_mlp_down_proj_weight/norm":6.28125,"train/tensor_param_model_layers_84_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_10_mlp_up_proj/max":3.171875,"train/tensor_act_model_layers_17_residual_pre_attn/max_abs":69.5,"train/tensor_param_model_layers_88_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_71_mlp_up_proj_weight/std":0.0380859375,"train/tensor_param_model_layers_45_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_8_mlp_gate_proj/max":3.203125,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_residual_pre_attn/min":11.375,"train/tensor_param_model_layers_20_mlp_down_proj_weight/min":-0.1201171875,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_13_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_81_residual_post_attn/mean":324,"train/tensor_act_model_layers_38_residual_pre_attn/mean":152,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_63_mlp/max":4,"train/tensor_act_model_layers_33_mlp_gate_proj/mean":0.0439453125,"train/tensor_act_model_layers_41_mlp/min":4,"train/tensor_param_model_layers_79_mlp_down_proj_weight/range":0.275390625,"train/tensor_param_model_layers_27_mlp_up_proj_weight/range":0.2626953125,"train/tensor_act_model_layers_99_mlp_down_proj/max_abs":17.75,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_53_residual_post_attn/mean":212,"train/tensor_param_model_layers_79_mlp_up_proj_weight/range":0.28125,"train/tensor_param_model_layers_51_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_88_residual_pre_attn/min":348,"train/tensor_act_model_layers_25_residual_post_mlp/max_abs":106,"train/tensor_act_model_layers_93_mlp_up_proj/max":4.375,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_16_residual_pre_attn/range":3,"train/tensor_act_model_layers_99_residual_pre_attn/max":398,"train/tensor_act_model_layers_49_mlp_up_proj/std":1.8984375,"train/tensor_act_model_layers_70_mlp_gate_proj/max_abs":3.875,"train/tensor_act_model_layers_28_mlp_up_proj/std":1.859375,"train/tensor_param_model_layers_34_mlp_up_proj_weight/std":0.03466796875,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_83_mlp_up_proj_weight/std":0.042236328125,"train/tensor_act_model_layers_45_mlp_gate_proj/min":-3.796875,"train/tensor_act_model_layers_30_mlp_gate_proj/std":1.828125,"train/tensor_act_model_layers_39_mlp_gate_proj/min":-3.3125,"train/tensor_act_model_layers_95_residual_pre_attn/min":376,"train/tensor_act_model_layers_59_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_12_mlp_down_proj/range":1.5,"train/tensor_act_model_layers_63_mlp_up_proj/mean":-0.0167236328125,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_residual_post_mlp/max_abs":246,"train/tensor_param_model_layers_45_mlp_down_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_35_mlp_down_proj_weight/mean":0.00040435791015625,"train/tensor_act_model_layers_67_mlp/std":0,"train/tensor_act_model_layers_54_residual_post_mlp/min":217,"train/tensor_act_model_layers_73_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_13_residual_post_attn/norm":106254.078125,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/norm":6.75,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_99_mlp_down_proj_weight/max_abs":0.1650390625,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_4_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_97_residual_post_mlp/range":6,"train/tensor_param_model_layers_24_mlp_up_proj_weight/mean":-0.001556396484375,"train/tensor_act_model_layers_5_mlp/max":4,"train/tensor_act_model_layers_31_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/mean":0.0003108978271484375,"train/tensor_act_model_layers_38_mlp_gate_proj/max":3.671875,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/min":-0.1416015625,"train/tensor_act_model_layers_26_residual_pre_attn/range":4,"train/tensor_act_model_layers_4_mlp_up_proj/range":7.078125,"train/tensor_act_model_layers_4_mlp/max":4,"train/tensor_act_model_layers_97_mlp/max":4,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_2_mlp_up_proj/max":3.875,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_residual_post_mlp/range":5,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_attn/range":5,"train/tensor_act_model_layers_90_mlp_down_proj/range":2.9375,"train/tensor_act_model_layers_1_mlp/std":0,"train/tensor_act_model_layers_30_residual_pre_attn/max":122.5,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_19_mlp_up_proj_weight/range":0.2392578125,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_down_proj/std":0.353515625,"train/tensor_param_model_layers_37_mlp_up_proj_weight/range":0.25,"train/tensor_param_model_layers_77_mlp_up_proj_weight/norm":7.03125,"train/tensor_act_model_layers_28_mlp_gate_proj/norm":5385.67333984375,"train/tensor_act_model_layers_98_residual_post_mlp/max_abs":398,"train/tensor_act_model_layers_92_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_residual_pre_attn/max":94,"train/tensor_param_model_layers_63_mlp_up_proj_weight/min":-0.125,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/mean":0.00116729736328125,"train/tensor_act_model_layers_16_mlp_up_proj/std":1.84375,"train/tensor_param_model_layers_95_mlp_down_proj_weight/max":0.1435546875,"train/tensor_act_model_layers_34_residual_post_mlp/norm":286181.65625,"train/tensor_act_model_layers_41_mlp_down_proj/range":1.875,"train/tensor_param_model_layers_38_mlp_up_proj_weight/max":0.12158203125,"train/tensor_act_model_layers_56_residual_post_mlp/std":0.8203125,"train/tensor_act_model_layers_25_mlp_down_proj/std":0.37890625,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_1_mlp_up_proj_weight/mean":-0.00164031982421875,"train/tensor_act_model_layers_37_residual_post_mlp/min":149,"train/tensor_act_model_layers_61_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_81_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_81_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_59_mlp_up_proj_weight/mean":-0.00093841552734375,"train/tensor_act_model_layers_81_mlp/norm":8192,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_27_mlp/std":0,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_8_mlp_down_proj/norm":30327.58203125,"train/tensor_act_model_layers_33_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_45_mlp/max":4,"train/tensor_act_model_layers_4_mlp/min":4,"train/tensor_param_model_layers_52_mlp_down_proj_weight/range":0.25146484375,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_40_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_32_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_28_residual_pre_attn/mean":112,"train/tensor_param_model_layers_32_mlp_down_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_34_residual_pre_attn/std":0.71484375,"train/tensor_act_model_layers_92_mlp/max":4,"train/tensor_param_model_layers_5_mlp_down_proj_weight/std":0.032470703125,"train/tensor_act_model_layers_34_residual_post_mlp/range":4,"train/tensor_act_model_layers_91_mlp/max_abs":4,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_26_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_31_residual_post_attn/norm":253338.609375,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_72_mlp_down_proj_weight/std":0.03759765625,"train/tensor_act_model_layers_61_residual_post_mlp/range":5,"train/tensor_act_model_layers_58_mlp/norm":8192,"train/tensor_act_model_layers_22_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_37_mlp_up_proj/max_abs":4,"train/grad_norm":13888,"train/tensor_act_model_layers_77_mlp/range":0,"train/tensor_act_model_layers_45_mlp_up_proj/mean":-0.11865234375,"train/tensor_act_model_layers_28_mlp/std":0,"train/tensor_act_model_layers_18_mlp/min":4,"train/tensor_act_model_layers_18_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_51_mlp/max":4,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_33_residual_pre_attn/min":130,"train/tensor_act_model_layers_3_residual_post_mlp/max":16.5,"train/tensor_act_model_layers_1_mlp_gate_proj/range":7.71875,"train/tensor_act_model_layers_13_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_75_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_3_mlp/max":4,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/max":0.1220703125,"train/tensor_act_model_layers_83_residual_pre_attn/mean":332,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_35_mlp_down_proj/std":0.4140625,"train/tensor_act_model_layers_1_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_86_mlp_up_proj/norm":5452.412109375,"train/tensor_act_model_layers_20_residual_post_mlp/max_abs":86,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_78_mlp_up_proj/norm":5261.263671875,"train/tensor_act_model_layers_89_residual_post_attn/norm":728685.8125,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/norm":6.75,"train/tensor_act_model_layers_79_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_81_residual_post_mlp/min":324,"train/tensor_act_model_layers_53_mlp_up_proj/range":7.1875,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_82_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_1_mlp_gate_proj/min":-3.6875,"train/tensor_act_model_layers_94_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_26_residual_post_mlp/norm":220754.3125,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_82_mlp_up_proj_weight/range":0.2744140625,"train/tensor_act_model_layers_86_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_83_mlp_down_proj/mean":15,"train/tensor_act_model_layers_36_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_83_residual_pre_attn/min":328,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/max":0.115234375,"train/tensor_param_model_layers_97_mlp_up_proj_weight/max_abs":0.150390625,"train/tensor_act_model_layers_18_mlp/max":4,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_67_mlp_down_proj_weight/std":0.0361328125,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_30_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/std":0.037353515625,"train/tensor_act_model_layers_49_mlp/min":4,"train/tensor_act_model_layers_37_residual_post_attn/norm":302566.5,"train/tensor_act_model_layers_41_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_residual_pre_attn/mean":192,"train/tensor_act_model_layers_20_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_7_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/max":0.130859375,"train/tensor_act_model_layers_18_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_60_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_mlp_up_proj/norm":5400.6376953125,"train/tensor_act_model_layers_79_residual_post_mlp/range":6,"train/tensor_param_model_layers_16_mlp_up_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_96_mlp/mean":4,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_96_mlp_down_proj_weight/norm":8.625,"train/tensor_param_model_layers_6_mlp_down_proj_weight/min":-0.111328125,"train/tensor_param_model_layers_19_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_46_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_82_residual_pre_attn/min":324,"train/tensor_act_model_layers_35_residual_pre_attn/mean":140,"train/tensor_act_model_layers_48_residual_pre_attn/range":5,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_99_mlp_up_proj/max":4.6875,"train/tensor_act_model_layers_74_residual_post_mlp/max_abs":302,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_56_mlp_up_proj_weight/min":-0.1318359375,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_15_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_90_residual_post_attn/norm":736877.8125,"train/tensor_act_model_layers_6_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_49_mlp_up_proj/range":7.578125,"train/tensor_act_model_layers_12_mlp_up_proj/mean":0.099609375,"train/tensor_act_model_layers_39_mlp_down_proj/range":2.1875,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/min":-0.1162109375,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_73_residual_pre_attn/range":6,"train/tensor_act_model_layers_8_mlp_up_proj/norm":5305.44873046875,"train/tensor_act_model_layers_55_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_mlp_down_proj/range":3.1875,"train/tensor_act_model_layers_12_residual_pre_attn/range":3,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_45_mlp_up_proj_weight/range":0.248046875,"train/tensor_act_model_layers_30_residual_post_attn/std":0.68359375,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/min":-0.1142578125,"train/tensor_act_model_layers_14_mlp/mean":4,"train/tensor_param_model_layers_24_mlp_down_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_15_mlp_gate_proj/min":-3.9375,"train/tensor_act_model_layers_24_mlp_up_proj/max":3.609375,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/mean":0.001068115234375,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/range":0.2587890625,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_86_mlp_down_proj_weight/norm":7.75,"train/tensor_act_model_layers_95_residual_post_attn/norm":777837.6875,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/mean":-0.00038909912109375,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_98_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_1_mlp_down_proj/max":23.375,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_6_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_70_mlp_down_proj_weight/mean":0.000911712646484375,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_88_mlp/std":0,"train/tensor_param_model_layers_9_mlp_down_proj_weight/mean":-0.0005340576171875,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_53_mlp_up_proj/mean":-0.041259765625,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/min":-0.1484375,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_9_mlp_down_proj/max_abs":17,"train/tensor_act_model_layers_34_mlp_up_proj/std":1.8125,"train/tensor_act_model_layers_12_mlp_down_proj/max":15.6875,"train/tensor_act_model_layers_91_mlp_down_proj/max":17.25,"train/tensor_param_model_layers_71_mlp_up_proj_weight/range":0.2646484375,"train/tensor_act_model_layers_96_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_mlp_down_proj/std":0.42578125,"train/tensor_act_model_layers_67_residual_pre_attn/mean":268,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_34_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_post_mlp/mean":108,"train/tensor_act_model_layers_80_mlp_up_proj/norm":5262.0732421875,"train/tensor_param_model_layers_0_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_14_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_11_mlp_gate_proj/max_abs":3.484375,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_72_mlp_down_proj/std":0.53125,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/norm":7.75,"train/tensor_act_model_layers_63_residual_post_mlp/max_abs":258,"train/tensor_act_model_layers_14_mlp_down_proj/max":16.5,"train/tensor_act_model_layers_34_residual_pre_attn/norm":277973.8125,"train/tensor_act_model_layers_79_mlp_gate_proj/mean":0.068359375,"train/tensor_act_model_layers_43_mlp/max":4,"train/tensor_act_model_layers_65_residual_pre_attn/range":6,"train/tensor_act_model_layers_2_residual_post_attn/max_abs":8.4375,"train/tensor_act_model_layers_57_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_32_residual_pre_attn/std":0.6484375,"train/tensor_act_model_layers_35_mlp_gate_proj/norm":5367.43603515625,"train/tensor_act_model_layers_12_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_63_mlp_gate_proj/min":-3.5,"train/tensor_act_model_layers_9_mlp_up_proj/norm":5370.212890625,"train/tensor_param_model_layers_36_mlp_down_proj_weight/range":0.2470703125,"train/tensor_act_model_layers_91_mlp_down_proj/min":14.0625,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_63_residual_pre_attn/min":248,"train/tensor_act_model_layers_80_mlp/std":0,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/std":0.037841796875,"train/tensor_act_model_layers_96_residual_pre_attn/max_abs":386,"train/tensor_act_model_layers_2_residual_post_attn/mean":8,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/max_abs":0.13671875,"train/tensor_act_model_layers_98_residual_post_mlp/max":398,"train/tensor_act_model_layers_9_residual_pre_attn/max":37.5,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_95_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp/norm":8192,"train/tensor_param_model_layers_17_mlp_up_proj_weight/max":0.12451171875,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/max":0.142578125,"train/tensor_act_model_layers_41_residual_post_attn/min":161,"train/tensor_act_model_layers_32_mlp_gate_proj/max_abs":3.90625,"train/tensor_act_model_layers_53_mlp_gate_proj/max":3.953125,"train/tensor_act_model_layers_78_mlp_down_proj/max_abs":16.875,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_55_residual_post_attn/min":217,"train/tensor_act_model_layers_81_mlp_up_proj/max_abs":3.859375,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_80_mlp_up_proj_weight/max":0.1376953125,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_92_mlp_up_proj_weight/min":-0.13671875,"train/tensor_act_model_layers_87_mlp_up_proj/min":-3.65625,"train/tensor_act_model_layers_3_residual_post_attn/std":0.1572265625,"train/tensor_param_model_layers_36_mlp_up_proj_weight/min":-0.12060546875,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_44_mlp/max":4,"train/tensor_act_model_layers_39_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_residual_post_attn/max":366,"train/tensor_param_model_layers_19_mlp_down_proj_weight/range":0.25,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_68_mlp_down_proj/norm":30570.796875,"train/tensor_act_model_layers_34_residual_post_mlp/mean":140,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_3_mlp/mean":4,"train/tensor_act_model_layers_21_mlp_down_proj/max_abs":16.625,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_residual_post_attn/min":360,"train/tensor_act_model_layers_57_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_5_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_81_mlp_up_proj/norm":5196.44384765625,"train/tensor_act_model_layers_90_mlp_gate_proj/max_abs":3.984375,"train/tensor_act_model_layers_41_residual_pre_attn/range":5,"train/tensor_act_model_layers_34_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_85_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_13_mlp_gate_proj/min":-3.71875,"train/tensor_act_model_layers_49_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_pre_attn/norm":212562.453125,"train/tensor_param_model_layers_74_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_16_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_9_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_85_mlp_up_proj_weight/std":0.04248046875,"train/tensor_act_model_layers_93_residual_post_mlp/max_abs":378,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_55_residual_post_attn/max":222,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_22_residual_post_attn/range":4,"train/tensor_act_model_layers_70_mlp/norm":8192,"train/tensor_act_model_layers_70_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_27_mlp_down_proj_weight/mean":-0.00103759765625,"train/tensor_act_model_layers_30_mlp_up_proj/max_abs":3.734375,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/mean":0.1845703125,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_59_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_mlp_gate_proj/range":7.296875,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_15_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_41_residual_post_attn/max":166,"train/tensor_act_model_layers_96_residual_pre_attn/mean":384,"train/tensor_act_model_layers_49_mlp_gate_proj/max":3.984375,"train/tensor_act_model_layers_59_residual_pre_attn/max_abs":238,"train/tensor_act_model_layers_12_residual_post_attn/range":3,"train/tensor_param_model_layers_0_mlp_up_proj_weight/min":-0.125,"train/tensor_param_model_layers_86_mlp_down_proj_weight/std":0.04296875,"train/tensor_act_model_layers_28_residual_post_attn/max_abs":114.5,"train/tensor_act_model_layers_46_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_88_residual_post_attn/range":6,"train/tensor_act_model_layers_80_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_98_mlp/mean":4,"train/tensor_act_model_layers_3_residual_post_mlp/min":15.375,"train/tensor_act_model_layers_92_residual_pre_attn/range":6,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/mean":0.00069427490234375,"train/tensor_act_model_layers_85_mlp_down_proj/max_abs":17.625,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_residual_pre_attn/min":138,"train/tensor_act_model_layers_44_residual_pre_attn/std":0.80859375,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_66_mlp_down_proj/range":2.6875,"train/tensor_act_model_layers_6_residual_post_attn/min":23,"train/tensor_act_model_layers_30_residual_post_mlp/norm":253346.515625,"train/tensor_act_model_layers_56_mlp_up_proj/norm":5287.08349609375,"train/tensor_param_model_layers_84_mlp_down_proj_weight/frac_near_user_limit":0,"eval/samples_per_second":589.567,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/std":0.035400390625,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_5_mlp_up_proj_weight/norm":5.96875,"train/tensor_act_model_layers_90_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_70_residual_pre_attn/mean":280,"train/tensor_act_model_layers_2_mlp_down_proj/max":17.625,"train/tensor_act_model_layers_56_mlp/std":0,"train/tensor_act_model_layers_69_residual_post_attn/mean":276,"train/tensor_param_model_layers_39_mlp_up_proj_weight/range":0.24169921875,"train/tensor_param_model_layers_29_mlp_down_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/mean":-0.0012054443359375,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_41_mlp_up_proj_weight/mean":0.00020885467529296875,"train/tensor_act_model_layers_20_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_67_residual_post_attn/norm":548462.375,"train/tensor_act_model_layers_36_mlp_down_proj/std":0.3828125,"train/tensor_act_model_layers_30_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_38_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_55_mlp_gate_proj/mean":0.01104736328125,"train/tensor_act_model_layers_44_residual_pre_attn/max_abs":178,"train/tensor_act_model_layers_19_mlp_gate_proj/max_abs":3.640625,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_3_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_60_mlp_up_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_59_mlp_up_proj/range":7.390625,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_post_attn/std":0.62109375,"train/tensor_act_model_layers_56_mlp_up_proj/max_abs":3.984375,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_43_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_99_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_88_mlp_up_proj/norm":5295.177734375,"train/tensor_act_model_layers_7_residual_pre_attn/range":2,"train/tensor_grad_model_layers_29_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_94_mlp_down_proj_weight/range":0.2900390625,"train/tensor_act_model_layers_78_mlp_gate_proj/max":3.78125,"train/tensor_act_model_layers_90_mlp_down_proj/max":17.125,"train/tensor_param_model_layers_24_mlp_down_proj_weight/range":0.244140625,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/max":0.1220703125,"train/tensor_act_model_layers_37_mlp_up_proj/mean":-0.08740234375,"train/tensor_act_model_layers_95_mlp/max_abs":4,"train/tensor_act_model_layers_20_mlp_down_proj/max":16.625,"train/tensor_param_model_layers_99_mlp_up_proj_weight/norm":8.8125,"train/tensor_act_model_layers_73_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_36_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_78_residual_post_attn/mean":312,"train/tensor_param_model_layers_32_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_24_residual_post_attn/norm":196203.3125,"train/tensor_grad_model_layers_81_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_57_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_43_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_40_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_51_residual_post_mlp/std":0.81640625,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_81_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_mlp/range":3,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_35_residual_pre_attn/range":4,"train/tensor_act_model_layers_36_mlp_gate_proj/norm":5352.640625,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_54_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_31_mlp_down_proj_weight/min":-0.125,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/max":0.123046875,"train/tensor_act_model_layers_80_residual_post_mlp/range":6,"train/tensor_act_model_layers_46_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_29_mlp_down_proj/range":2.125,"train/tensor_act_model_layers_21_residual_post_attn/mean":84,"train/tensor_act_model_layers_74_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/min":-0.1396484375,"train/tensor_act_model_layers_94_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_50_residual_post_mlp/max_abs":206,"train/tensor_param_model_layers_7_mlp_down_proj_weight/mean":-0.000164031982421875,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/mean":0.0006256103515625,"train/tensor_act_model_layers_81_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_15_mlp_up_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_10_mlp_gate_proj/min":-3.21875,"train/tensor_act_model_layers_5_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_73_residual_post_attn/max_abs":294,"train/tensor_act_model_layers_18_mlp_down_proj/max":15.875,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_64_mlp_gate_proj/norm":5309.96484375,"train/tensor_act_model_layers_9_mlp/max":4,"train/tensor_act_model_layers_82_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_89_mlp/range":0,"train/tensor_act_model_layers_85_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_48_mlp/range":0,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_67_mlp_gate_proj/max_abs":4.0625,"train/tensor_act_model_layers_11_residual_post_mlp/range":3,"train/tensor_act_model_layers_11_residual_pre_attn/mean":44,"train/tensor_act_model_layers_93_mlp_gate_proj/std":1.8515625,"train/tensor_act_model_layers_23_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_72_residual_pre_attn/max_abs":290,"train/tensor_act_model_layers_7_residual_post_mlp/max_abs":33.5,"train/tensor_param_model_layers_31_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_79_mlp_up_proj/min":-4.28125,"train/tensor_act_model_layers_74_residual_pre_attn/min":292,"train/tensor_param_model_layers_28_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_1_residual_post_mlp/range":0.9375,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_34_residual_post_attn/std":0.7109375,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_99_residual_post_attn/max_abs":398,"train/tensor_act_model_layers_88_mlp_gate_proj/min":-3.96875,"train/tensor_act_model_layers_72_residual_pre_attn/mean":288,"train/tensor_act_model_layers_85_mlp_down_proj/max":17.625,"train/tensor_act_model_layers_20_residual_post_attn/mean":80,"train/tensor_param_model_layers_24_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_residual_post_attn/max_abs":146,"train/tensor_act_model_layers_64_residual_post_attn/norm":523449.15625,"train/tensor_act_model_layers_87_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_29_residual_post_mlp/max":122.5,"train/tensor_act_model_layers_50_mlp_up_proj/mean":0.04150390625,"train/tensor_act_model_layers_38_mlp_up_proj/range":7.328125,"train/tensor_param_model_layers_86_mlp_up_proj_weight/max_abs":0.1474609375,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_65_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_18_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_94_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/max":0.130859375,"train/tensor_act_model_layers_71_mlp_up_proj/norm":5368.767578125,"train/tensor_act_model_layers_43_residual_post_mlp/range":5,"train/tensor_act_model_layers_52_residual_pre_attn/std":0.81640625,"train/tensor_param_model_layers_5_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_mlp_gate_proj/std":1.8515625,"train/tensor_act_model_layers_84_residual_post_mlp/min":336,"train/tensor_act_model_layers_50_mlp_up_proj/min":-3.90625,"train/tensor_act_model_layers_4_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/min":-390,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_35_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_3_residual_pre_attn/norm":24574.501953125,"train/tensor_act_model_layers_11_mlp_gate_proj/min":-3.484375,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_38_mlp_gate_proj/range":7.234375,"train/tensor_act_model_layers_42_mlp_gate_proj/norm":5406.24072265625,"train/tensor_param_model_layers_17_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_91_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_87_mlp_up_proj/max_abs":3.84375,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_38_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_53_mlp_gate_proj/max_abs":4,"train/tensor_act_model_layers_68_mlp_gate_proj/range":7.71875,"train/tensor_param_model_layers_12_mlp_down_proj_weight/norm":6.125,"train/tensor_param_model_layers_60_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_93_residual_post_mlp/range":6,"train/tensor_act_model_layers_41_residual_post_mlp/min":165,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/min":-0.134765625,"train/tensor_act_model_layers_73_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_95_mlp_down_proj/norm":30764.93359375,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_57_residual_post_mlp/range":5,"train/tensor_act_model_layers_8_mlp_up_proj/range":6.5625,"train/tensor_act_model_layers_5_mlp_up_proj/min":-3.15625,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_67_mlp/max":4,"train/tensor_act_model_layers_29_residual_post_attn/range":4.5,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_40_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_28_residual_post_attn/norm":228818.84375,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_77_mlp_up_proj/min":-4.125,"train/tensor_act_model_layers_0_residual_post_mlp/range":0.828125,"train/tensor_act_model_layers_83_mlp_up_proj/max_abs":4.15625,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_43_mlp_up_proj_weight/range":0.24267578125,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/min":-0.12890625,"train/tensor_act_model_layers_1_mlp/max":4,"train/tensor_act_model_layers_89_mlp_up_proj/mean":0.0888671875,"train/tensor_act_model_layers_94_residual_pre_attn/mean":376,"train/tensor_act_model_layers_56_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_42_mlp_gate_proj/min":-3.890625,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/max":408,"train/tensor_param_model_layers_26_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/min":-0.15234375,"train/tensor_param_model_layers_4_mlp_up_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_33_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_74_mlp_gate_proj/norm":5329.7353515625,"train/tensor_act_model_layers_11_residual_pre_attn/std":0.39453125,"train/tensor_param_model_layers_64_mlp_up_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_95_mlp_down_proj/std":0.640625,"train/tensor_act_model_layers_56_mlp_down_proj/norm":30549.01953125,"train/tensor_act_model_layers_64_residual_post_attn/max_abs":258,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_56_mlp_down_proj/max":16.75,"train/tensor_param_model_layers_78_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp/max":4,"train/tensor_act_model_layers_4_mlp/norm":8192,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_78_mlp_up_proj/range":7.921875,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_76_residual_post_attn/max":306,"train/tensor_act_model_layers_24_residual_post_mlp/min":98,"train/tensor_act_model_layers_77_residual_post_mlp/range":6,"train/tensor_act_model_layers_12_mlp_gate_proj/norm":5292.70849609375,"train/tensor_grad_model_layers_58_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_35_mlp_up_proj_weight/mean":-0.00055694580078125,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_18_mlp/max_abs":4,"_timestamp":1.7867428986313863e+09,"train/tensor_act_model_layers_56_mlp/norm":8192,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_15_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_73_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_55_mlp_down_proj_weight/mean":0.000835418701171875,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_59_mlp_gate_proj/max":4.03125,"train/tensor_act_model_layers_10_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_76_residual_pre_attn/max_abs":306,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_63_mlp_down_proj_weight/mean":0.00110626220703125,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/std":64.5,"train/tensor_act_model_layers_3_residual_post_mlp/mean":16,"train/tensor_act_model_layers_3_mlp_up_proj/std":1.8203125,"train/tensor_act_model_layers_43_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_down_proj_weight/min":-0.12060546875,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_87_residual_post_mlp/max_abs":354,"train/tensor_act_model_layers_16_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_25_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp/max_abs":4,"train/tensor_act_model_layers_14_residual_pre_attn/mean":56,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_71_residual_post_attn/max_abs":286,"train/tensor_act_model_layers_30_residual_pre_attn/norm":245186.5625,"train/tensor_act_model_layers_87_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_26_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_4_mlp_gate_proj/max":3.546875,"train/tensor_param_model_layers_58_mlp_up_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_26_residual_post_attn/range":4,"train/tensor_param_model_layers_49_mlp_up_proj_weight/range":0.236328125,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_2_mlp_up_proj/max_abs":4.03125,"train/tensor_param_model_layers_66_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_40_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_76_residual_post_attn/min":300,"train/tensor_param_model_layers_36_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_75_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_30_residual_post_mlp/mean":123.5,"train/tensor_act_model_layers_62_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_39_mlp_down_proj_weight/norm":6.25,"train/tensor_param_model_layers_38_mlp_up_proj_weight/range":0.24755859375,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_7_mlp_down_proj_weight/range":0.236328125,"train/tensor_param_model_layers_45_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_33_mlp_up_proj/max":3.546875,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_param_model_layers_92_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_50_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_78_residual_pre_attn/max":314,"train/tensor_param_model_layers_1_mlp_down_proj_weight/max_abs":0.12109375,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_25_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_7_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_47_residual_pre_attn/max":190,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_55_mlp_up_proj/range":7.296875,"train/tensor_act_model_layers_96_mlp_up_proj/min":-4.65625,"train/tensor_act_model_layers_74_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_72_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_16_mlp_gate_proj/norm":5321.03076171875,"train/tensor_act_model_layers_6_residual_post_mlp/max_abs":29,"train/tensor_act_model_layers_95_residual_post_mlp/max_abs":386,"train/tensor_act_model_layers_85_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_26_residual_post_attn/min":102,"train/tensor_act_model_layers_81_residual_post_mlp/max":330,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_40_mlp_gate_proj/norm":5367.4033203125,"train/tensor_act_model_layers_51_mlp_gate_proj/max_abs":3.75,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/std":0.040283203125,"train/tensor_act_model_layers_91_residual_pre_attn/norm":745069.8125,"train/tensor_act_model_layers_12_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_26_residual_post_attn/max":106,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_29_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_82_mlp_up_proj_weight/max":0.13671875,"train/tensor_act_model_layers_6_mlp_up_proj/norm":5288.47314453125,"train/tensor_param_model_layers_73_mlp_down_proj_weight/range":0.2666015625,"train/tensor_act_model_layers_20_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_43_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_49_mlp_up_proj_weight/min":-0.11376953125,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/max_abs":0.140625,"train/tensor_act_model_layers_90_residual_pre_attn/min":356,"train/tensor_param_model_layers_13_mlp_down_proj_weight/range":0.24609375,"train/tensor_act_model_layers_31_mlp_down_proj/max_abs":15.875,"train/tensor_act_model_layers_55_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_5_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_9_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_11_mlp/range":0,"train/tensor_act_model_layers_92_mlp_up_proj/norm":5326.20068359375,"train/tensor_act_model_layers_38_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_49_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_0_mlp_gate_proj/norm":4688.95263671875,"train/tensor_act_model_layers_92_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_97_mlp_down_proj_weight/max":0.146484375,"train/tensor_act_model_layers_99_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_40_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_residual_pre_attn/norm":450021.09375,"train/tensor_param_model_layers_15_mlp_up_proj_weight/norm":6.15625,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_mlp/std":0,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_89_mlp_up_proj/std":1.8203125,"train/tensor_act_model_layers_67_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_67_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_52_mlp_up_proj/max":3.734375,"train/tensor_act_model_layers_41_residual_pre_attn/norm":335334.0625,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_35_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_77_residual_pre_attn/max_abs":310,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_65_mlp_up_proj_weight/max":0.1259765625,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_21_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_20_mlp_up_proj/min":-3.15625,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_87_mlp/mean":4,"train/tensor_act_model_layers_31_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_18_residual_pre_attn/range":3.5,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_22_mlp/max":4,"train/tensor_act_model_layers_40_mlp_gate_proj/range":6.921875,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_15_mlp_up_proj/mean":-0.0810546875,"train/tensor_act_model_layers_17_residual_post_attn/norm":138933.890625,"train/tensor_param_model_layers_77_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_6_mlp_down_proj_weight/max_abs":0.12060546875,"train/tensor_act_model_layers_51_mlp_down_proj/range":1.6875,"train/tensor_act_model_layers_21_residual_pre_attn/max":86,"train/tensor_act_model_layers_62_residual_pre_attn/max_abs":250,"train/tensor_act_model_layers_79_mlp_down_proj/min":14.1875,"train/tensor_param_model_layers_30_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_41_mlp/max_abs":4,"train/tensor_act_model_layers_76_residual_post_mlp/norm":630382.0625,"train/tensor_act_model_layers_5_mlp_up_proj/max_abs":3.4375,"train/tensor_param_model_layers_73_mlp_up_proj_weight/mean":0.000308990478515625,"train/tensor_act_model_layers_77_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_74_mlp_up_proj/min":-4.03125,"train/tensor_act_model_layers_17_mlp_down_proj/std":0.37890625,"train/tensor_act_model_layers_72_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_0_mlp/max_abs":4,"train/tensor_act_model_layers_21_mlp_down_proj/norm":30452.419921875,"train/tensor_param_model_layers_72_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_mlp_up_proj/mean":0.16796875,"train/tensor_param_model_layers_9_mlp_up_proj_weight/min":-0.11962890625,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_91_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_90_mlp_up_proj/max":3.921875,"train/tensor_act_model_layers_83_mlp_gate_proj/norm":5152.57861328125,"train/tensor_act_model_layers_41_residual_post_mlp/max":170,"train/tensor_act_model_layers_35_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_mlp_up_proj/norm":5361.0322265625,"train/tensor_act_model_layers_1_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_41_mlp_up_proj/norm":5402.8916015625,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_19_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_20_mlp/norm":8192,"train/tensor_param_model_layers_47_mlp_down_proj_weight/range":0.2490234375,"train/tensor_param_model_layers_6_mlp_up_proj_weight/mean":0.000457763671875,"train/tensor_act_model_layers_66_mlp_gate_proj/max_abs":3.53125,"train/tensor_act_model_layers_90_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_87_mlp_down_proj/norm":30746.5390625,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/max":0.1357421875,"train/tensor_act_model_layers_57_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_50_mlp_up_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_48_mlp_gate_proj/std":1.84375,"train/tensor_act_model_layers_34_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_35_mlp_down_proj_weight/range":0.251953125,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_58_residual_pre_attn/std":0.8359375,"train/tensor_act_model_layers_60_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_32_mlp_up_proj_weight/mean":-0.00020313262939453125,"train/tensor_act_model_layers_32_mlp/range":0,"train/tensor_act_model_layers_63_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp/std":0,"train/tensor_act_model_layers_28_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_39_residual_post_mlp/max_abs":162,"train/tensor_act_model_layers_24_residual_post_mlp/max":102,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_1_mlp_gate_proj/norm":5421.07275390625,"train/tensor_act_model_layers_87_mlp_up_proj/norm":5257.25537109375,"train/tensor_act_model_layers_68_residual_post_attn/range":6,"train/tensor_act_model_layers_54_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_residual_post_attn/std":0.8203125,"train/tensor_param_model_layers_57_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_42_mlp/std":0,"train/tensor_act_model_layers_14_residual_pre_attn/std":0.46875,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_46_residual_post_attn/norm":376293.59375,"train/tensor_act_model_layers_79_mlp_down_proj/max":17.375,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_30_mlp/std":0,"train/tensor_act_model_layers_29_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_28_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_41_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_60_mlp_down_proj/range":2.375,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_99_mlp_down_proj_weight/min":-0.1416015625,"train/tensor_param_model_layers_23_mlp_down_proj_weight/mean":-0.00011301040649414062,"train/tensor_act_model_layers_42_mlp/range":0,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_47_residual_post_attn/norm":384485.53125,"train/tensor_act_model_layers_28_mlp_down_proj/mean":14.8125,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_81_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_residual_post_attn/max_abs":138,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_4_mlp_down_proj/norm":30598.3203125,"train/tensor_act_model_layers_45_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_80_mlp_gate_proj/std":1.8125,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_60_mlp/max":4,"train/tensor_act_model_layers_66_residual_pre_attn/max_abs":266,"train/tensor_act_model_layers_70_mlp_down_proj/norm":30714.142578125,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_32_mlp_up_proj_weight/range":0.24560546875,"train/tensor_act_model_layers_76_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_19_residual_pre_attn/mean":76,"train/tensor_act_model_layers_97_mlp_down_proj/norm":30587.279296875,"train/tensor_param_model_layers_7_mlp_down_proj_weight/norm":5.90625,"train/tensor_act_model_layers_43_residual_pre_attn/min":169,"train/tensor_param_model_layers_74_mlp_down_proj_weight/max_abs":0.1435546875,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_66_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_72_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_31_residual_post_attn/max_abs":126.5,"train/tensor_param_model_layers_43_mlp_up_proj_weight/max":0.126953125,"train/tensor_act_model_layers_97_residual_post_mlp/max":394,"train/tensor_act_model_layers_47_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_51_mlp/range":0,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_20_mlp_up_proj/max_abs":3.875,"train/tensor_act_model_layers_69_mlp/norm":8192,"train/tensor_param_model_layers_18_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_36_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_mlp_down_proj/max_abs":17.25,"train/tensor_act_model_layers_6_mlp/range":0,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_51_residual_post_attn/norm":417269.3125,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_60_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_48_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_post_mlp/max":222,"train/tensor_act_model_layers_65_mlp_gate_proj/mean":-0.04443359375,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_44_mlp/mean":4,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_mlp_down_proj/max_abs":15.875,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp/mean":4,"train/tensor_act_model_layers_12_residual_pre_attn/std":0.396484375,"train/tensor_act_model_layers_7_residual_pre_attn/max":29,"train/tensor_act_model_layers_13_residual_pre_attn/mean":52,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/min":-0.1357421875,"train/tensor_param_model_layers_16_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_4_mlp_down_proj_weight/min":-0.11767578125,"train/tensor_act_model_layers_64_mlp_up_proj/min":-3.765625,"train/tensor_param_model_layers_48_mlp_up_proj_weight/mean":0.0001125335693359375,"train/tensor_act_model_layers_62_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_22_residual_post_attn/norm":179867.25,"train/tensor_act_model_layers_47_mlp_gate_proj/min":-3.546875,"train/tensor_act_model_layers_86_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_45_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_21_residual_pre_attn/max_abs":86,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/max":0.12451171875,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_90_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_mlp_down_proj/max_abs":15.8125,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_56_mlp_down_proj/max_abs":16.75,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_76_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_0_mlp/norm":8191.18701171875,"train/tensor_param_model_layers_72_mlp_up_proj_weight/max_abs":0.130859375,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_30_residual_post_attn/min":118,"train/tensor_act_model_layers_79_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_78_mlp_down_proj_weight/range":0.2607421875,"train/tensor_act_model_layers_2_mlp/max":4,"train/tensor_act_model_layers_10_mlp/max_abs":4,"train/tensor_param_model_layers_14_mlp_up_proj_weight/max":0.12158203125,"train/tensor_param_model_layers_25_mlp_up_proj_weight/max":0.126953125,"train/tensor_act_model_layers_48_mlp/std":0,"train/tensor_act_model_layers_91_residual_pre_attn/min":360,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_36_residual_post_attn/range":5,"train/tensor_act_model_layers_9_mlp_up_proj/max_abs":3.765625,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_92_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_pre_attn/mean":216,"train/tensor_act_model_layers_45_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_75_mlp_up_proj/max_abs":4.40625,"train/tensor_act_model_layers_2_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_96_residual_pre_attn/max":386,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_79_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_93_mlp_down_proj/max_abs":16.625,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/max":0.1494140625,"train/tensor_param_model_layers_68_mlp_up_proj_weight/min":-0.1318359375,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_3_residual_post_mlp/norm":32756.287109375,"train/tensor_act_model_layers_70_residual_post_mlp/range":6,"train/tensor_act_model_layers_77_residual_post_attn/range":6,"train/tensor_act_model_layers_31_residual_pre_attn/mean":123.5,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_25_mlp_up_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_19_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_71_residual_pre_attn/min":280,"train/tensor_param_model_layers_58_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_74_mlp_gate_proj/range":7.421875,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_27_mlp_down_proj/std":0.3671875,"train/tensor_act_model_layers_49_residual_pre_attn/max":198,"train/tensor_act_model_layers_87_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_42_mlp_down_proj_weight/mean":0.00091552734375,"train/tensor_act_model_layers_79_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_22_mlp_down_proj_weight/mean":-0.000286102294921875,"train/tensor_act_model_layers_76_residual_pre_attn/norm":622190.125,"train/tensor_act_model_layers_3_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_residual_pre_attn/range":5,"train/tensor_param_model_layers_52_mlp_up_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_5_residual_pre_attn/mean":20,"train/tensor_act_model_layers_87_mlp/norm":8192,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_47_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_70_residual_pre_attn/min":276,"train/tensor_act_model_layers_6_residual_post_mlp/std":0.2412109375,"train/tensor_act_model_layers_83_mlp/mean":4,"train/tensor_param_model_layers_68_mlp_down_proj_weight/max":0.12890625,"train/tensor_act_model_layers_37_residual_pre_attn/mean":148,"train/tensor_act_model_layers_59_residual_post_mlp/range":5,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/std":0.047607421875,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_85_mlp_up_proj/mean":0.16796875,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/norm":6.90625,"train/tensor_act_model_layers_68_residual_post_attn/mean":272,"train/tensor_act_model_layers_1_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_13_mlp_up_proj/max_abs":3.78125,"train/tensor_act_model_layers_53_mlp_up_proj/max_abs":3.75,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_26_mlp/norm":8192,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/std":0.03515625,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_94_residual_post_attn/max_abs":378,"train/tensor_act_model_layers_59_mlp/max_abs":4,"train/tensor_act_model_layers_5_residual_post_attn/max":20.875,"train/tensor_act_model_layers_15_mlp_down_proj/max_abs":16,"train/tensor_param_model_layers_36_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/norm":6.3125,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_mlp_up_proj/max_abs":4.03125,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_58_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_59_residual_post_attn/max_abs":238,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_51_mlp_down_proj_weight/range":0.25390625,"train/tensor_act_model_layers_7_mlp_down_proj/range":1.75,"train/tensor_act_model_layers_13_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_11_mlp_up_proj/norm":5453.02099609375,"train/tensor_act_model_layers_30_residual_post_attn/max":122.5,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/range":0.2705078125,"train/tensor_act_model_layers_89_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_65_mlp/range":0,"train/tensor_act_model_layers_27_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_82_mlp_down_proj/range":2.5,"train/tensor_param_model_layers_69_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_75_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_48_mlp_up_proj_weight/min":-0.1220703125,"train/tensor_act_model_layers_26_mlp/min":4,"train/tensor_act_model_layers_43_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_31_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_44_mlp_up_proj_weight/range":0.24560546875,"train/tensor_param_model_layers_62_mlp_down_proj_weight/max":0.12255859375,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_37_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_76_mlp_gate_proj/std":1.78125,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_30_residual_pre_attn/range":4.5,"train/tensor_act_model_layers_38_residual_post_attn/mean":152,"train/tensor_act_model_layers_17_mlp_up_proj/min":-3.890625,"train/tensor_act_model_layers_37_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_37_mlp_up_proj_weight/mean":-0.000682830810546875,"train/tensor_param_model_layers_77_mlp_up_proj_weight/std":0.038818359375,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_52_mlp_up_proj/max_abs":3.734375,"train/tensor_param_model_layers_28_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_19_mlp/range":0,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_72_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_97_mlp/std":0,"train/tensor_act_model_layers_76_residual_post_mlp/mean":308,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_14_mlp/range":0,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_40_residual_post_mlp/range":5,"train/tensor_param_model_layers_83_mlp_down_proj_weight/min":-0.1396484375,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_48_residual_post_attn/max_abs":194,"train/tensor_act_model_layers_85_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_33_mlp_down_proj/max_abs":15.8125,"train/tensor_act_model_layers_26_mlp_up_proj/std":1.875,"train/tensor_act_model_layers_2_mlp_down_proj/std":0.703125,"train/tensor_param_model_layers_54_mlp_up_proj_weight/norm":6.375,"train/tensor_act_model_layers_73_mlp/max":4,"train/tensor_act_model_layers_63_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_57_mlp_gate_proj/norm":5332.88427734375,"train/tensor_act_model_layers_35_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/norm":6.34375,"train/tensor_param_model_layers_84_mlp_up_proj_weight/max":0.140625,"train/tensor_act_model_layers_75_residual_post_mlp/max_abs":306,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_84_residual_post_mlp/max":342,"train/tensor_act_model_layers_23_mlp_gate_proj/range":7.640625,"train/tensor_param_model_layers_13_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_57_mlp_up_proj/max":3.84375,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_32_residual_pre_attn/max_abs":130,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_residual_pre_attn/std":0.76953125,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/max":0.1142578125,"train/tensor_act_model_layers_52_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_mlp_up_proj/max":3.59375,"train/tensor_act_model_layers_98_mlp_up_proj/norm":5359.20849609375,"train/tensor_param_model_layers_58_mlp_up_proj_weight/max":0.126953125,"train/tensor_act_model_layers_0_mlp_up_proj/norm":4674.65869140625,"train/tensor_param_model_layers_66_mlp_up_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_11_mlp/max":4,"train/tensor_param_model_layers_74_mlp_down_proj_weight/std":0.037841796875,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_18_residual_post_attn/range":4,"train/tensor_act_model_layers_33_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_10_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_residual_post_mlp/min":118,"train/tensor_param_model_layers_50_mlp_down_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_63_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_54_mlp_down_proj_weight/mean":-0.00138092041015625,"train/tensor_act_model_layers_77_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_87_mlp_down_proj/max":16.125,"train/tensor_param_model_layers_12_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_76_residual_post_attn/mean":304,"train/tensor_act_model_layers_12_mlp/mean":4,"train/tensor_act_model_layers_26_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_25_mlp_down_proj_weight/mean":0.000499725341796875,"train/tensor_param_model_layers_74_mlp_down_proj_weight/range":0.2705078125,"train/tensor_param_model_layers_30_mlp_down_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_64_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_30_mlp_down_proj_weight/mean":0.000316619873046875,"train/tensor_act_model_layers_94_mlp_gate_proj/norm":5322.109375,"train/tensor_act_model_layers_45_mlp/mean":4,"train/tensor_act_model_layers_79_mlp_up_proj/max_abs":4.28125,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_74_mlp_up_proj/range":7.421875,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_82_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_11_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_63_residual_post_attn/range":6,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_67_residual_pre_attn/min":264,"train/tensor_act_model_layers_94_mlp/range":0,"train/tensor_act_model_layers_16_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_mlp_gate_proj/mean":0.0081787109375,"train/tensor_act_model_layers_98_mlp_gate_proj/min":-4.875,"train/tensor_act_model_layers_25_residual_post_mlp/std":0.62109375,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/mean":0.0016021728515625,"train/tensor_param_model_layers_89_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_3_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_76_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_up_proj/min":-3.6875,"train/tensor_param_model_layers_24_mlp_up_proj_weight/max":0.126953125,"train/tensor_param_model_layers_24_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_22_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_53_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_57_mlp_down_proj_weight/range":0.25,"train/tensor_act_model_layers_43_mlp_up_proj/mean":-0.0303955078125,"train/tensor_act_model_layers_71_residual_post_attn/range":6,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_6_mlp_down_proj_weight/range":0.23193359375,"train/tensor_act_model_layers_36_mlp/min":4,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_86_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/std":0.044921875,"train/tensor_act_model_layers_84_mlp_down_proj/max_abs":16.25,"train/tensor_act_model_layers_76_mlp_down_proj/norm":30778.47265625,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_98_residual_post_mlp/min":392,"train/tensor_act_model_layers_42_residual_pre_attn/mean":168,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/norm":11712,"train/tensor_act_model_layers_50_mlp_up_proj/range":7.59375,"train/tensor_act_model_layers_94_mlp_up_proj/norm":5319.42529296875,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_86_mlp_gate_proj/mean":0.1123046875,"train/tensor_grad_model_layers_91_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_82_mlp_down_proj/max":16.75,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/norm":13888,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_42_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_22_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_78_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_27_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/max_abs":0.1435546875,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_87_mlp_down_proj/min":14.125,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/range":0.2705078125,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/min":-0.12255859375,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_26_mlp_down_proj/max":16.5,"train/tensor_param_model_layers_62_mlp_down_proj_weight/norm":6.375,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_40_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/range":0.24267578125,"train/tensor_act_model_layers_83_mlp_down_proj/max_abs":16.625,"train/tensor_act_model_layers_55_residual_post_mlp/norm":458213.03125,"train/tensor_param_model_layers_1_mlp_down_proj_weight/range":0.2373046875,"train/tensor_param_model_layers_12_mlp_down_proj_weight/std":0.033935546875,"train/tensor_param_model_layers_46_mlp_down_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_70_mlp/std":0,"train/tensor_act_model_layers_97_mlp/min":4,"train/tensor_act_model_layers_3_mlp_down_proj/mean":15,"train/tensor_act_model_layers_55_mlp/mean":4,"train/tensor_param_model_layers_48_mlp_down_proj_weight/range":0.25390625,"train/tensor_act_model_layers_34_residual_post_mlp/max_abs":142,"train/tensor_act_model_layers_41_residual_post_mlp/norm":343525.96875,"train/tensor_act_model_layers_81_residual_pre_attn/max_abs":326,"train/tensor_act_model_layers_3_mlp_gate_proj/mean":-0.01287841796875,"train/tensor_act_model_layers_60_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_29_mlp_up_proj/norm":5270.93603515625,"train/tensor_act_model_layers_15_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/min":-0.1240234375,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_3_mlp/min":4,"train/tensor_act_model_layers_87_residual_post_attn/range":6,"train/tensor_act_model_layers_41_mlp_up_proj/std":1.8671875,"train/tensor_param_model_layers_33_mlp_down_proj_weight/mean":-0.0007781982421875,"train/tensor_act_model_layers_70_mlp/max_abs":4,"train/tensor_act_model_layers_13_mlp/max_abs":4,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_53_mlp_up_proj_weight/range":0.25048828125,"train/tensor_act_model_layers_89_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_55_mlp_gate_proj/max":3.78125,"train/tensor_act_model_layers_81_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_60_mlp_down_proj_weight/min":-0.123046875,"train/tensor_act_model_layers_81_mlp_gate_proj/max_abs":4.15625,"train/tensor_param_model_layers_21_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_92_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_59_residual_pre_attn/std":0.8359375,"train/tensor_act_model_layers_10_residual_post_attn/range":3,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_86_mlp_up_proj_weight/mean":0.000244140625,"train/tensor_param_model_layers_75_mlp_down_proj_weight/norm":6.875,"train/tensor_param_model_layers_77_mlp_down_proj_weight/norm":6.96875,"train/tensor_param_model_layers_47_mlp_up_proj_weight/norm":6.375,"train/tensor_param_model_layers_56_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_residual_pre_attn/std":0.8203125,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/min":-0.123046875,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_75_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_76_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_90_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/loss":7.948851776123047,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_5_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_65_mlp_gate_proj/std":1.8671875,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_85_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_57_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_91_residual_post_mlp/norm":753261.6875,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_35_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_10_mlp_gate_proj/max_abs":3.296875,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_42_mlp_up_proj/mean":0.154296875,"train/tensor_act_model_layers_22_residual_pre_attn/min":86,"train/tensor_act_model_layers_36_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_45_mlp/range":0,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/norm":6.375,"train/tensor_act_model_layers_1_mlp_down_proj/std":1.3828125,"train/tensor_act_model_layers_66_mlp/max":4,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_28_mlp_gate_proj/range":6.78125,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_91_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_30_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_86_residual_post_attn/mean":344,"train/tensor_act_model_layers_49_residual_post_attn/mean":196,"train/tensor_act_model_layers_23_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_61_residual_post_attn/max_abs":246,"train/tensor_act_model_layers_45_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_47_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_46_mlp/max_abs":4,"train/tensor_param_model_layers_23_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_residual_post_mlp/mean":168,"train/tensor_act_model_layers_89_residual_pre_attn/max_abs":358,"train/tensor_act_model_layers_74_mlp_gate_proj/max":3.515625,"train/tensor_act_model_layers_69_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_66_residual_post_attn/mean":264,"train/tensor_act_model_layers_18_residual_pre_attn/mean":72,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_40_mlp_up_proj_weight/norm":6.3125,"train/tensor_act_model_layers_98_mlp_down_proj/range":4.25,"train/tensor_act_model_layers_37_residual_post_mlp/max":154,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_79_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_attn/max_abs":33.5,"train/tensor_act_model_layers_93_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_24_mlp/min":4,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_39_mlp/max":4,"train/tensor_param_model_layers_5_mlp_up_proj_weight/max":0.125,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_75_mlp_gate_proj/norm":5215.18212890625,"train/tensor_act_model_layers_26_residual_post_mlp/std":0.62109375,"train/tensor_act_model_layers_42_mlp_up_proj/max":3.765625,"train/tensor_param_model_layers_0_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_55_residual_post_mlp/mean":224,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_15_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_55_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_59_mlp_up_proj_weight/range":0.25,"train/tensor_act_model_layers_16_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp_down_proj/range":2,"train/tensor_act_model_layers_59_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_65_residual_pre_attn/max":262,"train/tensor_act_model_layers_55_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/max":0.123046875,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_34_residual_post_attn/min":134,"train/tensor_act_model_layers_33_mlp/max":4,"train/tensor_act_model_layers_46_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_4_mlp_down_proj_weight/max_abs":0.11767578125,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_10_residual_pre_attn/std":0.373046875,"train/tensor_param_model_layers_9_mlp_down_proj_weight/norm":6.0625,"train/tensor_act_model_layers_34_residual_post_attn/max":138,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_22_mlp_up_proj/mean":-0.11376953125,"train/tensor_act_model_layers_49_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_50_mlp_gate_proj/max":3.546875,"train/tensor_act_model_layers_91_residual_post_mlp/mean":368,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_23_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_58_mlp_up_proj/max":3.296875,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_16_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_99_mlp_up_proj/mean":0.0252685546875,"train/tensor_act_model_layers_17_residual_pre_attn/std":0.55859375,"train/tensor_act_model_layers_11_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_27_mlp_up_proj_weight/max_abs":0.13671875,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_52_residual_post_attn/max_abs":210,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_residual_post_mlp/std":0.81640625,"train/tensor_act_model_layers_40_residual_pre_attn/max":162,"train/tensor_act_model_layers_21_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_64_mlp_down_proj/std":0.3984375,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/range":0.25146484375,"train/tensor_param_model_layers_14_mlp_up_proj_weight/mean":0.000274658203125,"train/tensor_act_model_layers_48_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_mlp_gate_proj/range":7.53125,"train/tensor_param_model_layers_60_mlp_up_proj_weight/mean":0.00013065338134765625,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_69_mlp_up_proj/max_abs":3.953125,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_83_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_54_mlp_up_proj/min":-3.53125,"train/tensor_act_model_layers_72_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp_up_proj/max_abs":3.546875,"train/tensor_act_model_layers_40_residual_pre_attn/max_abs":162,"train/tensor_act_model_layers_97_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_89_mlp_down_proj_weight/max_abs":0.142578125,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_7_mlp_up_proj/mean":0.07177734375,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_16_mlp_up_proj/norm":5340.6484375,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_67_mlp_down_proj/max":16.375,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/range":0.251953125,"train/tensor_act_model_layers_34_mlp/max_abs":4,"train/tensor_act_model_layers_68_residual_post_attn/max_abs":274,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_21_mlp/max":4,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/max":0.130859375,"train/tensor_act_model_layers_82_residual_post_attn/max_abs":330,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_95_mlp_gate_proj/range":9.03125,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_68_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_69_residual_post_mlp/min":276,"train/tensor_param_model_layers_10_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_53_mlp_up_proj/std":1.8671875,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_73_residual_post_mlp/max_abs":298,"train/tensor_act_model_layers_95_residual_pre_attn/norm":777837.6875,"train/tensor_act_model_layers_49_residual_pre_attn/mean":196,"train/tensor_param_model_layers_43_mlp_up_proj_weight/mean":-0.000240325927734375,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_75_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_67_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/max_abs":0.1298828125,"train/tensor_param_model_layers_79_mlp_up_proj_weight/max":0.1494140625,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_72_mlp/range":0,"train/tensor_act_model_layers_43_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_2_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp_down_proj/norm":30598.66015625,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/mean":0.00058746337890625,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_2_residual_post_mlp/std":0.1357421875,"train/tensor_act_model_layers_29_residual_post_mlp/range":4.5,"train/tensor_act_model_layers_15_mlp_up_proj/max_abs":3.625,"train/tensor_param_model_layers_87_mlp_up_proj_weight/max":0.140625,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_66_mlp_down_proj_weight/mean":-0.0001373291015625,"train/tensor_act_model_layers_6_mlp_up_proj/range":6.9375,"train/tensor_act_model_layers_57_mlp/max_abs":4,"train/tensor_act_model_layers_5_residual_pre_attn/range":1.375,"train/tensor_act_model_layers_33_residual_post_attn/mean":132,"train/tensor_act_model_layers_77_residual_pre_attn/min":304,"train/tensor_act_model_layers_29_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_97_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_80_residual_post_mlp/max":326,"train/tensor_param_model_layers_41_mlp_down_proj_weight/min":-0.125,"train/tensor_param_model_layers_58_mlp_down_proj_weight/max":0.1220703125,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_76_mlp_up_proj/std":1.8203125,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/mean":2.5987625122070312e-05,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_74_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_2_mlp_down_proj_weight/max_abs":0.1005859375,"train/tensor_act_model_layers_30_mlp_up_proj/min":-3.515625,"train/tensor_param_model_layers_78_mlp_up_proj_weight/max":0.1328125,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_31_mlp_gate_proj/std":1.8203125,"train/tensor_param_model_layers_89_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_mlp_up_proj/max_abs":3.859375,"train/tensor_act_model_layers_59_mlp/range":0,"train/tensor_act_model_layers_50_mlp_down_proj/range":1.8125,"train/tensor_act_model_layers_80_mlp_up_proj/min":-3.859375,"train/tensor_act_model_layers_40_mlp_up_proj/norm":5266.75732421875,"train/tensor_act_model_layers_88_mlp_up_proj/max_abs":4.1875,"train/tensor_act_model_layers_41_mlp_gate_proj/mean":-0.1728515625,"train/tensor_act_model_layers_46_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_22_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_mlp_up_proj/norm":5274.54296875,"train/tensor_act_model_layers_62_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_82_mlp_up_proj/range":7.546875,"train/tensor_act_model_layers_19_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_19_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_87_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_51_residual_pre_attn/range":5,"train/tensor_param_model_layers_95_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_26_mlp_up_proj/range":7.234375,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_49_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_88_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_90_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_13_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_68_mlp_gate_proj/min":-3.9375,"train/tensor_param_model_layers_11_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_96_mlp_gate_proj/mean":0.01019287109375,"train/tensor_act_model_layers_37_mlp/norm":8192,"train/tensor_param_model_layers_71_mlp_down_proj_weight/max":0.1328125,"eval/loss":7.94517707824707,"train/tensor_act_model_layers_11_mlp_down_proj/range":2.375,"train/tensor_act_model_layers_22_residual_pre_attn/range":4,"train/tensor_param_model_layers_54_mlp_down_proj_weight/range":0.2431640625,"train/tensor_act_model_layers_95_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_24_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_50_mlp_down_proj/max":16,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_4_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_1_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_70_mlp_gate_proj/norm":5398.74365234375,"train/tensor_param_model_layers_68_mlp_up_proj_weight/mean":-0.000774383544921875,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp/max":4,"train/tensor_act_model_layers_13_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_25_residual_post_attn/max_abs":102,"train/tensor_act_model_layers_10_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_mlp_down_proj/min":14.1875,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_5_residual_post_attn/max_abs":20.875,"train/tensor_act_model_layers_78_mlp_down_proj/range":2.5625,"train/tensor_param_model_layers_14_mlp_up_proj_weight/max_abs":0.1240234375,"train/tensor_act_model_layers_86_mlp/norm":8192,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_50_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_24_mlp_up_proj/max_abs":3.78125,"train/tensor_act_model_layers_86_mlp_gate_proj/std":1.8203125,"train/tensor_act_model_layers_76_residual_pre_attn/max":306,"train/tensor_act_model_layers_47_residual_pre_attn/range":5,"train/tensor_act_model_layers_87_mlp_up_proj/range":7.5,"train/tensor_act_model_layers_31_residual_pre_attn/max":126.5,"train/tensor_param_model_layers_31_mlp_down_proj_weight/mean":0.00164031982421875,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_50_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_97_residual_pre_attn/norm":794221.6875,"train/tensor_act_model_layers_84_mlp/max_abs":4,"train/tensor_param_model_layers_49_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_92_residual_post_mlp/min":368,"train/tensor_act_model_layers_97_mlp_gate_proj/norm":5352.30810546875,"train/tensor_param_model_layers_82_mlp_down_proj_weight/norm":7.46875,"train/tensor_act_model_layers_16_residual_post_mlp/max_abs":69.5,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_45_residual_pre_attn/max":182,"train/tensor_act_model_layers_22_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_mlp_down_proj/max":17.625,"train/tensor_act_model_layers_64_mlp_up_proj/max_abs":3.765625,"train/tensor_act_model_layers_57_mlp/norm":8192,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp/max_abs":4,"train/tensor_param_model_layers_29_mlp_up_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_55_mlp_up_proj/std":1.84375,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_62_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_75_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_94_mlp_up_proj/max":4.59375,"train/tensor_act_model_layers_13_mlp_up_proj/max":3.6875,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/norm":6.625,"train/tensor_param_model_layers_8_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_97_residual_post_mlp/mean":392,"train/tensor_act_model_layers_74_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_45_mlp_up_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_87_mlp_gate_proj/norm":5359.25146484375,"train/tensor_act_model_layers_70_mlp_up_proj/max_abs":4.09375,"train/tensor_act_model_layers_48_residual_pre_attn/min":189,"train/tensor_param_model_layers_31_mlp_up_proj_weight/mean":0.00147247314453125,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/norm":6.5,"train/tensor_param_model_layers_97_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_76_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_46_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_24_mlp_gate_proj/mean":0.0888671875,"train/tensor_act_model_layers_14_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_residual_post_attn/max":302,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/mean":-4.172325134277344e-05,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_88_mlp_down_proj_weight/max":0.138671875,"train/tensor_act_model_layers_46_residual_pre_attn/range":5,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/norm":8.625,"train/tensor_act_model_layers_61_mlp_gate_proj/mean":0.07763671875,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/max_abs":0.146484375,"train/tensor_act_model_layers_57_mlp/range":0,"train/tensor_param_model_layers_12_mlp_down_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_4_residual_pre_attn/min":15.375,"train/tensor_act_model_layers_17_residual_pre_attn/range":3.5,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_75_residual_post_attn/min":296,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_88_mlp_up_proj_weight/max":0.1572265625,"train/tensor_act_model_layers_92_mlp_gate_proj/min":-4.1875,"train/tensor_param_model_layers_20_mlp_up_proj_weight/norm":6.3125,"train/tensor_act_model_layers_82_mlp_up_proj/min":-3.96875,"train/tensor_act_model_layers_6_mlp/norm":8192,"train/tensor_param_model_layers_63_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_76_mlp_gate_proj/max_abs":4.03125,"train/tensor_act_model_layers_22_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_0_mlp_up_proj_weight/mean":8.881092071533203e-06,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_29_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_83_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_residual_post_attn/range":2,"train/tensor_act_model_layers_18_residual_post_attn/norm":147101.890625,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_55_mlp/min":4,"train/tensor_act_model_layers_79_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/max":0.125,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_26_residual_pre_attn/min":102,"train/tensor_param_model_layers_51_mlp_down_proj_weight/norm":6.3125,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_94_mlp_down_proj/std":0.640625,"train/tensor_act_model_layers_83_mlp_down_proj/norm":30768.591796875,"train/tensor_act_model_layers_41_mlp_gate_proj/std":1.8359375,"train/tensor_act_model_layers_62_mlp_gate_proj/min":-4.125,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_14_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_5_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_91_residual_post_mlp/range":6,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_78_mlp_down_proj_weight/mean":-0.0001239776611328125,"train/tensor_act_model_layers_8_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_5_mlp_gate_proj/mean":0.099609375,"train/tensor_param_model_layers_12_mlp_up_proj_weight/norm":6.21875,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_2_residual_post_attn/norm":16379.7919921875,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_53_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_99_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_mlp_gate_proj/max":4.0625,"train/tensor_act_model_layers_83_mlp_up_proj/norm":5420.91796875,"train/tensor_act_model_layers_20_residual_post_mlp/min":82,"train/tensor_param_model_layers_5_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_down_proj/max_abs":16.625,"train/tensor_act_model_layers_59_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_19_mlp/norm":8192,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_27_residual_pre_attn/max_abs":110,"train/tensor_act_model_layers_72_residual_post_mlp/range":6,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/range":0.23681640625,"train/tensor_act_model_layers_97_residual_post_mlp/norm":802413.6875,"train/tensor_act_model_layers_26_mlp_gate_proj/max":3.765625,"train/tensor_param_model_layers_48_mlp_up_proj_weight/max_abs":0.1220703125,"train/tensor_param_model_layers_23_mlp_down_proj_weight/std":0.034423828125,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_65_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_18_mlp_down_proj_weight/min":-0.12353515625,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_30_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp_gate_proj/max":4.40625,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/max_abs":0.150390625,"train/tensor_act_model_layers_62_residual_pre_attn/std":0.8359375,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_80_mlp_down_proj_weight/mean":-0.0004024505615234375,"train/tensor_act_model_layers_81_mlp_down_proj/std":0.38671875,"train/tensor_act_model_layers_22_residual_post_mlp/std":0.58203125,"train/tensor_param_model_layers_38_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_residual_pre_attn/std":0.8359375,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/range":0.248046875,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_76_mlp_up_proj_weight/max":0.1328125,"train/tensor_act_model_layers_4_residual_post_mlp/min":19.25,"train/tensor_act_model_layers_47_residual_post_attn/max_abs":190,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_4_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_18_residual_post_mlp/max":78,"train/tensor_act_model_layers_70_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_9_mlp_down_proj_weight/max":0.123046875,"train/tensor_param_model_layers_71_mlp_up_proj_weight/norm":6.90625,"train/tensor_act_model_layers_57_mlp_down_proj/norm":30506.041015625,"train/tensor_act_model_layers_21_mlp_gate_proj/norm":5358.1181640625,"train/tensor_param_model_layers_74_mlp_up_proj_weight/mean":2.6702880859375e-05,"train/tensor_param_model_layers_96_mlp_up_proj_weight/range":0.2998046875,"train/tensor_act_model_layers_14_residual_pre_attn/max_abs":57.5,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_24_mlp_down_proj_weight/mean":-0.000461578369140625,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_99_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_17_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_13_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_21_mlp_up_proj/max_abs":3.59375,"train/tensor_param_model_layers_19_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_30_residual_post_attn/norm":245154.65625,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_32_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_18_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_64_mlp/max":4,"train/tensor_act_model_layers_73_mlp_gate_proj/min":-3.625,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_15_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp/norm":8192,"train/tensor_act_model_layers_55_residual_post_mlp/range":5,"train/tensor_act_model_layers_24_mlp_gate_proj/min":-3.84375,"train/tensor_act_model_layers_40_mlp_up_proj/min":-3.625,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_75_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/std":0.035888671875,"train/tensor_act_model_layers_46_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_89_mlp_down_proj_weight/norm":8,"train/tensor_grad_model_layers_57_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_87_mlp_gate_proj/max":4.15625,"train/tensor_param_model_layers_94_mlp_down_proj_weight/max_abs":0.1513671875,"train/tensor_act_model_layers_51_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_mlp_up_proj/std":1.8046875,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/range":0.21630859375,"train/tensor_act_model_layers_55_residual_pre_attn/max":222,"train/tensor_act_model_layers_28_mlp_up_proj/mean":-0.0005645751953125,"train/tensor_act_model_layers_62_residual_post_attn/range":6,"train/tensor_act_model_layers_75_mlp_up_proj/std":1.8203125,"train/tensor_act_model_layers_73_mlp_up_proj/max_abs":3.78125,"train/tensor_act_model_layers_44_residual_post_attn/norm":359909.75,"train/tensor_act_model_layers_24_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_43_mlp_down_proj_weight/max":0.125,"train/tensor_param_model_layers_72_mlp_up_proj_weight/norm":6.875,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_44_mlp/range":0,"train/tensor_act_model_layers_66_residual_post_attn/range":6,"train/tensor_act_model_layers_13_residual_post_attn/max_abs":53.5,"train/tensor_act_model_layers_33_mlp_up_proj/norm":5263.46826171875,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_97_residual_post_mlp/max_abs":394,"train/tensor_param_model_layers_36_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_39_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/max":0.1357421875,"train/tensor_param_model_layers_83_mlp_up_proj_weight/mean":-0.0003108978271484375,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_49_mlp_gate_proj/range":7.28125,"train/tensor_param_model_layers_35_mlp_up_proj_weight/max":0.125,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/norm":6.34375,"train/tensor_param_model_layers_96_mlp_down_proj_weight/std":0.0478515625,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_55_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/range":0.23583984375,"train/tensor_act_model_layers_15_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_67_mlp_up_proj_weight/range":0.2568359375,"train/tensor_act_model_layers_11_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/mean":0.00103759765625,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_55_mlp_down_proj/max":16.125,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_76_mlp_up_proj/max_abs":3.765625,"train/tensor_act_model_layers_92_residual_pre_attn/max_abs":370,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_25_residual_pre_attn/min":98,"train/tensor_param_model_layers_44_mlp_down_proj_weight/min":-0.1279296875,"train/tensor_param_model_layers_39_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_67_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_27_mlp_up_proj/norm":5425.6552734375,"train/tensor_act_model_layers_60_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_67_mlp_up_proj/range":7.609375,"train/tensor_act_model_layers_69_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_22_mlp_gate_proj/min":-4.125,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_70_mlp_up_proj_weight/range":0.2626953125,"train/tensor_act_model_layers_68_mlp_down_proj/mean":14.9375,"train/tensor_param_model_layers_79_mlp_down_proj_weight/std":0.03955078125,"train/tensor_act_model_layers_30_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_residual_post_mlp/range":6,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_98_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp/std":0,"train/tensor_act_model_layers_60_residual_post_mlp/range":5,"train/tensor_act_model_layers_85_residual_pre_attn/max_abs":342,"train/tensor_act_model_layers_50_mlp_gate_proj/mean":0.05419921875,"train/tensor_act_model_layers_8_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_79_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_32_residual_post_attn/min":125.5,"train/tensor_param_model_layers_98_mlp_down_proj_weight/mean":-0.0004253387451171875,"train/tensor_act_model_layers_47_residual_pre_attn/max_abs":190,"train/tensor_act_model_layers_10_residual_pre_attn/mean":40,"train/tensor_act_model_layers_71_mlp_up_proj/min":-3.515625,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_79_mlp_up_proj/range":7.859375,"train/tensor_act_model_layers_68_mlp/max_abs":4,"train/tensor_act_model_layers_46_mlp_down_proj/max_abs":16.125,"train/tensor_param_model_layers_75_mlp_down_proj_weight/max_abs":0.1337890625,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/mean":0.00110626220703125,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_77_mlp_up_proj_weight/min":-0.134765625,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_28_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_68_mlp/std":0,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_28_mlp_down_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_44_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_45_residual_post_attn/min":177,"train/tensor_act_model_layers_60_mlp_down_proj/std":0.427734375,"train/tensor_act_model_layers_46_residual_post_attn/max":186,"train/tensor_act_model_layers_36_residual_post_attn/max":146,"train/tensor_param_model_layers_37_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_17_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_44_mlp_gate_proj/max_abs":3.90625,"train/tensor_param_model_layers_55_mlp_down_proj_weight/max":0.12451171875,"train/tensor_param_model_layers_39_mlp_down_proj_weight/max":0.1259765625,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_55_mlp/range":0,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_75_mlp_down_proj_weight/std":0.037841796875,"train/tensor_param_model_layers_48_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_84_mlp_down_proj/std":0.42578125,"train/tensor_param_model_layers_69_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/mean":0.0003376007080078125,"train/tensor_act_model_layers_34_mlp_gate_proj/range":6.984375,"train/tensor_param_model_layers_11_mlp_up_proj_weight/max":0.12353515625,"train/tensor_param_model_layers_57_mlp_up_proj_weight/max":0.130859375,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_48_mlp_up_proj/mean":0.01458740234375,"train/tensor_act_model_layers_60_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_82_mlp_down_proj_weight/std":0.041259765625,"train/tensor_param_model_layers_73_mlp_down_proj_weight/mean":-0.0009613037109375,"train/tensor_act_model_layers_45_residual_post_attn/max":182,"train/tensor_act_model_layers_36_residual_post_attn/min":141,"train/tensor_act_model_layers_46_residual_post_attn/min":181,"train/tensor_act_model_layers_72_residual_pre_attn/min":284,"train/tensor_param_model_layers_47_mlp_down_proj_weight/max":0.12451171875,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_56_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_39_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_44_mlp_gate_proj/mean":0.17578125,"train/tensor_act_model_layers_8_mlp/std":0,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_99_residual_post_mlp/range":6,"train/tensor_act_model_layers_67_residual_post_attn/max_abs":270,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_73_mlp_down_proj/max_abs":16.125,"train/tensor_param_model_layers_12_mlp_up_proj_weight/max_abs":0.1279296875,"train/tensor_param_model_layers_13_mlp_down_proj_weight/mean":0.0019073486328125,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/range":0.3251953125,"train/tensor_act_model_layers_32_residual_post_attn/max":130,"train/tensor_act_model_layers_15_mlp/std":0,"train/tensor_act_model_layers_7_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_49_mlp_up_proj_weight/norm":6.375,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_17_mlp_down_proj_weight/min":-0.1171875,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_88_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_74_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_45_mlp_gate_proj/range":7.125,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_50_mlp_up_proj_weight/mean":0.0003223419189453125,"train/tensor_param_model_layers_98_mlp_down_proj_weight/std":0.04736328125,"train/tensor_act_model_layers_53_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_61_residual_post_mlp/max":250,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/range":0.25390625,"train/tensor_param_model_layers_95_mlp_up_proj_weight/mean":0.00099945068359375,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_56_residual_pre_attn/max_abs":226,"train/tensor_act_model_layers_15_residual_pre_attn/max":61.5,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_68_mlp_up_proj_weight/std":0.037353515625,"train/tensor_param_model_layers_41_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_residual_post_attn/min":368,"train/tensor_param_model_layers_11_mlp_up_proj_weight/range":0.244140625,"train/tensor_act_model_layers_92_mlp_up_proj/min":-4.5625,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/mean":0.00148773193359375,"train/tensor_act_model_layers_61_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_8_mlp_up_proj_weight/norm":6.15625,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_76_mlp_gate_proj/norm":5204.3583984375,"train/tensor_act_model_layers_84_residual_post_mlp/max_abs":342,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_91_mlp_up_proj_weight/range":0.291015625,"train/tensor_act_model_layers_8_residual_post_attn/max":33.5,"train/tensor_act_model_layers_50_residual_pre_attn/std":0.81640625,"train/tensor_act_model_layers_45_mlp_gate_proj/mean":-0.015625,"train/tensor_act_model_layers_70_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_41_mlp_down_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_27_mlp_up_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_91_mlp_down_proj_weight/max":0.146484375,"train/tensor_act_model_layers_39_mlp_gate_proj/max_abs":3.875,"train/tensor_act_model_layers_4_mlp/range":0,"train/tensor_act_model_layers_44_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_residual_post_mlp/range":4,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/range":0.267578125,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_18_residual_pre_attn/max_abs":73.5,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_5_mlp_up_proj_weight/std":0.032958984375,"train/tensor_param_model_layers_60_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_94_mlp/norm":8192,"train/tensor_act_model_layers_8_residual_pre_attn/mean":32,"train/tensor_act_model_layers_71_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_12_mlp/max_abs":4,"train/tensor_act_model_layers_94_mlp_down_proj/norm":30612.333984375,"train/tensor_act_model_layers_66_residual_post_attn/min":260,"train/tensor_act_model_layers_87_mlp/max":4,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/max":0.125,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_66_residual_post_attn/max":266,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_28_residual_post_mlp/max":118.5,"train/tensor_param_model_layers_12_mlp_up_proj_weight/max":0.1279296875,"train/tensor_param_model_layers_55_mlp_down_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_95_residual_post_mlp/norm":786029.6875,"train/tensor_param_model_layers_2_mlp_down_proj_weight/std":0.03076171875,"train/tensor_act_model_layers_81_mlp_up_proj/min":-3.859375,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_89_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_19_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_28_mlp_down_proj/std":0.37109375,"train/tensor_act_model_layers_33_mlp_gate_proj/max_abs":4.125,"train/tensor_act_model_layers_82_mlp/max_abs":4,"train/tensor_act_model_layers_91_residual_post_mlp/std":0.7578125,"train/tensor_param_model_layers_81_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_up_proj/max":3.609375,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_33_residual_post_mlp/max_abs":138,"train/tensor_param_model_layers_91_mlp_down_proj_weight/mean":0.001190185546875,"train/tensor_param_model_layers_46_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_97_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_residual_post_attn/max":374,"train/tensor_param_model_layers_62_mlp_down_proj_weight/max_abs":0.1259765625,"train/tensor_act_model_layers_72_mlp_up_proj/mean":-0.0091552734375,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_12_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_56_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_10_mlp_gate_proj/range":6.515625,"train/tensor_act_model_layers_19_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_47_residual_post_attn/max":190,"train/tensor_param_model_layers_70_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_9_residual_post_attn/min":34.5,"train/tensor_act_model_layers_44_residual_post_attn/min":173,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_4_mlp_up_proj_weight/min":-0.103515625,"train/tensor_act_model_layers_79_mlp_gate_proj/max":4.03125,"train/tensor_act_model_layers_53_residual_pre_attn/max_abs":214,"train/tensor_act_model_layers_40_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_78_mlp_down_proj_weight/max_abs":0.1357421875,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_71_residual_pre_attn/max_abs":286,"train/tensor_act_model_layers_51_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_0_mlp_gate_proj/std":1.6171875,"train/tensor_act_model_layers_50_mlp/mean":4,"train/tensor_act_model_layers_72_mlp_gate_proj/max_abs":4.03125,"train/tensor_act_model_layers_62_mlp_up_proj/min":-3.859375,"train/tensor_act_model_layers_93_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/range":0.24951171875,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_11_mlp_up_proj_weight/mean":1.4185905456542969e-05,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_71_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_85_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_43_mlp_down_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_21_mlp_gate_proj/min":-3.578125,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_49_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_65_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_9_residual_pre_attn/std":0.373046875,"train/tensor_act_model_layers_38_mlp/norm":8192,"train/tensor_act_model_layers_53_mlp/max":4,"train/tensor_act_model_layers_46_mlp_down_proj/std":0.373046875,"train/tensor_act_model_layers_20_mlp_up_proj/norm":5477.2373046875,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_74_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_96_mlp_gate_proj/std":1.859375,"train/tensor_param_model_layers_35_mlp_down_proj_weight/std":0.034912109375,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_89_mlp_up_proj_weight/mean":0.000698089599609375,"train/tensor_act_model_layers_31_mlp_gate_proj/max_abs":3.75,"train/tensor_param_model_layers_90_mlp_up_proj_weight/max_abs":0.142578125,"train/tensor_param_model_layers_65_mlp_down_proj_weight/std":0.03515625,"train/tensor_act_model_layers_6_mlp_gate_proj/max":3.328125,"train/tensor_act_model_layers_43_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_residual_post_mlp/mean":384,"train/tensor_act_model_layers_69_mlp_gate_proj/min":-4,"train/tensor_grad_model_layers_60_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_29_mlp_down_proj_weight/max":0.126953125,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_99_mlp_gate_proj/min":-4.5625,"train/tensor_act_model_layers_66_mlp_up_proj/max":3.5625,"train/tensor_act_model_layers_43_residual_post_mlp/std":0.80859375,"train/tensor_param_model_layers_64_mlp_down_proj_weight/min":-0.126953125,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_22_mlp_up_proj/max":3.234375,"train/tensor_act_model_layers_73_mlp/std":0,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_44_residual_post_attn/max":178,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_43_mlp/std":0,"train/tensor_act_model_layers_95_mlp_down_proj/mean":15,"train/tensor_act_model_layers_43_mlp_gate_proj/range":7.59375,"train/tensor_param_model_layers_58_mlp_down_proj_weight/range":0.25390625,"train/tensor_act_model_layers_40_mlp_down_proj/norm":30336.728515625,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_63_residual_post_attn/mean":252,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_8_mlp/mean":4,"train/tensor_act_model_layers_7_mlp/norm":8192,"train/tensor_param_model_layers_30_mlp_down_proj_weight/range":0.24072265625,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_94_mlp_up_proj_weight/max":0.1416015625,"train/tensor_act_model_layers_47_residual_post_attn/min":185,"train/tensor_act_model_layers_38_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_4_residual_post_mlp/max":20.625,"train/tensor_act_model_layers_58_mlp_gate_proj/max_abs":3.9375,"train/tensor_act_model_layers_16_mlp_down_proj/norm":30384.935546875,"train/tensor_act_model_layers_52_residual_post_attn/mean":208,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_7_mlp_gate_proj/std":1.890625,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/norm":6.5,"train/tensor_param_model_layers_53_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_28_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_55_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/norm":6.5625,"train/tensor_act_model_layers_33_residual_post_mlp/std":0.71484375,"train/tensor_act_model_layers_30_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_20_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_mlp_up_proj/min":-3.734375,"train/tensor_act_model_layers_58_residual_post_attn/mean":232,"train/tensor_act_model_layers_17_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_mlp/mean":4,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/std":0.03173828125,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/max_abs":390,"train/tensor_act_model_layers_93_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_65_mlp_down_proj/range":2.0625,"train/tensor_act_model_layers_36_mlp_up_proj/norm":5351.44091796875,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_34_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_63_mlp_up_proj_weight/std":0.035888671875,"train/tensor_act_model_layers_46_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_36_residual_post_attn/std":0.80859375,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_30_mlp_down_proj/max_abs":16,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_14_mlp_down_proj_weight/min":-0.12060546875,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_89_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_9_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/mean":0.000926971435546875,"train/tensor_act_model_layers_15_mlp_gate_proj/mean":-0.091796875,"train/tensor_act_model_layers_12_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_77_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_45_residual_post_mlp/max_abs":186,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_67_mlp_up_proj/mean":-0.1416015625,"train/tensor_act_model_layers_40_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_mlp_gate_proj/std":1.8046875,"train/tensor_act_model_layers_26_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/max":0.1279296875,"train/tensor_act_model_layers_46_mlp_gate_proj/range":6.859375,"train/tensor_param_model_layers_20_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_56_mlp_up_proj/max":3.90625,"train/tensor_act_model_layers_45_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_56_residual_pre_attn/min":221,"train/tensor_act_model_layers_75_residual_post_mlp/norm":622190.125,"train/tensor_act_model_layers_13_residual_post_mlp/max_abs":57.5,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/max":0.1337890625,"train/tensor_act_model_layers_20_residual_post_mlp/max":86,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_26_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_58_mlp_up_proj/norm":5231.1708984375,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/max":0.1279296875,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_54_mlp/max":4,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/min":-0.123046875,"train/tensor_param_model_layers_9_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_residual_pre_attn/mean":172,"train/tensor_act_model_layers_8_mlp_gate_proj/max_abs":3.328125,"train/tensor_act_model_layers_74_mlp_down_proj/norm":30574.36328125,"train/tensor_act_model_layers_69_residual_pre_attn/max_abs":278,"train/tensor_act_model_layers_87_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_mlp_gate_proj/max_abs":4.125,"train/tensor_act_model_layers_40_mlp_down_proj/std":0.33984375,"train/tensor_act_model_layers_75_mlp_up_proj/mean":0.000949859619140625,"train/tensor_act_model_layers_28_residual_pre_attn/range":4,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/std":0.031494140625,"train/tensor_act_model_layers_32_residual_post_attn/std":0.65625,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_14_residual_post_attn/mean":55.75,"train/tensor_act_model_layers_43_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_90_mlp_down_proj_weight/max":0.14453125,"train/tensor_act_model_layers_78_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_mlp_gate_proj/norm":5307.35693359375,"train/tensor_act_model_layers_44_residual_post_mlp/norm":368101.6875,"train/tensor_param_model_layers_10_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_83_mlp_gate_proj/min":-4.28125,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_87_mlp_down_proj/mean":15,"train/tensor_act_model_layers_60_mlp_gate_proj/mean":-0.04833984375,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_61_mlp_gate_proj/min":-3.75,"train/tensor_act_model_layers_75_mlp/range":0,"train/tensor_act_model_layers_93_residual_post_mlp/mean":376,"train/tensor_act_model_layers_71_mlp_up_proj/max_abs":3.5625,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_47_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_81_mlp/std":0,"train/tensor_act_model_layers_32_mlp_gate_proj/min":-3.578125,"train/tensor_param_model_layers_45_mlp_up_proj_weight/min":-0.123046875,"train/tensor_param_model_layers_71_mlp_up_proj_weight/min":-0.12890625,"train/tensor_act_model_layers_27_residual_pre_attn/max":110,"train/tensor_act_model_layers_49_mlp_up_proj/mean":0.13671875,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_59_mlp_down_proj/max":16.25,"train/tensor_act_model_layers_12_residual_post_attn/norm":98137.9921875,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_95_residual_post_mlp/std":0.7578125,"train/tensor_param_model_layers_81_mlp_up_proj_weight/norm":7.375,"train/tensor_act_model_layers_89_mlp/max_abs":4,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_35_mlp_up_proj_weight/max_abs":0.12890625,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_36_residual_pre_attn/norm":294374.625,"train/tensor_act_model_layers_8_mlp_up_proj/max_abs":3.34375,"train/tensor_act_model_layers_46_residual_pre_attn/norm":376293.59375,"train/tensor_act_model_layers_93_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_90_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_15_mlp_down_proj_weight/norm":6.125,"train/tensor_act_model_layers_90_mlp_up_proj/range":8.390625,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/mean":-0.000720977783203125,"train/tensor_act_model_layers_11_residual_post_attn/max_abs":45.5,"train/tensor_act_model_layers_90_residual_pre_attn/range":6,"train/tensor_act_model_layers_73_residual_post_attn/norm":597614.1875,"train/tensor_act_model_layers_48_residual_post_attn/norm":392693.5,"train/tensor_act_model_layers_56_mlp_gate_proj/max":3.875,"train/tensor_act_model_layers_20_mlp_up_proj/range":7.03125,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_residual_post_mlp/mean":268,"train/tensor_act_model_layers_88_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_13_mlp/mean":4,"train/tensor_act_model_layers_97_mlp_gate_proj/min":-4.3125,"train/tensor_act_model_layers_17_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_71_mlp_down_proj_weight/max_abs":0.1337890625,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/norm":8.3125,"train/tensor_act_model_layers_32_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_8_mlp_up_proj/max":3.34375,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_86_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_52_mlp_down_proj/std":0.384765625,"train/tensor_param_model_layers_33_mlp_down_proj_weight/std":0.034912109375,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_99_mlp_gate_proj/max_abs":4.84375,"train/tensor_act_model_layers_59_mlp_down_proj/min":14.1875,"train/tensor_param_model_layers_83_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_45_residual_pre_attn/norm":368101.6875,"train/tensor_act_model_layers_89_residual_pre_attn/range":6,"train/tensor_act_model_layers_29_mlp/max":4,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_37_residual_pre_attn/range":5,"train/tensor_act_model_layers_6_mlp_up_proj/std":1.828125,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_2_mlp_gate_proj/norm":5336.8955078125,"train/tensor_act_model_layers_42_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_83_mlp_up_proj_weight/min":-0.1337890625,"train/tensor_param_model_layers_68_mlp_down_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_47_residual_post_mlp/norm":392677.4375,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_49_mlp_down_proj/max_abs":15.9375,"train/tensor_act_model_layers_89_residual_pre_attn/max":358,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/mean":-0.000579833984375,"train/tensor_act_model_layers_21_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_34_mlp_up_proj_weight/min":-0.12158203125,"train/tensor_act_model_layers_92_mlp_down_proj/max_abs":17,"train/tensor_act_model_layers_88_mlp/norm":8192,"train/tensor_act_model_layers_16_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_0_mlp_down_proj_weight/range":0.25,"train/tensor_param_model_layers_35_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_45_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_57_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp_gate_proj/max_abs":3.515625,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_59_mlp_up_proj/min":-3.96875,"train/tensor_act_model_layers_32_residual_pre_attn/norm":261421.984375,"train/tensor_act_model_layers_58_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_53_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/range":0.25,"train/tensor_act_model_layers_69_residual_post_mlp/max":282,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/std":0.04296875,"train/tensor_act_model_layers_31_residual_pre_attn/range":4.5,"train/tensor_act_model_layers_18_mlp_up_proj/max_abs":3.71875,"train/tensor_act_model_layers_38_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_70_residual_pre_attn/max_abs":282,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_41_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_32_residual_post_mlp/mean":132,"train/tensor_act_model_layers_50_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_61_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_71_mlp_gate_proj/std":1.859375,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_residual_post_attn/max_abs":302,"train/tensor_param_model_layers_64_mlp_down_proj_weight/range":0.255859375,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_86_residual_pre_attn/max":346,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_1_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_45_residual_post_mlp/mean":184,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/norm":7,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_8_mlp_down_proj_weight/max":0.123046875,"train/tensor_act_model_layers_85_mlp_up_proj/range":8.203125,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_param_model_layers_27_mlp_up_proj_weight/norm":6.40625,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_param_model_layers_65_mlp_up_proj_weight/range":0.259765625,"train/tensor_act_model_layers_66_residual_pre_attn/norm":540270.375,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/norm":6.3125,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/std":0.0361328125,"train/tensor_act_model_layers_6_residual_pre_attn/range":1.75,"train/tensor_act_model_layers_28_mlp_down_proj/norm":30379.91796875,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/min":-0.1298828125,"train/tensor_param_model_layers_95_mlp_up_proj_weight/max_abs":0.1513671875,"train/tensor_act_model_layers_36_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/range":0.2685546875,"train/tensor_param_model_layers_66_mlp_down_proj_weight/std":0.036376953125,"train/tensor_act_model_layers_7_mlp/std":0,"train/tensor_param_model_layers_1_mlp_down_proj_weight/std":0.033203125,"train/tensor_act_model_layers_46_residual_post_mlp/mean":188,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_36_residual_post_mlp/mean":148,"train/tensor_act_model_layers_62_mlp/max":4,"train/tensor_act_model_layers_70_mlp_down_proj/mean":15,"train/tensor_param_model_layers_7_mlp_up_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_47_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/min":-250,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_37_mlp_down_proj/std":0.392578125,"train/tensor_param_model_layers_76_mlp_down_proj_weight/min":-0.1328125,"train/tensor_act_model_layers_28_mlp/max_abs":4,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_62_mlp_gate_proj/mean":-0.00531005859375,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_97_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_65_mlp/std":0,"train/tensor_act_model_layers_80_residual_post_mlp/min":320,"train/tensor_act_model_layers_38_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_48_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_93_residual_pre_attn/norm":761453.75,"train/tensor_param_model_layers_85_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_18_mlp_up_proj/range":7,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/mean":-0.00011777877807617188,"train/tensor_act_model_layers_42_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_40_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_5_mlp_up_proj/range":6.59375,"train/tensor_param_model_layers_95_mlp_down_proj_weight/max_abs":0.1640625,"train/tensor_act_model_layers_68_residual_pre_attn/max":274,"train/tensor_act_model_layers_47_residual_pre_attn/norm":384485.53125,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_14_mlp_up_proj/std":1.859375,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/std":0.035888671875,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_13_residual_pre_attn/max":53.5,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_32_residual_post_mlp/norm":269781.90625,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_41_residual_post_attn/norm":335334.0625,"train/tensor_act_model_layers_71_mlp/norm":8192,"train/tensor_param_model_layers_42_mlp_up_proj_weight/max":0.12109375,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_8_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_23_residual_post_attn/mean":92,"train/tensor_act_model_layers_26_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_20_mlp_gate_proj/max":3.515625,"train/tensor_act_model_layers_56_mlp_up_proj/range":7.890625,"train/tensor_act_model_layers_18_mlp/range":0,"train/tensor_act_model_layers_25_mlp_up_proj/mean":0.185546875,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/max":0.138671875,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_78_mlp_down_proj/max":16.875,"train/tensor_act_model_layers_16_mlp_gate_proj/std":1.8359375,"train/tensor_param_model_layers_61_mlp_up_proj_weight/range":0.25390625,"train/tensor_act_model_layers_90_mlp_gate_proj/max":3.984375,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_64_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_75_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_25_residual_post_attn/range":4,"train/tensor_act_model_layers_74_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_45_residual_post_mlp/norm":376293.59375,"train/tensor_act_model_layers_86_mlp_gate_proj/range":7.734375,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_76_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_70_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_27_mlp_down_proj_weight/max_abs":0.12451171875,"train/tensor_act_model_layers_66_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/std":0.042236328125,"train/tensor_act_model_layers_74_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_46_residual_post_mlp/norm":384485.53125,"train/tensor_act_model_layers_36_residual_post_mlp/norm":302566.5,"train/tensor_act_model_layers_19_mlp/std":0,"train/tensor_param_model_layers_10_mlp_down_proj_weight/min":-0.11962890625,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_92_mlp_down_proj_weight/max_abs":0.1357421875,"train/tensor_act_model_layers_71_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_0_mlp_up_proj/max_abs":5.625,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/range":0.2998046875,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_68_mlp_down_proj_weight/norm":6.71875,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_87_residual_pre_attn/min":344,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/norm":6.875,"train/tensor_act_model_layers_6_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/std":0.0439453125,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_78_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_65_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/max_abs":0.12890625,"train/tensor_grad_model_layers_70_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_68_mlp_gate_proj/norm":5413.30712890625,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_99_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_10_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_44_residual_pre_attn/norm":359909.75,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_69_mlp_down_proj_weight/range":0.2578125,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_70_mlp_up_proj_weight/mean":0.000606536865234375,"train/tensor_param_model_layers_99_mlp_up_proj_weight/range":0.2978515625,"train/tensor_act_model_layers_22_mlp_gate_proj/range":8.015625,"train/tensor_act_model_layers_87_mlp/max_abs":4,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_34_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_70_mlp_gate_proj/min":-3.875,"train/tensor_act_model_layers_19_mlp_gate_proj/mean":0.034423828125,"train/tensor_param_model_layers_53_mlp_up_proj_weight/mean":-0.0003223419189453125,"train/tensor_act_model_layers_89_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/std":0.03466796875,"train/tensor_grad_model_layers_14_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_52_residual_post_mlp/max_abs":214,"train/tensor_param_model_layers_28_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_6_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_35_mlp_down_proj/norm":30473.431640625,"train/tensor_act_model_layers_5_mlp_gate_proj/max":3.21875,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_65_residual_post_attn/norm":532078.4375,"train/tensor_act_model_layers_51_residual_pre_attn/min":201,"train/tensor_act_model_layers_58_mlp_gate_proj/range":7.4375,"train/tensor_act_model_layers_44_residual_post_mlp/mean":180,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_51_mlp/min":4,"train/tensor_param_model_layers_48_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp_up_proj/mean":-0.19921875,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_residual_post_mlp/norm":769645.75,"train/tensor_param_model_layers_84_mlp_down_proj_weight/std":0.0419921875,"train/tensor_param_model_layers_69_mlp_down_proj_weight/std":0.036865234375,"train/tensor_act_model_layers_94_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp_up_proj/std":1.859375,"train/tensor_act_model_layers_58_mlp_down_proj/std":0.39453125,"train/tensor_param_model_layers_4_mlp_up_proj_weight/mean":0.000240325927734375,"train/tensor_param_model_layers_26_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_61_residual_post_mlp/min":245,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_63_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_32_mlp_down_proj/max_abs":16.25,"train/tensor_act_model_layers_85_mlp_gate_proj/max":4.21875,"train/tensor_param_model_layers_14_mlp_down_proj_weight/norm":6.03125,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_66_residual_post_mlp/norm":548462.375,"train/tensor_param_model_layers_63_mlp_down_proj_weight/min":-0.12353515625,"train/tensor_act_model_layers_95_residual_post_attn/max":382,"train/tensor_act_model_layers_61_mlp_gate_proj/range":7.421875,"train/tensor_param_model_layers_48_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_93_mlp_up_proj/max_abs":4.375,"train/tensor_param_model_layers_75_mlp_up_proj_weight/norm":6.9375,"train/tensor_act_model_layers_64_mlp_up_proj/mean":0.10302734375,"train/tensor_act_model_layers_22_mlp/range":0,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_93_mlp/norm":8192,"train/tensor_act_model_layers_95_residual_post_attn/min":376,"train/tensor_param_model_layers_4_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_54_mlp_gate_proj/max_abs":3.859375,"train/tensor_act_model_layers_0_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_42_mlp_down_proj/range":2.375,"train/tensor_act_model_layers_95_residual_pre_attn/range":6,"train/tensor_act_model_layers_63_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_73_mlp_up_proj/max":3.78125,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_18_mlp_gate_proj/max":3.421875,"train/tensor_act_model_layers_38_mlp/range":0,"train/tensor_act_model_layers_97_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_5_mlp_up_proj_weight/range":0.251953125,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/std":0.03515625,"train/tensor_param_model_layers_34_mlp_up_proj_weight/norm":6.28125,"train/tensor_act_model_layers_35_residual_post_mlp/std":0.80859375,"train/tensor_param_model_layers_97_mlp_down_proj_weight/mean":-0.0012359619140625,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_69_residual_pre_attn/mean":276,"train/tensor_act_model_layers_28_residual_post_mlp/min":114,"train/tensor_grad_model_layers_54_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_76_mlp/min":4,"train/tensor_act_model_layers_89_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_47_residual_post_mlp/mean":192,"train/tensor_param_model_layers_61_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_22_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_30_mlp/norm":8192,"train/tensor_act_model_layers_64_residual_pre_attn/range":6,"train/tensor_act_model_layers_87_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_47_mlp_up_proj/mean":0.044677734375,"train/tensor_act_model_layers_5_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_residual_pre_attn/std":0.80859375,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_mlp_up_proj/max":4.96875,"train/tensor_param_model_layers_72_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_76_mlp_down_proj/mean":15,"train/tensor_param_model_layers_13_mlp_down_proj_weight/max":0.12255859375,"train/tensor_act_model_layers_68_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_69_mlp_up_proj_weight/max":0.1279296875,"train/tensor_act_model_layers_4_residual_post_mlp/range":1.375,"train/tensor_act_model_layers_5_residual_pre_attn/max_abs":20.625,"train/tensor_act_model_layers_18_residual_post_mlp/min":74,"train/tensor_act_model_layers_17_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_73_mlp/min":4,"train/tensor_param_model_layers_28_mlp_down_proj_weight/min":-0.1298828125,"train/tensor_act_model_layers_56_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_97_residual_post_attn/min":384,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/range":0.24560546875,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_51_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_43_mlp_down_proj_weight/mean":-0.000125885009765625,"train/tensor_param_model_layers_36_mlp_up_proj_weight/std":0.034912109375,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_38_mlp_up_proj_weight/mean":-0.0003032684326171875,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_6_mlp/std":0,"train/tensor_act_model_layers_39_residual_post_mlp/range":5,"train/tensor_param_model_layers_92_mlp_up_proj_weight/std":0.0458984375,"train/tensor_param_model_layers_93_mlp_down_proj_weight/mean":-0.000537872314453125,"train/tensor_act_model_layers_42_mlp_down_proj/norm":30547.1796875,"train/tensor_act_model_layers_37_residual_pre_attn/max_abs":150,"train/tensor_param_model_layers_42_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_85_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_86_residual_pre_attn/max_abs":346,"train/tensor_act_model_layers_64_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_14_mlp_gate_proj/max":3.671875,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_28_mlp_gate_proj/max_abs":3.5,"train/tensor_act_model_layers_22_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_76_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_75_mlp_gate_proj/std":1.796875,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_4_residual_post_attn/max":16.625,"train/tensor_act_model_layers_29_mlp/norm":8192,"train/tensor_act_model_layers_75_residual_post_mlp/mean":304,"train/tensor_act_model_layers_75_mlp_down_proj/max_abs":16.375,"train/tensor_act_model_layers_45_residual_post_mlp/range":5,"train/tensor_param_model_layers_81_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/std":0.046142578125,"train/tensor_act_model_layers_85_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_42_mlp_down_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_27_mlp/mean":4,"train/tensor_act_model_layers_14_mlp_gate_proj/range":7.015625,"train/tensor_act_model_layers_71_mlp/min":4,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/range":0.22802734375,"train/tensor_act_model_layers_73_residual_post_attn/range":6,"train/tensor_act_model_layers_64_mlp_down_proj/norm":30503.572265625,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_mlp_gate_proj/max":4.21875,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_54_mlp_down_proj_weight/max_abs":0.12255859375,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/max":0.12451171875,"train/tensor_param_model_layers_61_mlp_down_proj_weight/min":-0.12451171875,"train/tensor_act_model_layers_51_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_92_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_residual_pre_attn/max":266,"train/tensor_act_model_layers_12_mlp_gate_proj/max":3.734375,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_3_mlp_down_proj_weight/max":0.10400390625,"train/tensor_act_model_layers_3_residual_post_attn/range":1.1875,"train/tensor_act_model_layers_97_residual_post_attn/max":390,"train/tensor_act_model_layers_1_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_85_mlp_up_proj_weight/min":-0.142578125,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_37_mlp_down_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_97_mlp_gate_proj/range":8.8125,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_44_mlp_up_proj/norm":5372.28076171875,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_11_mlp_down_proj_weight/mean":0.00084686279296875,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/max_abs":0.16796875,"train/tensor_act_model_layers_4_residual_pre_attn/mean":16,"train/tensor_param_model_layers_78_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_20_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/min":-0.1171875,"train/tensor_param_model_layers_82_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_12_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_23_mlp/norm":8192,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/norm":6.875,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_53_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_69_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_23_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_48_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_40_mlp/mean":4,"train/tensor_act_model_layers_24_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp_down_proj/max":15.9375,"train/tensor_act_model_layers_18_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_47_residual_post_attn/std":0.80859375,"train/tensor_act_model_layers_34_mlp_gate_proj/mean":0.1416015625,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_38_mlp_down_proj_weight/min":-0.12451171875,"train/tensor_grad_model_layers_61_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_29_mlp_down_proj/max_abs":16.375,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/max":0.125,"train/tensor_act_model_layers_88_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_34_mlp/min":4,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_45_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_54_mlp_gate_proj/min":-3.46875,"train/tensor_act_model_layers_81_mlp/min":4,"train/tensor_act_model_layers_22_mlp/max_abs":4,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_7_residual_pre_attn/std":0.2412109375,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_70_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_83_residual_post_attn/range":6,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_52_mlp_down_proj_weight/mean":-0.001220703125,"train/tensor_act_model_layers_77_residual_pre_attn/mean":308,"train/tensor_act_model_layers_7_mlp/max":4,"train/tensor_act_model_layers_63_residual_post_mlp/std":0.76953125,"train/tensor_act_model_layers_36_mlp_down_proj/norm":30473.3828125,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_73_mlp_up_proj_weight/max_abs":0.1279296875,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_35_residual_post_attn/norm":286182.75,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/range":0.2509765625,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_75_residual_pre_attn/norm":613998.125,"train/tensor_act_model_layers_72_mlp_up_proj/std":1.828125,"train/tensor_act_model_layers_31_residual_post_mlp/max_abs":130,"train/tensor_act_model_layers_98_mlp/norm":8192,"train/tensor_act_model_layers_52_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_92_mlp_up_proj_weight/mean":-0.00013446807861328125,"train/tensor_act_model_layers_83_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_24_residual_pre_attn/range":4,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_35_residual_post_attn/max_abs":142,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_62_mlp_down_proj_weight/range":0.24853515625,"train/tensor_act_model_layers_57_mlp/mean":4,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_95_mlp/std":0,"train/tensor_param_model_layers_45_mlp_down_proj_weight/mean":0.000926971435546875,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_92_residual_post_attn/norm":753261.6875,"train/tensor_act_model_layers_50_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_38_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_18_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_residual_pre_attn/max":78,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_99_mlp/max":4,"train/tensor_param_model_layers_24_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_44_residual_post_attn/std":0.80859375,"train/tensor_param_model_layers_94_mlp_down_proj_weight/mean":0.000316619873046875,"train/tensor_act_model_layers_36_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_75_residual_pre_attn/max":302,"train/tensor_act_model_layers_78_mlp_up_proj/max":4.03125,"train/tensor_act_model_layers_31_mlp_down_proj/range":1.625,"train/tensor_act_model_layers_57_mlp_gate_proj/min":-3.75,"train/tensor_act_model_layers_70_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_16_mlp_down_proj_weight/norm":6.15625,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_53_mlp_down_proj/max":16.25,"train/tensor_act_model_layers_67_residual_pre_attn/range":6,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_49_mlp_down_proj/min":14.3125,"train/tensor_param_model_layers_15_mlp_up_proj_weight/max":0.119140625,"train/tensor_act_model_layers_21_mlp_up_proj/min":-3.421875,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/min":-0.12060546875,"train/tensor_act_model_layers_90_mlp_up_proj/mean":-0.06591796875,"train/tensor_param_model_layers_25_mlp_gate_proj_weight/norm":6.3125,"train/tensor_act_model_layers_52_mlp/range":0,"train/tensor_act_model_layers_14_mlp_down_proj/max_abs":16.5,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/mean":-0.00165557861328125,"train/tensor_act_model_layers_99_mlp_down_proj/range":3.8125,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/mean":-0.000247955322265625,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/range":0.2587890625,"train/tensor_act_model_layers_33_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp/range":0,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_32_mlp_down_proj_weight/norm":6.28125,"train/tensor_act_model_layers_16_residual_post_attn/norm":130656.3203125,"train/tensor_act_model_layers_21_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/std":0.03662109375,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_89_mlp_down_proj/std":0.58203125,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_14_mlp/max_abs":4,"train/tensor_act_model_layers_26_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_92_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_4_residual_post_attn/range":1.3125,"train/tensor_act_model_layers_91_mlp_up_proj/max":4.0625,"train/tensor_act_model_layers_65_mlp_up_proj/max":3.453125,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_29_mlp/std":0,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/max_abs":0.11376953125,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/mean":-0.0003452301025390625,"train/tensor_act_model_layers_82_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_24_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_27_residual_post_attn/mean":108,"train/tensor_param_model_layers_22_mlp_down_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_2_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_post_attn/norm":212562.453125,"train/tensor_act_model_layers_14_mlp_up_proj/range":6.78125,"train/tensor_act_model_layers_61_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_34_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp_up_proj/mean":-0.060302734375,"train/tensor_param_model_layers_1_mlp_up_proj_weight/range":0.2373046875,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_94_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_44_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_85_mlp_up_proj/min":-4.21875,"train/tensor_act_model_layers_0_residual_post_mlp/std":0.09228515625,"train/tensor_param_model_layers_67_mlp_down_proj_weight/mean":-0.0028839111328125,"train/tensor_act_model_layers_11_mlp/norm":8192,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_19_residual_post_attn/mean":76,"train/tensor_act_model_layers_4_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_13_residual_pre_attn/range":3,"train/tensor_act_model_layers_16_residual_pre_attn/std":0.515625,"train/tensor_param_model_layers_94_mlp_up_proj_weight/norm":8.5625,"train/tensor_act_model_layers_7_residual_post_attn/max_abs":29.375,"train/tensor_act_model_layers_69_mlp_down_proj/norm":30544.89453125,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_residual_post_attn/max":8.4375,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_12_mlp_up_proj/range":7.15625,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/mean":0.0005340576171875,"train/tensor_act_model_layers_47_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_64_mlp_up_proj_weight/max_abs":0.1279296875,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/min":-0.1513671875,"train/tensor_act_model_layers_91_mlp_up_proj/range":8.125,"train/tensor_act_model_layers_66_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_31_mlp/max":4,"train/tensor_act_model_layers_92_residual_post_mlp/max":374,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_56_mlp_up_proj_weight/std":0.03564453125,"train/tensor_act_model_layers_83_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_66_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_94_residual_pre_attn/max":378,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_42_residual_post_attn/mean":168,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_51_mlp_down_proj_weight/std":0.03466796875,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_2_residual_pre_attn/mean":8,"train/tensor_act_model_layers_85_residual_pre_attn/mean":340,"train/tensor_act_model_layers_17_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_55_mlp_up_proj_weight/mean":0.000743865966796875,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_16_mlp_up_proj_weight/min":-0.125,"train/tensor_param_model_layers_24_mlp_up_proj_weight/max_abs":0.1279296875,"train/tensor_grad_model_layers_4_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_25_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/std":0.04052734375,"train/tensor_act_model_layers_53_residual_pre_attn/std":0.81640625,"train/tensor_param_model_layers_66_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_9_mlp_up_proj/max":3.5,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_7_mlp_up_proj/std":1.84375,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_93_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_12_mlp_down_proj/norm":30285.91796875,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_71_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/mean":-0.00109100341796875,"train/tensor_act_model_layers_15_mlp/norm":8192,"train/tensor_act_model_layers_57_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_56_mlp_gate_proj/norm":5353.33740234375,"train/tensor_act_model_layers_21_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_85_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_56_residual_post_mlp/norm":466404.96875,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_21_residual_post_mlp/range":4,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/min":-0.12158203125,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_22_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_20_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_up_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_29_residual_pre_attn/mean":115.5,"train/tensor_act_model_layers_33_residual_post_attn/min":130,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_13_mlp_down_proj/norm":30575.9375,"train/tensor_grad_model_layers_55_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_17_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_83_mlp_up_proj_weight/max_abs":0.1337890625,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/mean":0.00244140625,"train/tensor_act_model_layers_61_residual_post_mlp/norm":507396.8125,"train/tensor_act_model_layers_84_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_residual_post_attn/mean":44,"train/tensor_act_model_layers_61_mlp_up_proj/mean":-0.2109375,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_92_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_88_mlp/max_abs":4,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_86_mlp_down_proj_weight/max":0.1376953125,"train/tensor_act_model_layers_68_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_35_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_75_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_36_mlp/norm":8192,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/std":0.043212890625,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_70_mlp_up_proj/std":1.8828125,"train/tensor_param_model_layers_80_mlp_up_proj_weight/std":0.039794921875,"train/tensor_act_model_layers_90_mlp_up_proj/min":-4.46875,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_20_residual_post_mlp/range":4,"train/tensor_act_model_layers_92_mlp/std":0,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/mean":-0.000659942626953125,"train/tensor_act_model_layers_51_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_20_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_69_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_33_residual_post_attn/max":134,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_53_residual_post_mlp/max_abs":218,"train/tensor_act_model_layers_74_mlp_up_proj/max":3.390625,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_90_mlp_down_proj/mean":15.0625,"train/tensor_act_model_layers_71_residual_post_mlp/max_abs":290,"train/tensor_act_model_layers_69_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_5_residual_post_mlp/min":23.125,"train/tensor_act_model_layers_93_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_mlp_gate_proj/norm":5399.68359375,"train/tensor_act_model_layers_76_mlp_gate_proj/max":4.03125,"train/tensor_grad_model_layers_73_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_80_mlp_up_proj/range":7.390625,"train/tensor_act_model_layers_56_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_5_mlp_down_proj/std":0.4140625,"train/tensor_act_model_layers_89_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_67_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_69_mlp_up_proj_weight/min":-0.1259765625,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_22_residual_post_mlp/mean":92,"train/tensor_act_model_layers_20_mlp_up_proj/max":3.875,"train/tensor_act_model_layers_43_residual_post_attn/mean":172,"train/tensor_act_model_layers_92_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_8_mlp_down_proj_weight/range":0.24169921875,"train/tensor_act_model_layers_12_mlp/max":4,"train/tensor_act_model_layers_98_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/mean":0.146484375,"train/tensor_param_model_layers_0_mlp_down_proj_weight/min":-0.125,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_57_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_31_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_21_mlp_up_proj_weight/norm":6.34375,"train/tensor_param_model_layers_3_mlp_down_proj_weight/std":0.0308837890625,"train/tensor_act_model_layers_21_residual_post_attn/max_abs":86,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_54_mlp_down_proj/range":2.0625,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_52_mlp_up_proj/min":-3.671875,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_72_mlp/max_abs":4,"train/tensor_act_model_layers_17_mlp_gate_proj/min":-3.546875,"train/tensor_act_model_layers_72_residual_post_mlp/min":288,"train/tensor_grad_model_layers_82_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_59_mlp_down_proj/max_abs":16.25,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_61_mlp_up_proj_weight/max_abs":0.12890625,"train/tensor_param_model_layers_4_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_51_mlp_up_proj/mean":0.062255859375,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_91_mlp_down_proj/mean":15,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_18_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_98_mlp_down_proj_weight/range":0.298828125,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_30_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_12_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_96_residual_pre_attn/range":6,"train/tensor_act_model_layers_43_mlp_down_proj/norm":30423.771484375,"train/tensor_param_model_layers_77_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_81_residual_pre_attn/mean":324,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_42_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_38_residual_post_attn/max_abs":154,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp/max_abs":4,"train/tensor_param_model_layers_90_mlp_down_proj_weight/norm":7.96875,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_89_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_18_mlp/std":0,"train/tensor_act_model_layers_45_mlp/max_abs":4,"train/tensor_act_model_layers_87_residual_post_mlp/max":354,"train/tensor_act_model_layers_11_residual_pre_attn/max":45.5,"train/tensor_act_model_layers_96_mlp_up_proj/max":4.40625,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_70_mlp_down_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_29_residual_post_mlp/std":0.67578125,"train/tensor_act_model_layers_88_mlp_down_proj/max_abs":17.25,"train/tensor_act_model_layers_88_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/norm":7.3125,"train_runtime":353.9153,"train/tensor_param_model_layers_56_mlp_down_proj_weight/min":-0.1298828125,"train/tensor_act_model_layers_33_mlp_up_proj/min":-3.84375,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_38_mlp_up_proj/max_abs":3.703125,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_86_mlp_down_proj/max_abs":17.75,"train/tensor_act_model_layers_8_mlp_gate_proj/norm":5385.3369140625,"train/tensor_act_model_layers_22_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_41_mlp_down_proj_weight/norm":6.1875,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_8_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_18_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_34_mlp_down_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_14_mlp_down_proj_weight/max_abs":0.12060546875,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/mean":-0.0002765655517578125,"train/tensor_act_model_layers_21_mlp_up_proj/mean":0.0194091796875,"train/tensor_act_model_layers_54_mlp_up_proj/max":3.515625,"train/tensor_act_model_layers_64_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_22_residual_post_attn/std":0.58203125,"train/tensor_act_model_layers_23_residual_pre_attn/std":0.58203125,"train/tensor_act_model_layers_99_residual_pre_attn/mean":396,"train/tensor_param_model_layers_97_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_residual_post_attn/max_abs":366,"train/tensor_act_model_layers_12_mlp_gate_proj/max_abs":3.734375,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/min":-0.1318359375,"train/tensor_act_model_layers_47_mlp_gate_proj/norm":5459.138671875,"train/tensor_act_model_layers_32_mlp_gate_proj/mean":-0.042236328125,"train/tensor_act_model_layers_1_mlp_up_proj/min":-4.1875,"train/tensor_act_model_layers_73_mlp_gate_proj/mean":0.048095703125,"train/tensor_param_model_layers_84_mlp_down_proj_weight/mean":-3.266334533691406e-05,"train/tensor_grad_model_layers_17_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_65_mlp_down_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_23_mlp_down_proj/max_abs":16.75,"train/tensor_act_model_layers_1_mlp_up_proj/mean":-0.2109375,"train/tensor_act_model_layers_98_residual_post_attn/norm":802413.6875,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_66_mlp_up_proj/norm":5253.078125,"train/tensor_param_model_layers_39_mlp_up_proj_weight/mean":-0.00150299072265625,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/std":0.042236328125,"train/tensor_act_model_layers_65_mlp_gate_proj/max":3.8125,"train/tensor_act_model_layers_99_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_15_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_58_mlp_up_proj/min":-4.1875,"train/tensor_param_model_layers_3_mlp_down_proj_weight/max_abs":0.10400390625,"train/tensor_act_model_layers_16_residual_pre_attn/max_abs":65.5,"train/tensor_act_model_layers_12_residual_post_attn/min":46.5,"train/tensor_act_model_layers_84_mlp/std":0,"train/tensor_act_model_layers_32_mlp_up_proj/norm":5423.91943359375,"train/tensor_act_model_layers_88_mlp/max":4,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_44_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_71_residual_post_attn/norm":581230.25,"train/tensor_act_model_layers_12_residual_post_attn/max":49.5,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/mean":0.000274658203125,"train/tensor_act_model_layers_4_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_69_mlp_up_proj/mean":-0.044921875,"train/tensor_act_model_layers_55_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_47_mlp_down_proj/std":0.41796875,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_93_mlp_up_proj_weight/mean":0.00147247314453125,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_84_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_38_mlp_down_proj/mean":14.8125,"train/tensor_param_model_layers_95_mlp_up_proj_weight/range":0.2939453125,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_25_residual_post_attn/norm":204370.59375,"train/tensor_grad_model_layers_15_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_mlp_up_proj/std":1.890625,"train/tensor_act_model_layers_56_mlp_gate_proj/range":7.75,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/norm":5.75,"train/tensor_act_model_layers_55_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp_up_proj/min":-3.359375,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_51_residual_post_mlp/max":210,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_85_mlp/max_abs":4,"train/tensor_param_model_layers_39_mlp_down_proj_weight/range":0.24951171875,"train/tensor_param_model_layers_42_mlp_up_proj_weight/min":-0.12255859375,"train/tensor_act_model_layers_74_residual_pre_attn/mean":296,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_19_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/mean":-0.002044677734375,"train/tensor_act_model_layers_9_residual_post_attn/max_abs":37.5,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_50_mlp_down_proj_weight/norm":6.25,"train/tensor_act_model_layers_5_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_92_residual_post_mlp/max_abs":374,"train/tensor_act_model_layers_46_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_28_mlp_up_proj_weight/range":0.24267578125,"train/tensor_param_model_layers_91_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_2_residual_pre_attn/range":0.9375,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_17_mlp/mean":4,"train/tensor_act_model_layers_76_mlp/norm":8192,"train/tensor_act_model_layers_87_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_19_mlp_down_proj/max":16,"train/tensor_act_model_layers_19_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_50_residual_post_attn/mean":200,"train/tensor_act_model_layers_14_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_17_residual_post_attn/max":69.5,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_8_mlp_down_proj_weight/std":0.03369140625,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/mean":-0.00135040283203125,"train/tensor_act_model_layers_27_mlp_gate_proj/min":-3.515625,"train/tensor_act_model_layers_13_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_71_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_73_residual_pre_attn/max":294,"train/tensor_act_model_layers_67_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp_down_proj/min":14.125,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_param_model_layers_33_mlp_up_proj_weight/mean":-0.00122833251953125,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_9_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_29_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_45_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_residual_post_attn/max_abs":174,"train/tensor_act_model_layers_16_residual_post_attn/range":3.25,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_13_mlp_up_proj_weight/norm":6.1875,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/max":215,"train/tensor_param_model_layers_40_mlp_gate_proj_weight/range":0.24609375,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_mlp_gate_proj/std":1.8671875,"train/tensor_act_model_layers_85_residual_post_mlp/max_abs":346,"train/tensor_act_model_layers_61_residual_post_attn/mean":244,"train/tensor_param_model_layers_58_mlp_up_proj_weight/mean":-0.000530242919921875,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/norm":6.25,"train/tensor_param_model_layers_57_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_residual_post_attn/mean":136,"train/tensor_act_model_layers_83_residual_pre_attn/range":6,"train/tensor_act_model_layers_18_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_11_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_54_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_77_mlp_down_proj_weight/min":-0.13671875,"train/tensor_act_model_layers_61_residual_post_attn/std":0.8359375,"train/tensor_param_model_layers_6_mlp_up_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_54_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_5_residual_post_attn/min":19.125,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_86_mlp_down_proj/max":17.75,"train/tensor_act_model_layers_83_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_77_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_mlp_down_proj/mean":15,"train/tensor_act_model_layers_19_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_85_mlp/norm":8192,"train/tensor_act_model_layers_97_residual_pre_attn/max_abs":390,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/range":0.2578125,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_5_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_68_mlp_up_proj_weight/range":0.2578125,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_44_residual_pre_attn/max":178,"train/tensor_param_model_layers_77_mlp_up_proj_weight/max_abs":0.134765625,"train/tensor_act_model_layers_17_residual_post_attn/min":66,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_37_mlp_down_proj/norm":30498.689453125,"train/tensor_act_model_layers_56_residual_post_attn/std":0.8203125,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/std":76.5,"train/tensor_act_model_layers_96_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_34_mlp_gate_proj_weight/max_abs":0.1240234375,"train/tensor_param_model_layers_1_mlp_down_proj_weight/max":0.1162109375,"train/tensor_param_model_layers_14_mlp_up_proj_weight/range":0.24560546875,"train/tensor_act_model_layers_74_mlp_gate_proj/std":1.8359375,"train/tensor_param_model_layers_77_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_68_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_86_mlp/std":0,"train/tensor_act_model_layers_78_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_50_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_64_mlp_down_proj_weight/mean":-0.0009307861328125,"train/tensor_act_model_layers_77_residual_post_mlp/min":308,"train/tensor_act_model_layers_19_residual_post_attn/range":4,"train/tensor_grad_model_layers_13_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_up_proj_weight/max":0.1298828125,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_5_residual_post_mlp/max":24.875,"train/tensor_param_model_layers_83_mlp_up_proj_weight/max":0.1337890625,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_52_residual_pre_attn/max":210,"train/tensor_act_model_layers_84_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_90_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_51_residual_pre_attn/mean":204,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/min":-0.12060546875,"train/tensor_act_model_layers_41_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_68_mlp_up_proj/max_abs":4.09375,"train/tensor_act_model_layers_77_residual_post_attn/max_abs":310,"train/tensor_act_model_layers_87_mlp_gate_proj/range":8.03125,"train/tensor_param_model_layers_45_mlp_down_proj_weight/max":0.1240234375,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/mean":0.00034332275390625,"train/tensor_act_model_layers_64_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_35_mlp_down_proj_weight/max_abs":0.1279296875,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_21_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_42_mlp/max":4,"train/tensor_grad_model_layers_21_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_66_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_95_mlp_up_proj/std":1.859375,"train/tensor_act_model_layers_65_mlp_up_proj/range":7.140625,"train/tensor_act_model_layers_79_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_56_mlp/mean":4,"train/tensor_act_model_layers_85_residual_post_attn/norm":695917.875,"train/tensor_act_model_layers_83_mlp_gate_proj/range":8.40625,"train/tensor_act_model_layers_91_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_42_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/max":0.12890625,"train/tensor_act_model_layers_19_mlp_up_proj/max":3.625,"train/tensor_param_model_layers_40_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_27_residual_post_mlp/max_abs":114,"train/tensor_param_model_layers_39_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_12_mlp/std":0,"train/tensor_act_model_layers_79_residual_post_attn/mean":316,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_16_residual_post_attn/max_abs":65.5,"train/tensor_act_model_layers_16_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_1_mlp/max_abs":4,"train/tensor_act_model_layers_33_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_22_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_31_mlp/min":4,"train/tensor_param_model_layers_74_mlp_down_proj_weight/mean":-0.0011444091796875,"train/tensor_act_model_layers_0_mlp/range":0.125,"train/tensor_act_model_layers_53_residual_pre_attn/mean":212,"train/tensor_act_model_layers_61_mlp/min":4,"train/tensor_grad_model_layers_85_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_10_mlp_up_proj_weight/min":-0.125,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/max_abs":0.1494140625,"train/tensor_param_model_layers_26_mlp_up_proj_weight/mean":0.0003833770751953125,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_41_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_78_mlp/min":4,"train/tensor_act_model_layers_14_residual_post_mlp/max_abs":61.5,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_5_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_2_mlp_down_proj/min":14.125,"train/tensor_grad_model_layers_92_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_mlp_up_proj/std":1.8359375,"train/tensor_act_model_layers_69_residual_pre_attn/min":272,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_71_mlp_up_proj_weight/max":0.1357421875,"train/tensor_param_model_layers_45_mlp_up_proj_weight/max":0.125,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_50_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_10_mlp/norm":8192,"train/tensor_act_model_layers_93_residual_post_attn/mean":372,"train/tensor_act_model_layers_68_residual_post_attn/min":268,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/range":0.2548828125,"train/tensor_act_model_layers_56_residual_post_mlp/mean":228,"train/tensor_act_model_layers_78_residual_pre_attn/norm":638574.0625,"train/tensor_param_model_layers_32_mlp_down_proj_weight/max":0.12890625,"train/tensor_act_model_layers_98_mlp_down_proj/std":0.62890625,"train/tensor_act_model_layers_57_mlp/std":0,"train/tensor_param_model_layers_88_mlp_down_proj_weight/range":0.28125,"train/tensor_act_model_layers_39_mlp_up_proj/min":-3.578125,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_51_mlp_down_proj/max":15.9375,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_9_mlp_down_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_72_mlp_down_proj_weight/max":0.1240234375,"train/tensor_act_model_layers_93_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_85_mlp_down_proj_weight/range":0.275390625,"train/tensor_act_model_layers_5_residual_post_attn/norm":40940.84375,"train/tensor_act_model_layers_61_residual_post_mlp/mean":248,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp/max":4,"train/tensor_act_model_layers_20_residual_pre_attn/mean":80,"train/tensor_act_model_layers_41_residual_post_attn/range":5,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_7_residual_post_attn/mean":28,"train/tensor_grad_model_layers_17_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_57_mlp_up_proj/min":-3.734375,"train/tensor_act_model_layers_20_mlp_down_proj/max_abs":16.625,"train/tensor_act_model_layers_86_residual_post_attn/max_abs":346,"train/tensor_param_model_layers_6_mlp_down_proj_weight/mean":0.00066375732421875,"train/tensor_act_model_layers_71_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_41_mlp/max":4,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_96_mlp_down_proj/min":14,"train/tensor_act_model_layers_48_mlp_gate_proj/max":3.90625,"train/tensor_grad_model_layers_39_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/max":0.1044921875,"train/tensor_act_model_layers_92_residual_pre_attn/min":364,"train/tensor_act_model_layers_32_residual_pre_attn/max":130,"train/tensor_act_model_layers_43_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/std":0.0341796875,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_57_residual_post_attn/mean":228,"train/tensor_param_model_layers_2_mlp_down_proj_weight/range":0.19921875,"train/tensor_param_model_layers_44_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_79_residual_pre_attn/norm":646766,"train/tensor_act_model_layers_33_residual_pre_attn/max_abs":134,"train/tensor_act_model_layers_5_residual_pre_attn/min":19.25,"train/tensor_act_model_layers_52_mlp_down_proj/norm":30482.8515625,"train/tensor_act_model_layers_25_mlp/range":0,"train/tensor_act_model_layers_96_mlp_down_proj/max":17.25,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_88_residual_pre_attn/mean":352,"train/tensor_act_model_layers_38_mlp/mean":4,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/norm":6.375,"train/tensor_act_model_layers_56_residual_post_mlp/max_abs":230,"train/tensor_act_model_layers_53_mlp_down_proj/max_abs":16.25,"train/tensor_act_model_layers_46_mlp_up_proj/std":1.859375,"train/tensor_act_model_layers_43_mlp_down_proj/max":16.25,"train/tensor_param_model_layers_20_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_48_mlp_gate_proj_weight/mean":6.723403930664062e-05,"train/tensor_grad_model_layers_67_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_53_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_74_mlp_up_proj_weight/max_abs":0.1318359375,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_84_residual_pre_attn/max_abs":338,"train/tensor_act_model_layers_77_residual_post_attn/mean":308,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/max":0.1142578125,"train/tensor_act_model_layers_13_mlp_gate_proj/max_abs":3.71875,"train/tensor_act_model_layers_78_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_89_mlp_down_proj/range":3,"train/tensor_act_model_layers_65_residual_pre_attn/mean":260,"train/tensor_act_model_layers_23_mlp_up_proj/mean":0.265625,"train/tensor_grad_model_layers_49_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_64_mlp/max_abs":4,"train/tensor_act_model_layers_22_residual_post_mlp/norm":188059.078125,"train/tensor_act_model_layers_4_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_95_mlp_down_proj_weight/std":0.047119140625,"train/tensor_param_model_layers_64_mlp_up_proj_weight/norm":6.59375,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_80_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_17_mlp_up_proj/max":3.703125,"train/tensor_act_model_layers_51_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_68_residual_post_attn/max":274,"train/tensor_param_model_layers_38_mlp_up_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_25_mlp/min":4,"train/tensor_act_model_layers_26_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_2_mlp_down_proj_weight/mean":-0.0010528564453125,"train/tensor_act_model_layers_87_residual_post_attn/max_abs":350,"train/tensor_act_model_layers_81_residual_post_attn/max_abs":326,"train/tensor_act_model_layers_18_residual_post_mlp/max_abs":78,"train/tensor_act_model_layers_11_mlp_up_proj/std":1.8828125,"train/tensor_act_model_layers_61_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_31_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_79_mlp_gate_proj_weight/std":0.0400390625,"train/tensor_grad_model_layers_78_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_65_mlp_up_proj_weight/max_abs":0.1337890625,"train/tensor_act_model_layers_54_mlp_up_proj/mean":0.193359375,"train/tensor_act_model_layers_52_residual_post_attn/max":210,"train/tensor_param_model_layers_16_mlp_up_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_6_residual_pre_attn/std":0.23046875,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_25_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_33_mlp_up_proj/range":7.390625,"train/tensor_grad_model_layers_51_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_82_residual_pre_attn/range":6,"train/tensor_act_model_layers_30_mlp_gate_proj/norm":5308.6162109375,"train/tensor_act_model_layers_54_residual_post_attn/norm":441829.125,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_residual_post_mlp/range":5,"train/tensor_act_model_layers_20_residual_pre_attn/min":78,"train/tensor_act_model_layers_0_residual_pre_attn/max":0.271484375,"train/tensor_param_model_layers_9_mlp_up_proj_weight/std":0.033935546875,"train/tensor_grad_model_layers_18_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_46_mlp_down_proj_weight/mean":-0.0003986358642578125,"train/tensor_act_model_layers_61_mlp/norm":8192,"train/tensor_act_model_layers_84_residual_post_attn/min":332,"train/tensor_act_model_layers_56_residual_post_mlp/max":230,"train/tensor_param_model_layers_6_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/mean":0.001708984375,"train/tensor_act_model_layers_81_residual_post_mlp/range":6,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_97_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_29_mlp_down_proj/min":14.25,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_47_mlp_down_proj_weight/mean":-0.0013275146484375,"train/tensor_param_model_layers_24_mlp_down_proj_weight/max":0.119140625,"train/tensor_act_model_layers_28_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_residual_post_mlp/min":106,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_55_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_69_mlp/std":0,"train/tensor_act_model_layers_79_residual_post_mlp/mean":320,"train/tensor_act_model_layers_48_residual_post_mlp/min":193,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_47_residual_post_mlp/max_abs":194,"train/tensor_act_model_layers_63_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_residual_pre_attn/mean":92,"train/tensor_act_model_layers_91_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_7_mlp_down_proj_weight/min":-0.1220703125,"train/tensor_act_model_layers_16_mlp/mean":4,"train/tensor_act_model_layers_6_mlp_down_proj/max":16.375,"train/tensor_act_model_layers_79_mlp_gate_proj/max_abs":4.03125,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/mean":0.00067138671875,"train/tensor_act_model_layers_97_mlp_down_proj/max_abs":18,"train/tensor_param_model_layers_93_mlp_down_proj_weight/range":0.2802734375,"train/tensor_grad_model_layers_72_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_24_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_mlp_down_proj/max":16.375,"train/tensor_act_model_layers_1_mlp_down_proj/min":15.0625,"train/tensor_act_model_layers_61_residual_pre_attn/norm":499204.84375,"train/tensor_act_model_layers_77_mlp/norm":8192,"train/tensor_act_model_layers_94_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_96_mlp_down_proj/range":3.25,"train/tensor_act_model_layers_84_residual_post_attn/max":338,"train/tensor_act_model_layers_19_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_82_mlp_down_proj_weight/max_abs":0.1328125,"train/tensor_param_model_layers_91_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/std":0.0341796875,"train/tensor_param_model_layers_78_mlp_up_proj_weight/norm":7.125,"train/tensor_act_model_layers_97_residual_post_attn/mean":388,"train/tensor_act_model_layers_20_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_37_mlp_gate_proj/range":7.3125,"train/tensor_param_model_layers_69_mlp_gate_proj_weight/max":0.12890625,"train/tensor_act_model_layers_31_mlp_down_proj/max":15.875,"train/tensor_grad_model_layers_58_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_80_mlp_gate_proj/max":3.90625,"train/tensor_act_model_layers_26_mlp/std":0,"train/tensor_act_model_layers_78_residual_post_mlp/mean":316,"train/tensor_param_model_layers_67_mlp_down_proj_weight/max":0.126953125,"train/tensor_act_model_layers_52_residual_post_attn/min":205,"train/tensor_act_model_layers_56_residual_pre_attn/norm":458213.03125,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/mean":0.00026702880859375,"train/tensor_act_model_layers_29_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_76_mlp_up_proj/range":7.4375,"train/tensor_param_model_layers_90_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_74_mlp/norm":8192,"train/tensor_act_model_layers_81_mlp_gate_proj/range":7.8125,"train/tensor_act_model_layers_22_residual_pre_attn/norm":179867.25,"train/tensor_param_model_layers_14_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_46_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_92_residual_pre_attn/mean":368,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_49_mlp/std":0,"train/tensor_act_model_layers_29_mlp_gate_proj/max_abs":3.46875,"train/tensor_act_model_layers_64_mlp_up_proj/max":3.515625,"train/tensor_act_model_layers_14_mlp_up_proj/mean":0.03515625,"train/tensor_param_model_layers_15_mlp_up_proj_weight/min":-0.125,"train/tensor_act_model_layers_95_mlp_gate_proj/max_abs":4.5625,"train/tensor_act_model_layers_36_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_11_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/max":0.1279296875,"train/tensor_act_model_layers_71_residual_post_attn/min":280,"train/tensor_act_model_layers_44_mlp/std":0,"train/tensor_act_model_layers_69_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_17_mlp_up_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_82_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_mlp_up_proj/range":7.765625,"train/tensor_act_model_layers_76_mlp_up_proj/norm":5280.08544921875,"train/tensor_param_model_layers_18_mlp_down_proj_weight/mean":-0.00048828125,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_95_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_87_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_residual_pre_attn/max":138,"train/tensor_param_model_layers_79_mlp_down_proj_weight/mean":-0.00064849853515625,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/norm":6.15625,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_90_mlp/max_abs":4,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/norm":6.40625,"train/tensor_act_model_layers_23_mlp/min":4,"train/tensor_act_model_layers_69_residual_post_mlp/range":6,"train/tensor_act_model_layers_80_mlp_gate_proj/mean":-0.076171875,"train/tensor_act_model_layers_31_mlp_gate_proj/mean":-0.154296875,"train/tensor_param_model_layers_27_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_96_mlp_up_proj_weight/max_abs":0.15625,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_34_mlp_up_proj_weight/max_abs":0.1298828125,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/mean":0.0003414154052734375,"train/tensor_act_model_layers_57_residual_post_attn/max_abs":230,"train/tensor_act_model_layers_30_mlp/max_abs":4,"train/tensor_act_model_layers_82_mlp_gate_proj/norm":5263.25927734375,"train/tensor_act_model_layers_64_mlp_gate_proj/min":-4.03125,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_14_mlp_down_proj/norm":30368.904296875,"train/tensor_act_model_layers_39_mlp/min":4,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_21_residual_post_attn/range":4,"train/tensor_param_model_layers_83_mlp_down_proj_weight/norm":7.59375,"train/tensor_act_model_layers_71_residual_post_attn/max":286,"train/tensor_act_model_layers_79_residual_post_mlp/norm":654958,"train/tensor_act_model_layers_86_mlp/min":4,"train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit":0.9999692440032959,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_10_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/max_abs":0.1494140625,"train/tensor_grad_model_layers_41_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_27_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_30_mlp_up_proj_weight/mean":-0.0004863739013671875,"train/tensor_act_model_layers_20_residual_post_attn/max_abs":82,"train/tensor_act_model_layers_78_residual_post_mlp/range":6,"train/tensor_act_model_layers_41_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_80_residual_post_attn/mean":320,"train/tensor_act_model_layers_73_residual_pre_attn/max_abs":294,"train/tensor_act_model_layers_55_mlp_up_proj/max_abs":3.953125,"train/tensor_act_model_layers_78_residual_post_mlp/norm":646766,"train/tensor_act_model_layers_3_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_mlp_gate_proj/min":-3.453125,"train/tensor_act_model_layers_51_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_7_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_25_mlp/std":0,"train/tensor_act_model_layers_44_mlp_down_proj/range":1.5,"train/tensor_param_model_layers_91_mlp_down_proj_weight/range":0.2958984375,"train/tensor_param_model_layers_55_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_30_residual_pre_attn/std":0.67578125,"train/tensor_act_model_layers_92_mlp_down_proj/mean":14.9375,"train/tensor_grad_model_layers_73_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_7_mlp_gate_proj/mean":-0.032470703125,"train/tensor_param_model_layers_49_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_48_mlp_up_proj/std":1.84375,"train/tensor_act_model_layers_30_mlp_gate_proj/range":7.5,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/norm":6.375,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/range":0.25,"train/tensor_act_model_layers_55_mlp_gate_proj/max_abs":3.78125,"train/tensor_param_model_layers_73_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_down_proj/max_abs":16,"train/tensor_param_model_layers_85_mlp_up_proj_weight/max_abs":0.142578125,"train/tensor_act_model_layers_68_mlp_up_proj/norm":5317.88525390625,"train/tensor_param_model_layers_98_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_25_mlp_down_proj_weight/min":-0.11328125,"train/tensor_param_model_layers_89_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_85_mlp_up_proj_weight/max":0.1328125,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_25_residual_post_mlp/min":102,"train/tensor_act_model_layers_33_mlp_gate_proj/max":3.59375,"train/tensor_act_model_layers_31_mlp_up_proj/mean":0.1884765625,"train/tensor_act_model_layers_29_mlp_down_proj/norm":30446.552734375,"train/tensor_param_model_layers_16_mlp_down_proj_weight/min":-0.11962890625,"train/tensor_param_model_layers_53_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_47_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_59_mlp_down_proj_weight/mean":0.000858306884765625,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/mean":0.00115966796875,"train/tensor_param_model_layers_0_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_84_mlp_up_proj_weight/std":0.042236328125,"train/tensor_act_model_layers_20_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_20_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_88_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_41_mlp/range":0,"train/tensor_param_model_layers_11_mlp_up_proj_weight/min":-0.12060546875,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_45_mlp_down_proj_weight/range":0.2470703125,"train/tensor_param_model_layers_52_mlp_down_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_12_residual_post_mlp/norm":106329.6875,"train/tensor_act_model_layers_79_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_12_residual_pre_attn/max":49.5,"train/tensor_param_model_layers_57_mlp_up_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_99_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_66_mlp_up_proj_weight/std":0.036376953125,"train/tensor_act_model_layers_28_residual_pre_attn/min":110,"train/tensor_param_model_layers_58_mlp_up_proj_weight/std":0.035400390625,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_74_mlp_down_proj_weight/max":0.1435546875,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_52_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_41_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_param_model_layers_37_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_86_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_0_mlp_down_proj/norm":83135.953125,"train/tensor_param_model_layers_75_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_mlp_up_proj/min":-3.46875,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_31_mlp/range":0,"train/tensor_act_model_layers_56_residual_pre_attn/range":5,"train/tensor_act_model_layers_75_residual_post_attn/norm":613998.125,"train/tensor_act_model_layers_4_residual_post_mlp/max_abs":20.625,"train/tensor_act_model_layers_10_mlp_gate_proj/norm":5310.103515625,"train/tensor_act_model_layers_43_residual_post_attn/range":5,"train/tensor_act_model_layers_94_residual_post_attn/norm":769645.75,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_62_mlp_down_proj_weight/std":0.03515625,"train/tensor_grad_model_layers_52_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_69_mlp/max":4,"train/tensor_act_model_layers_6_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_60_residual_pre_attn/mean":240,"train/tensor_act_model_layers_75_mlp_gate_proj/range":8.0625,"train/tensor_act_model_layers_13_mlp_up_proj/norm":5338.8564453125,"train/tensor_act_model_layers_51_residual_post_mlp/min":205,"train/tensor_act_model_layers_32_mlp_up_proj/std":1.875,"train/tensor_grad_model_layers_74_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_mlp_up_proj/max_abs":4.03125,"train/tensor_act_model_layers_76_mlp/max_abs":4,"train/tensor_act_model_layers_82_mlp_down_proj/mean":15,"train/tensor_act_model_layers_50_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_27_residual_post_attn/range":4,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/mean":0.00154876708984375,"train/tensor_act_model_layers_70_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_79_mlp/std":0,"train/tensor_act_model_layers_85_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_92_mlp_gate_proj/norm":5318.26318359375,"train/tensor_param_model_layers_68_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_13_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_49_residual_post_mlp/range":5,"train/tensor_act_model_layers_86_residual_post_mlp/range":6,"train/tensor_act_model_layers_17_residual_post_mlp/mean":72,"train/tensor_act_model_layers_44_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_67_mlp_gate_proj/mean":0.11865234375,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_67_mlp_gate_proj/range":7.6875,"train/tensor_act_model_layers_94_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_68_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_25_mlp_down_proj_weight/range":0.2392578125,"train/tensor_act_model_layers_36_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_82_mlp_up_proj/norm":5410.17626953125,"train/tensor_act_model_layers_5_mlp_up_proj/mean":0.0947265625,"train/tensor_act_model_layers_53_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_78_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_mlp_down_proj/std":0.3984375,"train/tensor_act_model_layers_13_mlp_gate_proj/range":7.390625,"train/tensor_param_model_layers_77_mlp_up_proj_weight/max":0.130859375,"train/tensor_act_model_layers_56_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_61_residual_pre_attn/min":241,"train/tensor_act_model_layers_53_mlp_gate_proj/range":7.953125,"train/tensor_act_model_layers_78_residual_post_attn/max":314,"train/tensor_param_model_layers_6_mlp_down_proj_weight/norm":5.84375,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_3_residual_post_mlp/range":1.125,"train/tensor_param_model_layers_57_mlp_up_proj_weight/mean":-0.00016498565673828125,"train/tensor_act_model_layers_45_mlp_down_proj/std":0.39453125,"train/tensor_act_model_layers_5_mlp_up_proj/std":1.8046875,"train/tensor_act_model_layers_79_mlp/mean":4,"train/tensor_act_model_layers_5_mlp_down_proj/max":16.125,"train/tensor_act_model_layers_10_mlp/min":4,"train/tensor_act_model_layers_25_mlp_gate_proj/min":-3.6875,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/max":0.138671875,"train/tensor_act_model_layers_47_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_47_mlp_down_proj/range":2.1875,"train/tensor_param_model_layers_40_mlp_down_proj_weight/mean":0.000514984130859375,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_95_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_98_mlp_up_proj_weight/mean":0.0008697509765625,"train/tensor_act_model_layers_40_residual_pre_attn/std":0.80859375,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_36_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_34_mlp_gate_proj/min":-3.484375,"train/tensor_grad_model_layers_91_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_88_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_70_residual_post_attn/mean":280,"train/tensor_act_model_layers_16_mlp_down_proj/max":16,"train/tensor_act_model_layers_17_residual_pre_attn/max":69.5,"train/tensor_act_model_layers_84_residual_post_mlp/mean":340,"train/tensor_act_model_layers_34_mlp_up_proj/norm":5255.57177734375,"train/tensor_param_model_layers_78_mlp_up_proj_weight/range":0.2626953125,"train/tensor_act_model_layers_78_mlp_gate_proj/std":1.8203125,"train/tensor_act_model_layers_6_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/max_abs":0.1337890625,"train/tensor_act_model_layers_72_mlp/std":0,"train/tensor_grad_model_layers_80_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_mlp_up_proj/min":-4.0625,"train/tensor_act_model_layers_71_residual_post_mlp/norm":589422.1875,"train/tensor_act_model_layers_72_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_12_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_7_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_6_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_79_residual_post_attn/max":318,"train/tensor_act_model_layers_1_mlp_down_proj/norm":38212.4375,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/mean":-0.0005950927734375,"train/tensor_act_model_layers_10_residual_pre_attn/max":41.5,"train/tensor_act_model_layers_79_residual_post_attn/min":312,"train/tensor_act_model_layers_1_mlp/mean":4,"train/tensor_act_model_layers_59_residual_post_attn/mean":236,"train/tensor_act_model_layers_6_mlp/max":4,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_50_mlp/max":4,"train/tensor_act_model_layers_35_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_mlp_gate_proj/norm":5360.43994140625,"train/tensor_act_model_layers_86_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_down_proj_weight/mean":-0.00154876708984375,"train/tensor_act_model_layers_15_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_mlp/max_abs":37.5,"train/tensor_act_model_layers_70_residual_post_mlp/max_abs":286,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_30_mlp_down_proj_weight/max":0.12158203125,"train/tensor_act_model_layers_65_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_40_mlp/max":4,"train/tensor_param_model_layers_35_mlp_up_proj_weight/min":-0.12890625,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_2_mlp_down_proj_weight/norm":5.59375,"train/tensor_act_model_layers_78_mlp/mean":4,"train/tensor_grad_model_layers_80_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_17_mlp_up_proj_weight/mean":-0.0004825592041015625,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_16_mlp/max":4,"train/tensor_grad_model_layers_72_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_32_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_38_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_35_mlp_gate_proj/min":-3.640625,"train/tensor_act_model_layers_16_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_18_residual_pre_attn/min":70,"train/tensor_grad_model_layers_36_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_28_mlp_up_proj/max_abs":3.515625,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_24_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_35_mlp/max_abs":4,"train/tensor_act_model_layers_95_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_10_mlp_up_proj/min":-3.265625,"train/tensor_param_model_layers_24_mlp_up_proj_weight/std":0.03515625,"train/tensor_act_model_layers_68_residual_post_mlp/range":6,"train/tensor_act_model_layers_78_residual_post_attn/min":308,"train/tensor_act_model_layers_93_residual_post_attn/range":6,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/min":-0.109375,"train/tensor_grad_model_layers_72_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_62_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_residual_post_mlp/mean":212,"train/tensor_act_model_layers_68_residual_pre_attn/norm":556654.3125,"train/tensor_param_model_layers_9_mlp_down_proj_weight/std":0.033447265625,"train/tensor_act_model_layers_38_residual_post_mlp/range":5,"train/tensor_act_model_layers_15_residual_pre_attn/range":3,"train/tensor_act_model_layers_67_mlp/mean":4,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/norm":7.0625,"train/tensor_act_model_layers_68_residual_post_mlp/mean":276,"train/tensor_act_model_layers_52_residual_pre_attn/norm":425461.25,"train/tensor_act_model_layers_56_residual_post_attn/min":221,"train/tensor_act_model_layers_69_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_1_mlp_down_proj/max_abs":23.375,"train/tensor_act_model_layers_34_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_79_mlp_gate_proj/range":8.03125,"train/tensor_grad_model_layers_26_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_97_mlp_up_proj/std":1.8359375,"train/tensor_act_model_layers_23_mlp_up_proj/std":1.84375,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_residual_pre_attn/max":154,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_65_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_17_residual_post_attn/std":0.5625,"train/tensor_param_model_layers_67_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_40_residual_post_attn/mean":160,"train/tensor_grad_model_layers_11_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_61_residual_post_attn/min":241,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/mean":-0.0004062652587890625,"train/tensor_act_model_layers_53_mlp_gate_proj/mean":-0.00031280517578125,"train/tensor_act_model_layers_25_residual_post_mlp/max":106,"train/tensor_act_model_layers_42_mlp/norm":8192,"train/tensor_act_model_layers_12_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_33_residual_post_mlp/norm":277973.8125,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_30_mlp_up_proj/range":7.25,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_mlp_gate_proj/norm":5447.4326171875,"train/tensor_act_model_layers_18_mlp_gate_proj/norm":5307.703125,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/mean":-0.000579833984375,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_87_mlp_up_proj_weight/range":0.279296875,"train/tensor_act_model_layers_63_mlp_gate_proj/mean":-0.1455078125,"train/tensor_act_model_layers_25_residual_pre_attn/mean":100,"train/tensor_grad_model_layers_61_mlp_up_proj_weight/range":0,"_wandb":{"runtime":369},"train/tensor_param_model_layers_78_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_34_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_36_residual_post_attn/norm":294374.625,"train/tensor_act_model_layers_59_mlp_down_proj/norm":30496.927734375,"train/tensor_param_model_layers_12_mlp_down_proj_weight/max":0.125,"train/tensor_param_model_layers_55_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_55_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_30_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_95_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_93_mlp_gate_proj/max":4.34375,"train/tensor_act_model_layers_61_residual_post_attn/max":246,"train/tensor_grad_model_layers_10_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_84_residual_pre_attn/norm":687725.9375,"train/tensor_act_model_layers_47_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_92_mlp_down_proj/range":2.875,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_mlp_gate_proj/norm":5320.83251953125,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_77_mlp_up_proj/max":4,"train/tensor_param_model_layers_73_mlp_down_proj_weight/min":-0.134765625,"train/tensor_act_model_layers_96_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_61_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_17_mlp_down_proj/norm":30383.896484375,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/max_abs":0.1298828125,"train/tensor_act_model_layers_0_residual_post_attn/max_abs":0.373046875,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_78_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_56_residual_post_attn/max":226,"train/tensor_act_model_layers_88_residual_pre_attn/range":6,"train/tensor_param_model_layers_52_mlp_up_proj_weight/mean":0.0009765625,"train/tensor_act_model_layers_31_mlp_up_proj/std":1.828125,"train/tensor_param_model_layers_76_mlp_up_proj_weight/min":-0.12890625,"train/tensor_grad_model_layers_1_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_77_mlp_gate_proj/max_abs":3.890625,"train/tensor_act_model_layers_22_mlp_gate_proj/norm":5424.94775390625,"train/tensor_act_model_layers_77_mlp_up_proj/range":8.125,"train/tensor_act_model_layers_22_residual_post_attn/max":90,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_50_mlp_up_proj/max":3.6875,"train/tensor_act_model_layers_23_mlp_gate_proj/min":-3.609375,"train/tensor_act_model_layers_69_mlp/range":0,"train/tensor_param_model_layers_60_mlp_down_proj_weight/mean":0.0002689361572265625,"train/tensor_param_model_layers_96_mlp_up_proj_weight/mean":-0.0004711151123046875,"train/tensor_grad_model_layers_89_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_54_mlp/mean":4,"train/tensor_param_model_layers_69_mlp_down_proj_weight/mean":-0.00012874603271484375,"train/tensor_act_model_layers_68_residual_post_mlp/norm":564846.3125,"train/tensor_act_model_layers_84_residual_post_attn/range":6,"train/tensor_act_model_layers_72_mlp_gate_proj/min":-4.03125,"train/tensor_param_model_layers_26_mlp_down_proj_weight/norm":6.34375,"train/tensor_param_model_layers_42_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_78_mlp/max":4,"train/tensor_act_model_layers_36_mlp_down_proj/range":2.375,"train/tensor_act_model_layers_85_mlp_down_proj/range":3.375,"train/tensor_act_model_layers_6_mlp_down_proj/std":0.42578125,"train/tensor_act_model_layers_59_mlp_gate_proj/std":1.8359375,"train/tensor_param_model_layers_88_mlp_up_proj_weight/min":-0.146484375,"train/tensor_act_model_layers_15_mlp_down_proj/std":0.369140625,"train/tensor_act_model_layers_33_residual_post_mlp/mean":136,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_1_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_27_mlp/min":4,"train/tensor_act_model_layers_35_mlp/range":0,"train/tensor_act_model_layers_31_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_81_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_24_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_10_residual_pre_attn/range":3,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp/range":0,"train/tensor_act_model_layers_27_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_residual_post_attn/norm":49096.33984375,"train/tensor_grad_model_layers_34_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_12_residual_post_attn/std":0.4296875,"train/tensor_param_model_layers_10_mlp_down_proj_weight/norm":6.03125,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_82_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/min":-0.1298828125,"train/tensor_act_model_layers_39_mlp_down_proj/max_abs":16.5,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/max_abs":0.13671875,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/range":0.2529296875,"train/tensor_act_model_layers_79_mlp_up_proj/max":3.578125,"train/tensor_act_model_layers_47_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_26_mlp_down_proj/norm":30536.705078125,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_1_residual_pre_attn/max_abs":4.375,"train/tensor_act_model_layers_43_mlp/mean":4,"train/tensor_act_model_layers_47_mlp/max_abs":4,"train/tensor_act_model_layers_79_mlp/max_abs":4,"train/tensor_act_model_layers_71_residual_pre_attn/norm":581230.25,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_45_residual_post_attn/range":5,"train/tensor_act_model_layers_48_mlp_gate_proj/mean":0.0086669921875,"train/tensor_grad_model_layers_19_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_95_mlp/max":4,"train/tensor_act_model_layers_38_residual_post_attn/range":5,"train/tensor_grad_model_layers_56_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_21_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_56_mlp_down_proj_weight/norm":6.40625,"train/tensor_act_model_layers_22_residual_post_attn/min":86,"train/tensor_act_model_layers_56_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_80_mlp_up_proj_weight/mean":-8.940696716308594e-06,"train/tensor_act_model_layers_88_mlp_gate_proj/mean":-0.07373046875,"train/tensor_act_model_layers_63_mlp_down_proj/std":0.431640625,"train/tensor_act_model_layers_6_mlp_gate_proj/mean":-0.0498046875,"train/tensor_param_model_layers_34_mlp_up_proj_weight/range":0.25146484375,"train/tensor_grad_model_layers_52_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_38_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_56_residual_post_mlp/min":225,"train/tensor_act_model_layers_37_residual_post_mlp/std":0.80859375,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_63_mlp/mean":4,"train/tensor_act_model_layers_75_mlp/min":4,"train/tensor_act_model_layers_32_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_71_mlp_down_proj/max":16,"train/tensor_param_model_layers_23_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_38_mlp_gate_proj/max_abs":3.671875,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_82_mlp_up_proj_weight/std":0.04150390625,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/norm":6.34375,"train/tensor_act_model_layers_26_residual_post_mlp/max":110,"train/tensor_act_model_layers_79_mlp/min":4,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_47_mlp_down_proj/max_abs":16.5,"train/tensor_act_model_layers_30_mlp_gate_proj/max":3.78125,"train/tensor_grad_model_layers_3_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_71_residual_post_mlp/mean":288,"train/tensor_act_model_layers_5_mlp_gate_proj/min":-3.25,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/norm":6.1875,"train/tensor_act_model_layers_74_residual_pre_attn/range":6,"train/tensor_param_model_layers_97_mlp_down_proj_weight/std":0.047119140625,"train/tensor_act_model_layers_23_mlp/max":4,"train/tensor_act_model_layers_87_residual_post_attn/norm":712301.875,"train/tensor_param_model_layers_64_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/norm":8.75,"train/tensor_act_model_layers_54_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_9_residual_post_mlp/max_abs":41.5,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_20_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_96_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_84_residual_post_mlp/norm":695917.875,"train/tensor_act_model_layers_94_mlp_down_proj/range":4.625,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_residual_pre_attn/max":230,"train/tensor_param_model_layers_79_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_52_residual_pre_attn/max_abs":210,"train/tensor_act_model_layers_44_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_pre_attn/max":238,"train/tensor_act_model_layers_11_mlp_gate_proj/norm":5412.6689453125,"train/tensor_act_model_layers_2_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_93_mlp_up_proj/range":8.4375,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_36_mlp_down_proj_weight/norm":6.25,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/range":0.3037109375,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/range":0.2685546875,"train/tensor_act_model_layers_0_mlp/mean":4,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_33_residual_pre_attn/norm":269781.90625,"train/tensor_act_model_layers_73_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_77_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_41_mlp/std":0,"train/tensor_act_model_layers_71_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_48_residual_post_mlp/max":198,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_46_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_26_mlp_up_proj/norm":5422.9189453125,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_77_mlp_gate_proj/max":3.796875,"train/tensor_act_model_layers_13_mlp_down_proj/std":0.50390625,"train/tensor_act_model_layers_10_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_52_residual_post_mlp/norm":433653.1875,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_72_mlp_gate_proj/norm":5279.17529296875,"train/tensor_grad_model_layers_11_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_95_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_63_mlp_up_proj/max_abs":4.09375,"train/tensor_param_model_layers_80_mlp_gate_proj_weight/max":0.134765625,"train/tensor_act_model_layers_69_mlp_gate_proj/range":7.515625,"train/tensor_act_model_layers_15_residual_post_attn/range":3,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/norm":5.6875,"train/tensor_act_model_layers_99_mlp/max_abs":4,"train/tensor_act_model_layers_74_mlp/std":0,"train/tensor_act_model_layers_70_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_36_mlp_gate_proj/min":-3.84375,"train/tensor_param_model_layers_92_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_73_mlp_up_proj_weight/range":0.2548828125,"train/tensor_grad_model_layers_90_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_30_mlp_down_proj/mean":14.8125,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_31_mlp/max_abs":4,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/std":0.03857421875,"train/tensor_act_model_layers_27_residual_pre_attn/mean":108,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_66_residual_post_mlp/range":6,"train/tensor_param_model_layers_50_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_6_residual_pre_attn/mean":24,"train/tensor_param_model_layers_78_mlp_up_proj_weight/min":-0.1298828125,"train/tensor_act_model_layers_56_mlp_gate_proj/max_abs":3.875,"train/tensor_param_model_layers_20_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_67_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_residual_post_attn/max_abs":16.625,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_54_mlp_gate_proj/mean":-0.259765625,"train/tensor_param_model_layers_79_mlp_up_proj_weight/min":-0.1318359375,"train/tensor_grad_model_layers_23_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_55_residual_post_attn/max_abs":222,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_58_mlp_down_proj_weight/norm":6.40625,"train/tensor_act_model_layers_75_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_25_mlp_down_proj/range":2,"train/tensor_act_model_layers_8_mlp/max":4,"train/tensor_act_model_layers_86_residual_post_attn/range":6,"train/tensor_act_model_layers_20_mlp_gate_proj/range":6.90625,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_residual_post_attn/max":25,"train/tensor_act_model_layers_97_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_12_residual_post_mlp/mean":52,"train/tensor_act_model_layers_1_residual_post_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_87_mlp_up_proj_weight/min":-0.138671875,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_79_mlp_down_proj/max_abs":17.375,"train/tensor_param_model_layers_70_mlp_gate_proj_weight/mean":0.0002574920654296875,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_71_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_64_residual_pre_attn/mean":256,"train/tensor_act_model_layers_44_mlp_down_proj/max":15.8125,"train/tensor_param_model_layers_91_mlp_up_proj_weight/max_abs":0.154296875,"train/tensor_act_model_layers_49_mlp_down_proj/norm":30559.908203125,"train/tensor_act_model_layers_78_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_49_residual_pre_attn/std":0.81640625,"train/tensor_act_model_layers_44_mlp_down_proj/min":14.3125,"train/tensor_grad_model_layers_37_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_77_mlp/mean":4,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_48_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_54_residual_post_mlp/std":0.8203125,"train/tensor_param_model_layers_51_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_18_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_96_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_90_mlp/range":0,"train/tensor_act_model_layers_87_mlp_up_proj/max":3.84375,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/norm":6.15625,"train/tensor_act_model_layers_33_residual_post_attn/std":0.71484375,"train/tensor_param_model_layers_65_mlp_up_proj_weight/std":0.035888671875,"train/tensor_act_model_layers_2_residual_pre_attn/max":8.4375,"train/tensor_act_model_layers_37_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_37_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_43_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_6_residual_post_attn/max_abs":25,"train/tensor_param_model_layers_32_mlp_down_proj_weight/range":0.2568359375,"train/tensor_param_model_layers_46_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_5_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_98_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_93_mlp_down_proj_weight/min":-0.14453125,"train/tensor_act_model_layers_84_mlp_gate_proj/mean":-0.05224609375,"train/tensor_param_model_layers_17_mlp_down_proj_weight/mean":-0.000579833984375,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_79_residual_pre_attn/max":318,"train/tensor_act_model_layers_31_residual_pre_attn/max_abs":126.5,"train/tensor_act_model_layers_17_residual_post_mlp/norm":147125.609375,"train/tensor_grad_model_layers_28_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_20_mlp_gate_proj/norm":5378.904296875,"train/tensor_act_model_layers_84_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_40_mlp_up_proj/max_abs":3.625,"train/tensor_param_model_layers_87_mlp_up_proj_weight/mean":0.00167083740234375,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_75_mlp_gate_proj_weight/range":0.275390625,"train/tensor_act_model_layers_18_mlp_down_proj/max_abs":15.875,"train/tensor_act_model_layers_54_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_94_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_mlp/std":0,"train/tensor_act_model_layers_92_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_94_mlp_up_proj_weight/min":-0.1435546875,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_30_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_80_mlp_up_proj_weight/max_abs":0.1376953125,"train/tensor_act_model_layers_9_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_77_residual_post_mlp/max":314,"train/tensor_act_model_layers_64_residual_pre_attn/max":258,"train/tensor_param_model_layers_63_mlp_up_proj_weight/norm":6.5,"train/tensor_grad_model_layers_69_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_12_mlp_up_proj/std":1.8828125,"train/tensor_act_model_layers_10_mlp/mean":4,"train/tensor_act_model_layers_12_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/norm":7.0625,"train/tensor_act_model_layers_10_mlp_down_proj/std":0.400390625,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/range":772,"train/tensor_act_model_layers_12_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_8_mlp/max_abs":4,"train/tensor_act_model_layers_80_residual_pre_attn/min":316,"train/tensor_act_model_layers_71_mlp_down_proj/norm":30588.375,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/norm":6.28125,"train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_63_mlp_gate_proj_weight/std":0.035888671875,"train/tensor_act_model_layers_66_mlp_down_proj/min":14.1875,"train/tensor_grad_model_layers_2_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_95_mlp_gate_proj_weight/min":-0.14453125,"train/tensor_act_model_layers_25_mlp_gate_proj/mean":0.0830078125,"train/tensor_act_model_layers_41_residual_post_mlp/std":0.80859375,"train/tensor_act_model_layers_40_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/std":0.03515625,"train/tensor_param_model_layers_71_mlp_down_proj_weight/norm":6.78125,"train/tensor_act_model_layers_85_residual_pre_attn/max":342,"train/tensor_act_model_layers_22_mlp_down_proj/max_abs":16.5,"train/tensor_act_model_layers_22_mlp/std":0,"train/tensor_act_model_layers_93_mlp/max":4,"train/tensor_act_model_layers_66_mlp_down_proj/max":16.875,"train/tensor_act_model_layers_12_residual_pre_attn/norm":98207.4609375,"train/tensor_act_model_layers_41_mlp/mean":4,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/min":-0.1240234375,"train/tensor_act_model_layers_93_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_71_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_43_mlp_down_proj_weight/max_abs":0.1259765625,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_11_mlp_down_proj/std":0.4296875,"train/tensor_act_model_layers_38_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_76_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_mlp_up_proj/std":1.8671875,"train/tensor_param_model_layers_18_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_26_mlp/max_abs":4,"train/tensor_act_model_layers_24_residual_post_mlp/std":0.61328125,"train/tensor_act_model_layers_88_mlp_down_proj/std":0.4921875,"train/tensor_act_model_layers_2_residual_pre_attn/frac_near_user_limit":0,"train/tensor_act_model_layers_90_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_16_mlp_up_proj_weight/range":0.2509765625,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_86_residual_post_mlp/max_abs":350,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_43_mlp_up_proj_weight/std":0.034423828125,"train/tensor_param_model_layers_97_mlp_up_proj_weight/mean":-0.00128173828125,"train/tensor_act_model_layers_24_mlp_up_proj/min":-3.78125,"train/tensor_act_model_layers_37_residual_post_mlp/max_abs":154,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/mean":-9.918212890625e-05,"train/tensor_grad_model_layers_22_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_91_mlp_up_proj/norm":5309.185546875,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/max_abs":0.12255859375,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_35_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_residual_post_attn/std":0.81640625,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_74_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_37_mlp/range":0,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_3_mlp/norm":8192,"train/tensor_act_model_layers_42_mlp_down_proj/std":0.419921875,"train/tensor_act_model_layers_17_residual_pre_attn/norm":138949.8125,"train/tensor_act_model_layers_13_residual_pre_attn/max_abs":53.5,"train/tensor_act_model_layers_4_mlp_up_proj/norm":5382.72998046875,"train/tensor_act_model_layers_76_residual_post_mlp/range":6,"train/tensor_param_model_layers_68_mlp_down_proj_weight/std":0.037109375,"train/tensor_act_model_layers_25_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_52_mlp_gate_proj/range":6.984375,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_14_residual_pre_attn/max":57.5,"train/tensor_grad_model_layers_0_mlp_up_proj_weight/max":380,"train/tensor_param_model_layers_0_mlp_up_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/range":0.2529296875,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/max":0.1640625,"train/tensor_act_model_layers_69_residual_post_mlp/max_abs":282,"train/tensor_act_model_layers_29_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_58_residual_pre_attn/max":234,"train/tensor_act_model_layers_72_residual_post_mlp/max":294,"train/tensor_param_model_layers_49_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_99_mlp_up_proj/min":-4.53125,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_37_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_55_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_12_mlp_up_proj_weight/min":-0.115234375,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/max":0.1240234375,"train/tensor_grad_model_layers_97_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_2_mlp/range":0,"train/tensor_act_model_layers_46_mlp/range":0,"train/tensor_act_model_layers_75_mlp_down_proj/max":16.375,"train/tensor_param_model_layers_85_mlp_down_proj_weight/norm":7.65625,"train/tensor_grad_model_layers_93_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_92_mlp/norm":8192,"train/tensor_act_model_layers_62_residual_pre_attn/range":5,"train/tensor_act_model_layers_27_mlp_down_proj/norm":30436.857421875,"train/tensor_act_model_layers_23_mlp_up_proj/max_abs":3.703125,"train/tensor_grad_model_layers_79_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_27_mlp_up_proj_weight/min":-0.13671875,"train/tensor_param_model_layers_2_mlp_down_proj_weight/max":0.1005859375,"train/tensor_act_model_layers_75_mlp_down_proj/min":14.1875,"train/tensor_param_model_layers_25_mlp_up_proj_weight/std":0.034912109375,"train/tensor_param_model_layers_14_mlp_up_proj_weight/std":0.033935546875,"train/tensor_act_model_layers_90_mlp_gate_proj/norm":5264.29248046875,"train/tensor_act_model_layers_85_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_60_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_68_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_49_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_84_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_90_mlp_down_proj_weight/mean":0,"train/tensor_param_model_layers_62_mlp_up_proj_weight/range":0.255859375,"train/tensor_act_model_layers_84_residual_post_attn/std":0.7578125,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_41_mlp_up_proj_weight/range":0.2509765625,"train/tensor_param_model_layers_0_mlp_up_proj_weight/std":0.05712890625,"train/tensor_act_model_layers_98_residual_post_attn/max_abs":394,"train/tensor_act_model_layers_21_residual_pre_attn/std":0.58203125,"train/tensor_param_model_layers_68_mlp_down_proj_weight/range":0.25390625,"train/tensor_act_model_layers_95_mlp_gate_proj/min":-4.46875,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_42_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/norm":6.46875,"train/tensor_param_model_layers_16_mlp_down_proj_weight/max_abs":0.1240234375,"train/tensor_act_model_layers_20_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_77_mlp/max":4,"train/tensor_act_model_layers_60_mlp/max_abs":4,"train/tensor_param_model_layers_85_mlp_gate_proj_weight/std":0.04296875,"train/tensor_act_model_layers_93_mlp_down_proj/mean":15,"train/tensor_act_model_layers_2_mlp_up_proj/mean":0.040283203125,"train/tensor_grad_model_layers_35_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_5_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_45_residual_pre_attn/max_abs":182,"train/tensor_param_model_layers_31_mlp_gate_proj_weight/range":0.251953125,"train/tensor_act_model_layers_87_residual_post_mlp/min":348,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_46_mlp_gate_proj/norm":5436.96533203125,"train/tensor_act_model_layers_3_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_33_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_68_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_91_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_24_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit":0,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/min":-0.125,"train/tensor_act_model_layers_50_mlp/min":4,"train/tensor_act_model_layers_49_mlp_gate_proj/std":1.84375,"train/tensor_grad_model_layers_33_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_73_mlp_down_proj_weight/max_abs":0.134765625,"train/tensor_act_model_layers_3_mlp/std":0,"train/tensor_act_model_layers_88_mlp_up_proj/std":1.828125,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/norm":6.1875,"train/tensor_act_model_layers_18_mlp_gate_proj/std":1.828125,"train/tensor_act_model_layers_23_mlp_gate_proj/mean":0.0703125,"train/tensor_act_model_layers_62_residual_post_attn/norm":507333.25,"train/tensor_act_model_layers_70_mlp_up_proj/norm":5450.5380859375,"train/tensor_act_model_layers_53_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_90_mlp_down_proj/norm":30856.623046875,"train/tensor_act_model_layers_22_residual_post_mlp/max":94,"train/tensor_grad_model_layers_59_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_44_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_76_residual_post_mlp/max_abs":310,"train/tensor_act_model_layers_87_residual_pre_attn/max_abs":350,"train/tensor_act_model_layers_77_residual_post_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_18_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_2_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_93_mlp_down_proj/max":16.625,"train/tensor_act_model_layers_32_residual_post_attn/mean":127.5,"train/tensor_act_model_layers_18_residual_post_mlp/std":0.58203125,"train/tensor_param_model_layers_8_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_2_mlp_gate_proj/max":3.421875,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_40_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_79_mlp_down_proj_weight/max":0.1328125,"train/tensor_act_model_layers_25_mlp/norm":8192,"train/tensor_grad_model_layers_65_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_7_residual_post_attn/norm":57319.37890625,"train/tensor_param_model_layers_66_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_40_mlp_gate_proj/max_abs":3.65625,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_5_residual_post_attn/mean":20,"train/tensor_param_model_layers_71_mlp_gate_proj_weight/min":-0.12890625,"train/tensor_grad_model_layers_22_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_1_residual_post_attn/std":0.115234375,"train/tensor_grad_model_layers_20_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_24_mlp_up_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_40_mlp_gate_proj/min":-3.265625,"train/tensor_act_model_layers_29_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_24_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_8_mlp_up_proj_weight/max_abs":0.123046875,"train/tensor_param_model_layers_32_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_94_mlp_gate_proj/max_abs":4.65625,"train/tensor_param_model_layers_28_mlp_down_proj_weight/norm":6.28125,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_46_residual_pre_attn/max":186,"train/tensor_act_model_layers_31_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_85_mlp_gate_proj/std":1.8125,"train/tensor_param_model_layers_78_mlp_up_proj_weight/max_abs":0.1328125,"train/tensor_act_model_layers_21_mlp_gate_proj/range":7.3125,"train/tensor_param_model_layers_19_mlp_up_proj_weight/norm":6.25,"train/tensor_param_model_layers_42_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_62_mlp/max_abs":4,"train/tensor_grad_model_layers_13_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_43_mlp_down_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_7_residual_post_mlp/mean":32,"train/tensor_act_model_layers_93_mlp_down_proj/min":14.0625,"train/tensor_grad_model_layers_79_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_21_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_40_mlp/max_abs":4,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/range":0.26953125,"train/tensor_act_model_layers_75_mlp/max_abs":4,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_91_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_41_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_55_mlp_gate_proj_weight/std":0.03564453125,"train/tensor_param_model_layers_88_mlp_down_proj_weight/norm":7.90625,"train/tensor_param_model_layers_35_mlp_up_proj_weight/std":0.03515625,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/mean":-0.0001220703125,"train/tensor_param_model_layers_88_mlp_up_proj_weight/max_abs":0.1572265625,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_71_mlp_up_proj/range":7.078125,"train/tensor_act_model_layers_74_mlp_gate_proj/max_abs":3.90625,"train/tensor_act_model_layers_90_residual_pre_attn/norm":736877.8125,"train/tensor_grad_model_layers_31_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_13_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_67_mlp/min":4,"train/tensor_act_model_layers_7_mlp_up_proj/min":-3.453125,"train/tensor_act_model_layers_87_residual_post_mlp/norm":720493.8125,"train/tensor_param_model_layers_96_mlp_down_proj_weight/max":0.1435546875,"train/tensor_grad_model_layers_88_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/min":-0.1201171875,"train/tensor_param_model_layers_14_mlp_down_proj_weight/range":0.23583984375,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_9_mlp_up_proj_weight/mean":-0.0004634857177734375,"train/tensor_act_model_layers_1_mlp_up_proj/range":8.28125,"train/tensor_act_model_layers_99_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_residual_post_mlp/std":0.8359375,"train/tensor_act_model_layers_43_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_57_mlp_up_proj/norm":5379.5849609375,"train/tensor_act_model_layers_42_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_58_mlp_up_proj_weight/min":-0.1259765625,"train/tensor_act_model_layers_89_residual_pre_attn/mean":356,"train/tensor_param_model_layers_82_mlp_down_proj_weight/max":0.12890625,"train/tensor_act_model_layers_36_residual_pre_attn/range":5,"train/tensor_act_model_layers_30_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_24_residual_post_mlp/norm":204395.15625,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/norm":0,"train/tensor_grad_model_layers_69_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_66_mlp_up_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_47_mlp_down_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_45_mlp_up_proj/max_abs":4.15625,"train/tensor_act_model_layers_97_mlp_down_proj/std":0.65625,"train/tensor_act_model_layers_33_mlp/min":4,"train/tensor_act_model_layers_59_mlp_gate_proj/norm":5333.6103515625,"train/tensor_act_model_layers_55_residual_post_attn/range":5,"train/tensor_param_model_layers_11_mlp_up_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_86_residual_post_attn/min":340,"train/tensor_param_model_layers_57_mlp_up_proj_weight/std":0.035400390625,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_77_mlp_down_proj_weight/max_abs":0.13671875,"train/tensor_act_model_layers_12_mlp_gate_proj/std":1.828125,"train/tensor_act_model_layers_56_mlp_up_proj/min":-3.984375,"train/tensor_act_model_layers_70_residual_post_mlp/max":286,"train/tensor_act_model_layers_83_mlp_down_proj/std":0.470703125,"train/tensor_act_model_layers_74_residual_pre_attn/max_abs":298,"train/tensor_grad_model_layers_43_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_48_mlp_down_proj_weight/norm":6.28125,"train/tensor_grad_model_layers_23_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_37_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_91_mlp_gate_proj/std":1.8828125,"train/tensor_act_model_layers_99_mlp_gate_proj/mean":-0.0927734375,"train/tensor_act_model_layers_88_residual_post_attn/norm":720493.8125,"train/tensor_act_model_layers_61_mlp_down_proj/std":0.396484375,"train/tensor_act_model_layers_22_mlp_down_proj/norm":30525.333984375,"train/tensor_act_model_layers_66_mlp_gate_proj/mean":-0.05810546875,"train/tensor_act_model_layers_32_mlp_down_proj/std":0.388671875,"train/tensor_act_model_layers_17_mlp_gate_proj/norm":5212.3740234375,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_64_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_31_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_39_mlp_gate_proj/mean":0.21875,"train/tensor_act_model_layers_11_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_84_mlp_up_proj_weight/min":-0.1337890625,"train/tensor_act_model_layers_37_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_92_residual_post_attn/range":6,"train/tensor_param_model_layers_42_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_22_residual_post_attn/max_abs":90,"train/tensor_act_model_layers_29_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_23_mlp/max_abs":4,"train/tensor_act_model_layers_75_mlp_gate_proj/max":4.15625,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/mean":0,"train/tensor_param_model_layers_58_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_15_mlp_up_proj_weight/range":0.244140625,"train/tensor_act_model_layers_71_residual_post_mlp/max":290,"train/tensor_act_model_layers_20_residual_pre_attn/max_abs":82,"train/tensor_param_model_layers_42_mlp_up_proj_weight/mean":0.0012054443359375,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/std":0.033447265625,"train/tensor_param_model_layers_92_mlp_down_proj_weight/min":-0.1357421875,"train/tensor_param_model_layers_13_mlp_up_proj_weight/max_abs":0.12890625,"train/tensor_act_model_layers_14_mlp_gate_proj/std":1.8359375,"train/tensor_param_model_layers_37_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_28_residual_post_mlp/std":0.67578125,"train/tensor_act_model_layers_35_residual_post_mlp/min":141,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_59_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_30_mlp/mean":4,"train/tensor_act_model_layers_17_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_33_mlp/std":0,"train/tensor_act_model_layers_52_residual_post_attn/range":5,"train/tensor_act_model_layers_95_mlp_up_proj/norm":5399.09375,"train/tensor_act_model_layers_44_mlp_up_proj/range":7.8125,"train/tensor_act_model_layers_80_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_69_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_61_mlp_up_proj/std":1.859375,"train/tensor_param_model_layers_39_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_residual_post_attn/norm":122574.4765625,"train/tensor_act_model_layers_39_mlp_up_proj/norm":5377.22314453125,"train/tensor_param_model_layers_86_mlp_gate_proj_weight/mean":0.00087738037109375,"train/tensor_act_model_layers_60_mlp_up_proj/max":3.484375,"train/tensor_act_model_layers_91_residual_post_mlp/max":370,"train/tensor_act_model_layers_91_mlp_down_proj/norm":30730.876953125,"train/tensor_param_model_layers_39_mlp_down_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_87_mlp_up_proj_weight/max_abs":0,"train/tensor_param_model_layers_55_mlp_down_proj_weight/std":0.035400390625,"train/tensor_param_model_layers_48_mlp_up_proj_weight/max":0.12109375,"train/tensor_act_model_layers_86_residual_post_attn/max":346,"train/tensor_act_model_layers_52_mlp_gate_proj/min":-3.546875,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_18_mlp_down_proj/norm":30443.2578125,"train/tensor_param_model_layers_61_mlp_down_proj_weight/max_abs":0.12451171875,"train/tensor_param_model_layers_53_mlp_down_proj_weight/mean":0.00102996826171875,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_75_mlp_down_proj_weight/max":0.12890625,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_55_residual_post_attn/mean":220,"train/tensor_param_model_layers_21_mlp_down_proj_weight/min":-0.11669921875,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_37_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_90_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_16_mlp_up_proj_weight/norm":6.25,"train/tensor_act_model_layers_0_residual_post_mlp/norm":8196.392578125,"train/tensor_act_model_layers_80_mlp/range":0,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_48_residual_post_attn/range":5,"train/tensor_act_model_layers_1_mlp_gate_proj/max":4.03125,"train/tensor_grad_model_layers_11_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_12_mlp_up_proj_weight/std":0.034423828125,"train/tensor_act_model_layers_1_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_30_mlp_gate_proj_weight/std":0.03515625,"train/tensor_param_model_layers_23_mlp_up_proj_weight/norm":6.34375,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_20_mlp_gate_proj/max_abs":3.515625,"train/tensor_grad_model_layers_30_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_39_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_15_mlp_gate_proj/max_abs":3.9375,"train/tensor_act_model_layers_80_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_3_mlp/range":0,"train/tensor_act_model_layers_38_mlp_down_proj/norm":30354.720703125,"train/tensor_act_model_layers_61_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_1_residual_pre_attn/norm":8196.392578125,"train/tensor_act_model_layers_16_mlp_down_proj/range":1.75,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/min":-0.12890625,"train/tensor_grad_model_layers_4_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_32_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_31_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_17_mlp_up_proj/norm":5247.92138671875,"train/tensor_grad_model_layers_70_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_66_mlp_up_proj/range":7.296875,"train/tensor_grad_model_layers_42_mlp_gate_proj_weight/max_abs":0,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_74_mlp_up_proj_weight/max":0,"train/tensor_param_model_layers_42_mlp_down_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_84_residual_pre_attn/min":332,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_15_residual_post_mlp/norm":130693.09375,"train/tensor_act_model_layers_82_residual_post_attn/min":324,"train/tensor_act_model_layers_6_mlp_up_proj/min":-3.5,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_18_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_residual_post_attn/max":394,"train/tensor_param_model_layers_84_mlp_up_proj_weight/max_abs":0.140625,"train/tensor_param_model_layers_98_mlp_down_proj_weight/max":0.1474609375,"train/tensor_act_model_layers_37_mlp_gate_proj/min":-3.4375,"train/tensor_act_model_layers_83_mlp/min":4,"train/tensor_param_model_layers_66_mlp_up_proj_weight/mean":-0.00128173828125,"train/tensor_param_model_layers_94_mlp_gate_proj_weight/norm":8.5625,"train/tensor_act_model_layers_85_mlp_up_proj/max":3.984375,"train/tensor_act_model_layers_98_residual_post_attn/min":388,"train/tensor_act_model_layers_89_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_98_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_64_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_47_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_83_mlp/max_abs":4,"train/tensor_act_model_layers_82_residual_post_attn/max":330,"train/tensor_param_model_layers_68_mlp_up_proj_weight/max":0.1259765625,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_58_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_mlp_up_proj/max_abs":3.453125,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_residual_pre_attn/max_abs":394,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/mean":0.00010442733764648438,"train/tensor_act_model_layers_16_residual_post_mlp/norm":138949.8125,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_74_residual_post_mlp/min":296,"train/tensor_act_model_layers_0_residual_pre_attn/std":0.09130859375,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_91_mlp_up_proj/min":-4.0625,"train/tensor_act_model_layers_60_residual_post_attn/mean":240,"train/tensor_act_model_layers_39_residual_pre_attn/mean":156,"train/tensor_act_model_layers_56_mlp/range":0,"train/tensor_param_model_layers_74_mlp_up_proj_weight/range":0.2568359375,"train/tensor_act_model_layers_83_mlp_up_proj/std":1.875,"train/tensor_act_model_layers_65_mlp_up_proj/min":-3.6875,"train/tensor_act_model_layers_84_residual_post_attn/mean":336,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_73_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_92_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_77_mlp_up_proj/mean":-0.2265625,"train/tensor_act_model_layers_78_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_14_mlp_up_proj_weight/min":-0.1240234375,"train/tensor_param_model_layers_25_mlp_up_proj_weight/min":-0.1240234375,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_88_mlp_up_proj_weight/mean":-0.000274658203125,"train/tensor_param_model_layers_27_mlp_up_proj_weight/std":0.035400390625,"train/tensor_param_model_layers_91_mlp_down_proj_weight/std":0.044921875,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_85_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_17_residual_post_attn/range":3.5,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_6_residual_pre_attn/max":24.875,"train/tensor_param_model_layers_65_mlp_down_proj_weight/max":0.11962890625,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_1_residual_post_mlp/mean":8,"train/tensor_param_model_layers_35_mlp_down_proj_weight/max":0.1240234375,"train/tensor_param_model_layers_29_mlp_down_proj_weight/std":0.03466796875,"train/tensor_act_model_layers_86_mlp_up_proj/std":1.8828125,"train/tensor_param_model_layers_25_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_21_mlp_up_proj/max":3.59375,"train/tensor_act_model_layers_95_mlp_up_proj/max_abs":4.40625,"train/tensor_param_model_layers_3_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_11_mlp_down_proj/norm":30547.74609375,"train/tensor_act_model_layers_52_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_54_mlp_gate_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_57_residual_post_attn/min":225,"train/tensor_act_model_layers_24_residual_pre_attn/min":94,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_87_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_7_residual_post_attn/std":0.3125,"train/tensor_act_model_layers_65_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_93_residual_pre_attn/max_abs":374,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/mean":0.0003204345703125,"train/tensor_act_model_layers_78_mlp_up_proj/min":-3.890625,"train/tensor_act_model_layers_17_mlp/std":0,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_59_mlp_up_proj/norm":5288.80029296875,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_72_mlp_down_proj/norm":30700.228515625,"train/tensor_act_model_layers_8_mlp/range":0,"train/tensor_act_model_layers_39_residual_pre_attn/max":158,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_15_residual_post_mlp/mean":63.75,"train/tensor_act_model_layers_52_mlp/mean":4,"train/tensor_param_model_layers_94_mlp_up_proj_weight/std":0.047119140625,"train/tensor_grad_model_layers_62_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/min":0,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_53_mlp/min":4,"train/tensor_grad_model_layers_51_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_37_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_29_mlp_gate_proj_weight/std":0.03515625,"train/tensor_grad_model_layers_38_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_59_residual_post_mlp/max_abs":242,"train/tensor_act_model_layers_98_mlp_down_proj/max_abs":18.25,"train/tensor_act_model_layers_87_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_81_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_7_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_residual_post_attn/norm":605806.125,"train/tensor_act_model_layers_41_residual_pre_attn/min":161,"train/tensor_param_model_layers_10_mlp_up_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_6_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_mlp_gate_proj/range":7.484375,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_80_residual_post_mlp/std":0.7578125,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_16_residual_post_mlp/mean":68,"train/tensor_param_model_layers_43_mlp_up_proj_weight/min":-0.11572265625,"train/tensor_act_model_layers_66_residual_post_attn/max_abs":266,"train/tensor_act_model_layers_44_mlp_gate_proj/max":3.578125,"train/tensor_act_model_layers_7_residual_post_mlp/std":0.314453125,"train/tensor_param_model_layers_31_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_mlp_gate_proj/max_abs":3.71875,"train/tensor_act_model_layers_70_mlp_down_proj/std":0.453125,"train/tensor_param_model_layers_47_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_75_mlp_up_proj/range":8.28125,"train/tensor_param_model_layers_18_mlp_up_proj_weight/min":-0.1328125,"train/tensor_act_model_layers_57_residual_post_attn/max":230,"train/tensor_param_model_layers_90_mlp_gate_proj_weight/range":0.28515625,"train/tensor_act_model_layers_62_mlp_gate_proj/range":7.875,"train/tensor_param_model_layers_64_mlp_gate_proj_weight/min":-0.126953125,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_88_mlp_gate_proj/range":8,"train/tensor_act_model_layers_25_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_27_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_87_residual_pre_attn/range":6,"train/tensor_act_model_layers_94_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_75_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_4_mlp_up_proj/min":-3.53125,"train/tensor_grad_model_layers_97_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_24_residual_pre_attn/norm":196250.90625,"train/tensor_act_model_layers_58_residual_post_attn/std":0.8359375,"train/tensor_act_model_layers_26_mlp_gate_proj/norm":5268.5517578125,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/frac_near_user_limit":0.983642578125,"train/tensor_act_model_layers_24_residual_post_mlp/mean":100,"train/tensor_grad_model_layers_44_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_58_mlp_gate_proj/min":-3.9375,"train/tensor_act_model_layers_82_mlp/min":4,"train/tensor_act_model_layers_97_mlp/mean":4,"train/tensor_param_model_layers_87_mlp_up_proj_weight/std":0.042724609375,"train/tensor_act_model_layers_63_residual_pre_attn/max_abs":254,"train/tensor_act_model_layers_45_mlp_up_proj/std":1.8828125,"train/tensor_param_model_layers_28_mlp_gate_proj_weight/std":0.03515625,"train/tensor_param_model_layers_58_mlp_down_proj_weight/std":0.035400390625,"train/tensor_act_model_layers_28_mlp/range":0,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/norm":6.5,"train/tensor_act_model_layers_95_residual_post_mlp/min":380,"train/tensor_act_model_layers_10_mlp_up_proj/max_abs":3.265625,"train/tensor_act_model_layers_84_mlp/max":4,"train/tensor_grad_model_layers_78_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_18_mlp_up_proj_weight/mean":-0.0002536773681640625,"train/tensor_grad_model_layers_20_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_84_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_62_residual_post_attn/max":250,"train/tensor_param_model_layers_79_mlp_up_proj_weight/std":0.0400390625,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_88_residual_post_mlp/min":352,"train/tensor_param_model_layers_13_mlp_gate_proj_weight/range":0.2490234375,"train/tensor_act_model_layers_48_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_32_mlp/max_abs":4,"train/tensor_act_model_layers_9_mlp_gate_proj/std":1.8203125,"train/tensor_act_model_layers_99_residual_post_attn/range":6,"train/tensor_grad_model_layers_81_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_63_mlp/norm":8192,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/mean":0.001373291015625,"train/tensor_act_model_layers_89_mlp_gate_proj/max_abs":4.09375,"train/tensor_act_model_layers_87_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_50_mlp_up_proj_weight/min":-0.1171875,"train/tensor_act_model_layers_94_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_18_mlp_up_proj_weight/max_abs":0,"train/tensor_grad_model_layers_33_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/frac_near_user_limit":0.9967041015625,"train/tensor_param_model_layers_78_mlp_up_proj_weight/std":0.039306640625,"train/tensor_act_model_layers_66_mlp_gate_proj/max":3.53125,"train/tensor_param_model_layers_41_mlp_down_proj_weight/range":0.2412109375,"train/tensor_act_model_layers_16_residual_pre_attn/norm":130693.09375,"train/tensor_param_model_layers_59_mlp_down_proj_weight/max":0.12451171875,"train/tensor_act_model_layers_54_residual_pre_attn/min":213,"train/tensor_act_model_layers_9_mlp/norm":8192,"train/tensor_act_model_layers_83_mlp/std":0,"train/tensor_act_model_layers_86_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_28_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_95_mlp_gate_proj_weight/range":0,"train/tensor_param_model_layers_41_mlp_up_proj_weight/max_abs":0.1259765625,"train/tensor_param_model_layers_38_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_65_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_60_mlp_up_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_20_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_18_residual_post_attn/max_abs":74,"train/tensor_act_model_layers_80_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_77_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_96_residual_post_mlp/max_abs":390,"train/tensor_act_model_layers_97_residual_pre_attn/min":384,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_20_mlp_up_proj_weight/range":0.25,"train/tensor_grad_model_layers_96_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_65_mlp/min":4,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_96_mlp/norm":8192,"train/tensor_param_model_layers_39_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_68_mlp_up_proj/std":1.8359375,"train/tensor_act_model_layers_15_residual_pre_attn/norm":122617.9140625,"train/tensor_grad_model_layers_89_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_98_mlp_gate_proj/mean":-0.21484375,"train/tensor_act_model_layers_50_residual_pre_attn/max":202,"train/tensor_act_model_layers_43_residual_post_mlp/max":178,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/range":0.2529296875,"train/tensor_param_model_layers_14_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_72_residual_post_attn/mean":288,"train/tensor_param_model_layers_28_mlp_up_proj_weight/norm":6.34375,"train/tensor_act_model_layers_85_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_17_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_67_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_65_mlp_up_proj_weight/min":-0.1337890625,"train/tensor_act_model_layers_43_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_57_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_residual_post_attn/min":244,"train/tensor_act_model_layers_10_residual_post_attn/norm":81864.0625,"train/tensor_param_model_layers_84_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_38_mlp_down_proj_weight/mean":0.0011444091796875,"train/tensor_act_model_layers_98_mlp_up_proj/min":-4.5625,"train/tensor_param_model_layers_37_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_15_residual_pre_attn/std":0.494140625,"train/tensor_param_model_layers_4_mlp_gate_proj_weight/max_abs":0.1142578125,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_76_mlp_down_proj_weight/norm":7,"train/tensor_param_model_layers_11_mlp_down_proj_weight/min":-0.11328125,"train/tensor_act_model_layers_91_residual_post_attn/mean":364,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_49_residual_post_attn/range":5,"train/tensor_act_model_layers_29_mlp/min":4,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_43_mlp_up_proj/max_abs":3.78125,"train/tensor_param_model_layers_27_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_58_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_57_mlp_gate_proj_weight/norm":6.40625,"train/tensor_param_model_layers_12_mlp_gate_proj_weight/range":0.2470703125,"train/tensor_act_model_layers_68_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_63_residual_post_attn/min":248,"train/tensor_act_model_layers_58_residual_pre_attn/norm":474629.03125,"train/tensor_grad_model_layers_59_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_23_mlp_down_proj_weight/range":0.24853515625,"train/tensor_act_model_layers_70_mlp_gate_proj/range":7.609375,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_77_mlp_down_proj/range":3,"train/tensor_act_model_layers_3_mlp_up_proj/max_abs":3.5,"train/tensor_grad_model_layers_96_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_64_mlp_gate_proj/range":7.75,"train/tensor_act_model_layers_73_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_90_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_5_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_97_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_91_mlp_up_proj_weight/norm":8.3125,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_14_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_33_mlp_gate_proj_weight/range":0.2607421875,"train/tensor_act_model_layers_44_mlp/min":4,"train/tensor_param_model_layers_8_mlp_down_proj_weight/mean":0.0001049041748046875,"train/tensor_param_model_layers_82_mlp_up_proj_weight/min":-0.1376953125,"train/tensor_param_model_layers_50_mlp_down_proj_weight/max_abs":0.126953125,"train/tensor_act_model_layers_4_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_84_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_1_mlp_down_proj/range":8.3125,"train/tensor_act_model_layers_73_mlp_up_proj/min":-3.640625,"train/tensor_act_model_layers_8_mlp_down_proj/max":16.125,"train/tensor_param_model_layers_49_mlp_up_proj_weight/max":0.12255859375,"train/tensor_act_model_layers_97_residual_post_attn/range":6,"train/tensor_act_model_layers_33_residual_post_mlp/min":134,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_41_residual_post_attn/max_abs":166,"train/tensor_param_model_layers_53_mlp_down_proj_weight/min":-0.12353515625,"train/tensor_act_model_layers_66_mlp/range":0,"train/tensor_param_model_layers_99_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_residual_post_attn/std":0.546875,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_24_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_7_residual_pre_attn/norm":57287.984375,"train/tensor_param_model_layers_33_mlp_up_proj_weight/max_abs":0.1328125,"train/tensor_act_model_layers_56_mlp_gate_proj/std":1.84375,"train/tensor_param_model_layers_38_mlp_down_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_27_mlp_up_proj_weight/max":0,"train/tensor_grad_model_layers_49_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_8_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_16_mlp_up_proj/range":7.3125,"train/tensor_param_model_layers_49_mlp_down_proj_weight/min":-0.1298828125,"train/tensor_act_model_layers_80_residual_pre_attn/norm":654958,"train/tensor_grad_model_layers_29_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_50_residual_pre_attn/max_abs":202,"train/tensor_grad_model_layers_77_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_97_mlp_up_proj/mean":-0.1640625,"train/tensor_grad_model_layers_82_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_58_mlp/std":0,"train/tensor_param_model_layers_26_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_11_mlp_down_proj_weight/max_abs":0.11328125,"train/tensor_act_model_layers_15_residual_post_attn/std":0.52734375,"train/tensor_grad_model_layers_78_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_52_residual_post_mlp/range":5,"train/tensor_act_model_layers_60_residual_pre_attn/max":242,"train/tensor_act_model_layers_71_mlp_gate_proj/norm":5392.5625,"train/tensor_act_model_layers_90_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_93_mlp_gate_proj/max_abs":4.34375,"train/tensor_act_model_layers_62_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_4_residual_post_attn/min":15.3125,"train/tensor_grad_model_layers_10_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_52_mlp_down_proj/mean":14.875,"train/tensor_act_model_layers_62_residual_pre_attn/max":250,"train/tensor_act_model_layers_87_residual_post_mlp/mean":352,"train/tensor_act_model_layers_7_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_residual_post_mlp/range":6,"train/tensor_act_model_layers_37_residual_pre_attn/min":145,"train/tensor_act_model_layers_3_residual_post_attn/max_abs":12.5625,"train/tensor_act_model_layers_72_residual_post_mlp/max_abs":294,"train/tensor_act_model_layers_63_residual_post_attn/max":254,"train/tensor_act_model_layers_48_mlp_gate_proj/max_abs":3.90625,"train/tensor_param_model_layers_71_mlp_down_proj_weight/std":0.037353515625,"train/tensor_act_model_layers_47_mlp_up_proj/std":1.8203125,"train/tensor_grad_model_layers_64_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_27_mlp/range":0,"train/tensor_grad_model_layers_16_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_31_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_51_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_27_mlp_down_proj_weight/min":-0.12451171875,"train/tensor_act_model_layers_82_residual_post_mlp/min":328,"train/tensor_act_model_layers_27_mlp_down_proj/range":1.625,"train/tensor_act_model_layers_55_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_residual_pre_attn/mean":200,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_87_residual_pre_attn/norm":712301.875,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_50_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_49_mlp_gate_proj/norm":5342.87158203125,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_90_residual_post_mlp/norm":745069.8125,"train/tensor_param_model_layers_88_mlp_up_proj_weight/std":0.044189453125,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_37_mlp_down_proj_weight/norm":6.21875,"train/tensor_act_model_layers_77_mlp_down_proj/max":17.25,"train/tensor_act_model_layers_47_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_24_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_94_mlp_down_proj_weight/min":-0.138671875,"train/tensor_param_model_layers_17_mlp_gate_proj_weight/max":0.125,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/max_abs":0,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/std":0.03564453125,"train/tensor_act_model_layers_90_residual_post_mlp/max":366,"train/tensor_act_model_layers_20_mlp_down_proj/norm":30493.419921875,"train/tensor_grad_model_layers_3_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_89_mlp_down_proj_weight/min":-0.142578125,"train/tensor_act_model_layers_30_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_18_mlp_up_proj/mean":-0.0322265625,"train/tensor_act_model_layers_80_residual_post_mlp/mean":324,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/norm":6.0625,"train/tensor_act_model_layers_85_mlp/max":4,"train/tensor_param_model_layers_98_mlp_up_proj_weight/max_abs":0.1630859375,"train/tensor_act_model_layers_73_mlp_up_proj/mean":0.039306640625,"train/tensor_act_model_layers_60_mlp/range":0,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_54_mlp_down_proj/std":0.435546875,"train/tensor_act_model_layers_92_residual_post_attn/max_abs":370,"train/tensor_act_model_layers_59_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_33_residual_post_mlp/range":4,"train/tensor_param_model_layers_59_mlp_gate_proj_weight/norm":6.4375,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_60_residual_post_attn/range":5,"train/tensor_act_model_layers_5_mlp/norm":8192,"train/tensor_param_model_layers_48_mlp_down_proj_weight/max_abs":0.12890625,"train/tensor_param_model_layers_0_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_81_residual_pre_attn/min":320,"train/tensor_param_model_layers_52_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_24_mlp_gate_proj/range":7.609375,"train/tensor_act_model_layers_72_mlp/max":4,"train/tensor_act_model_layers_74_mlp/range":0,"train/tensor_act_model_layers_29_residual_post_attn/mean":115.5,"train/tensor_act_model_layers_84_mlp_up_proj/std":1.7734375,"train/tensor_act_model_layers_48_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_29_mlp_up_proj_weight/mean":-0.000209808349609375,"train/tensor_act_model_layers_83_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_mlp_down_proj/std":0.3671875,"train/tensor_grad_model_layers_56_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_49_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_9_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_30_mlp_down_proj/max":16,"train/tensor_act_model_layers_83_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_38_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_10_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_16_mlp_gate_proj_weight/range":0.24755859375,"train/tensor_param_model_layers_92_mlp_down_proj_weight/norm":8.1875,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_77_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_71_mlp_gate_proj/max":3.53125,"train/tensor_grad_model_layers_28_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_59_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_83_residual_post_mlp/min":332,"train/tensor_act_model_layers_4_mlp_down_proj/max":16.5,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_37_mlp_down_proj/max_abs":16.375,"train/tensor_grad_model_layers_0_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_38_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_10_mlp_down_proj_weight/range":0.244140625,"train/tensor_act_model_layers_55_residual_pre_attn/std":0.8203125,"train/tensor_act_model_layers_24_residual_post_attn/std":0.61328125,"train/tensor_param_model_layers_5_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_87_mlp/min":4,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_57_mlp_down_proj_weight/norm":6.3125,"train/tensor_act_model_layers_58_residual_post_mlp/mean":236,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/mean":0,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_73_mlp_gate_proj/range":7.296875,"train/tensor_grad_model_layers_6_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_25_residual_pre_attn/max_abs":102,"train/tensor_act_model_layers_61_mlp/max":4,"train/tensor_grad_model_layers_88_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_45_residual_pre_attn/std":0.80859375,"train/tensor_act_model_layers_90_residual_post_mlp/mean":364,"train/tensor_act_model_layers_4_mlp/max_abs":4,"train/tensor_act_model_layers_14_mlp_down_proj/mean":14.8125,"train/tensor_param_model_layers_10_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_50_mlp/norm":8192,"train/tensor_act_model_layers_58_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_44_residual_post_mlp/max_abs":182,"train/tensor_param_model_layers_88_mlp_down_proj_weight/std":0.043701171875,"train/tensor_param_model_layers_83_mlp_gate_proj_weight/max_abs":0.1416015625,"train/tensor_grad_model_layers_43_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_10_mlp_down_proj/norm":30492.388671875,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_75_residual_pre_attn/range":6,"train/tensor_param_model_layers_44_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_56_residual_post_attn/norm":458213.03125,"train/tensor_act_model_layers_94_residual_pre_attn/range":6,"train/tensor_param_model_layers_89_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_69_mlp/mean":4,"train/tensor_act_model_layers_31_residual_pre_attn/std":0.68359375,"train/tensor_act_model_layers_80_residual_post_mlp/norm":663150,"train/tensor_act_model_layers_49_mlp_up_proj/max_abs":4.1875,"train/tensor_param_model_layers_80_mlp_down_proj_weight/min":-0.142578125,"train/tensor_act_model_layers_60_residual_pre_attn/max_abs":242,"train/tensor_act_model_layers_76_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_2_mlp_up_proj/range":7.90625,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/max":0.126953125,"train/tensor_act_model_layers_89_mlp_gate_proj/min":-3.640625,"train/tensor_param_model_layers_69_mlp_up_proj_weight/mean":-0.0003509521484375,"train/tensor_grad_model_layers_87_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_88_mlp_down_proj_weight/max_abs":0.142578125,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/norm":8.75,"train/tensor_act_model_layers_92_mlp_down_proj/norm":30614.21484375,"train/tensor_param_model_layers_54_mlp_up_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_residual_post_mlp/min":252,"train/tensor_act_model_layers_33_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_60_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_84_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_75_mlp/mean":4,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_15_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_56_mlp/max_abs":4,"train/tensor_param_model_layers_76_mlp_up_proj_weight/std":0.0390625,"train/tensor_act_model_layers_35_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_82_mlp_up_proj_weight/norm":7.53125,"train/tensor_param_model_layers_23_mlp_down_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_29_mlp_up_proj/std":1.8203125,"train/tensor_act_model_layers_53_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_46_mlp_down_proj/norm":30481.92578125,"train/tensor_act_model_layers_90_mlp_gate_proj/std":1.8046875,"train/tensor_act_model_layers_84_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_13_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_88_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_mlp_gate_proj/max":3.609375,"train/tensor_act_model_layers_4_mlp_gate_proj/min":-3.703125,"train/tensor_act_model_layers_76_mlp_gate_proj/range":7.84375,"train/tensor_grad_model_layers_50_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_22_mlp_up_proj_weight/mean":-0.000888824462890625,"train/tensor_act_model_layers_67_residual_post_mlp/max":274,"train/tensor_act_model_layers_89_mlp_gate_proj/mean":0.10595703125,"train/tensor_act_model_layers_20_mlp_gate_proj/std":1.8515625,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/norm":0,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/norm":6.3125,"train/tensor_act_model_layers_10_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_58_residual_post_mlp/norm":482821,"train/tensor_act_model_layers_78_mlp_gate_proj/max_abs":3.78125,"train/tensor_act_model_layers_17_residual_post_mlp/max_abs":73.5,"train/tensor_act_model_layers_6_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_residual_post_attn/min":7.40625,"train/tensor_grad_model_layers_99_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_49_mlp_gate_proj_weight/min":-0.12451171875,"train/tensor_act_model_layers_19_mlp/max":4,"train/tensor_grad_model_layers_94_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_8_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_69_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_99_residual_post_attn/mean":396,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/range":0.2685546875,"train/tensor_param_model_layers_23_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_64_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_8_mlp_gate_proj/min":-3.328125,"train/tensor_act_model_layers_30_mlp_up_proj/norm":5329.65234375,"train/tensor_act_model_layers_37_residual_post_attn/max_abs":150,"train/tensor_act_model_layers_59_mlp_up_proj/max":3.421875,"train/tensor_act_model_layers_10_mlp_up_proj/norm":5391.54541015625,"train/tensor_grad_model_layers_74_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_63_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_72_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_27_mlp_gate_proj/norm":5398.1640625,"train/tensor_act_model_layers_62_residual_pre_attn/norm":507396.8125,"train/tensor_act_model_layers_57_residual_post_mlp/mean":232,"train/tensor_act_model_layers_27_residual_pre_attn/std":0.62109375,"train/tensor_act_model_layers_13_mlp_gate_proj/norm":5311.38330078125,"train/tensor_grad_model_layers_53_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_2_residual_post_mlp/max_abs":12.4375,"train/tensor_act_model_layers_82_mlp_down_proj/norm":30720.765625,"train/tensor_grad_model_layers_66_mlp_down_proj_weight/max_abs":0,"train/tensor_param_model_layers_69_mlp_up_proj_weight/range":0.25390625,"train/tensor_act_model_layers_40_mlp/std":0,"train/tensor_act_model_layers_14_mlp_gate_proj/norm":5362.7744140625,"train/tensor_param_model_layers_5_mlp_down_proj_weight/max_abs":0.1201171875,"train/tensor_grad_model_layers_57_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_77_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_39_mlp/mean":4,"train/tensor_param_model_layers_29_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_59_residual_pre_attn/mean":236,"train/tensor_act_model_layers_86_mlp_up_proj/max_abs":4.0625,"train/tensor_act_model_layers_29_residual_pre_attn/min":114,"train/tensor_act_model_layers_65_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_4_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_8_residual_pre_attn/min":30.875,"train/tensor_act_model_layers_82_residual_post_mlp/norm":679533.9375,"train/tensor_act_model_layers_15_residual_post_attn/min":58.5,"train/tensor_param_model_layers_15_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_6_mlp_up_proj_weight/max":0,"train/tensor_act_model_layers_40_residual_post_mlp/max_abs":166,"train/tensor_act_model_layers_80_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_42_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_22_mlp_down_proj_weight/min":0,"train/tensor_param_model_layers_79_mlp_down_proj_weight/max_abs":0.142578125,"train/tensor_param_model_layers_16_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_35_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_50_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_40_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_77_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_residual_post_attn/max":65.5,"train/tensor_param_model_layers_2_mlp_up_proj_weight/mean":0.0003147125244140625,"train/tensor_act_model_layers_89_residual_post_mlp/max_abs":362,"train/tensor_param_model_layers_28_mlp_up_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_74_residual_post_mlp/max":302,"train/tensor_act_model_layers_7_mlp_gate_proj/range":7.03125,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/max":0.12060546875,"train/tensor_grad_model_layers_7_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_11_mlp/max_abs":4,"train/tensor_param_model_layers_25_mlp_up_proj_weight/mean":0.00145721435546875,"train/tensor_grad_model_layers_83_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_42_mlp_up_proj_weight/max_abs":0.12255859375,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_61_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_16_residual_post_attn/min":62.25,"train/tensor_act_model_layers_97_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_65_mlp_down_proj/max":16.25,"train/tensor_act_model_layers_19_mlp_up_proj/mean":-0.23046875,"train/tensor_act_model_layers_57_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_44_mlp_up_proj_weight/norm":6.375,"train/tensor_act_model_layers_75_mlp/std":0,"train/tensor_param_model_layers_69_mlp_down_proj_weight/max":0.1318359375,"train/tensor_act_model_layers_67_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_31_mlp_down_proj_weight/max_abs":0.125,"train/tensor_param_model_layers_46_mlp_gate_proj_weight/mean":9.298324584960938e-05,"train/tensor_act_model_layers_98_residual_post_mlp/norm":810605.625,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_84_mlp_down_proj_weight/max":0.1337890625,"train/tensor_act_model_layers_82_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_5_mlp_gate_proj_weight/std":0.032958984375,"train/tensor_act_model_layers_69_mlp/min":4,"train/tensor_act_model_layers_86_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_92_mlp_down_proj/frac_near_user_limit":1,"train/tensor_grad_model_layers_60_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_56_mlp_up_proj_weight/norm":6.4375,"train/tensor_act_model_layers_50_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_22_mlp_up_proj/max_abs":4.125,"train/tensor_param_model_layers_1_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_residual_post_attn/std":0.91015625,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_48_mlp_down_proj/range":1.875,"train/tensor_grad_model_layers_41_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_15_residual_post_attn/max":61.5,"train/tensor_grad_model_layers_79_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_36_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_45_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_30_mlp_up_proj/max":3.734375,"train/tensor_act_model_layers_25_residual_pre_attn/range":4,"train/tensor_act_model_layers_58_residual_post_attn/max":234,"train/tensor_act_model_layers_80_mlp_up_proj/max":3.53125,"train/tensor_param_model_layers_44_mlp_down_proj_weight/norm":6.28125,"train/tensor_param_model_layers_21_mlp_down_proj_weight/mean":0.000701904296875,"train/tensor_param_model_layers_92_mlp_gate_proj_weight/max":0.150390625,"train/tensor_param_model_layers_8_mlp_down_proj_weight/norm":6.09375,"train/tensor_act_model_layers_98_residual_pre_attn/min":388,"train/tensor_act_model_layers_78_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_53_residual_pre_attn/max":214,"train/tensor_param_model_layers_77_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_78_mlp/max_abs":4,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_68_mlp_down_proj/std":0.404296875,"train/tensor_act_model_layers_0_residual_post_attn/std":0.09228515625,"train/tensor_act_model_layers_51_mlp_up_proj/std":1.828125,"train/tensor_param_model_layers_26_mlp_gate_proj_weight/std":0.03515625,"train/tensor_act_model_layers_2_residual_pre_attn/std":0.11669921875,"train/tensor_act_model_layers_62_residual_post_attn/std":0.89453125,"train/tensor_grad_model_layers_70_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_63_residual_post_mlp/max":258,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_5_mlp_up_proj_weight/min":-0.126953125,"train/tensor_act_model_layers_96_mlp_up_proj/range":9.0625,"train/tensor_param_model_layers_37_mlp_down_proj_weight/std":0.034423828125,"train/tensor_grad_model_layers_25_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_94_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_8_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_51_residual_post_attn/range":5,"train/tensor_act_model_layers_90_mlp/mean":4,"train/tensor_act_model_layers_43_mlp_up_proj/range":7.28125,"train/tensor_act_model_layers_80_residual_post_attn/min":316,"train/tensor_act_model_layers_32_mlp/mean":4,"train/tensor_grad_model_layers_42_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_2_mlp_down_proj/norm":31035.6640625,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/max_abs":0.140625,"train/tensor_grad_model_layers_19_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_79_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_23_residual_post_mlp/range":4,"train/tensor_act_model_layers_21_residual_post_mlp/max_abs":90,"train/tensor_act_model_layers_47_mlp_down_proj/norm":30472.732421875,"train/tensor_act_model_layers_16_residual_pre_attn/max":65.5,"train/tensor_param_model_layers_36_mlp_up_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_80_residual_post_attn/max":322,"train/tensor_act_model_layers_0_residual_post_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_42_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_2_mlp_gate_proj_weight/max_abs":0.11181640625,"train/tensor_grad_model_layers_86_mlp_down_proj_weight/std":0,"train/tensor_param_model_layers_13_mlp_down_proj_weight/std":0.03369140625,"train/tensor_act_model_layers_87_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_69_mlp_up_proj_weight/std":0.03759765625,"train/tensor_act_model_layers_66_residual_post_mlp/max_abs":270,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_19_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_83_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_45_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_85_mlp/min":4,"train/tensor_act_model_layers_77_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_34_mlp_down_proj_weight/range":0.24755859375,"train/tensor_param_model_layers_52_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_8_residual_post_mlp/max":37.5,"train/tensor_act_model_layers_71_residual_pre_attn/range":6,"train/tensor_param_model_layers_52_mlp_down_proj_weight/min":-0.12451171875,"train/tensor_act_model_layers_7_residual_post_attn/range":2.5,"train/tensor_act_model_layers_51_mlp_up_proj/range":7.421875,"train/tensor_param_model_layers_10_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_74_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_80_mlp_up_proj_weight/min":-0.12890625,"train/tensor_act_model_layers_58_residual_post_attn/min":229,"train/tensor_act_model_layers_15_mlp/min":4,"train/tensor_grad_model_layers_12_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_45_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_67_residual_post_mlp/min":268,"train/tensor_act_model_layers_63_residual_pre_attn/norm":515525.21875,"train/tensor_act_model_layers_98_residual_pre_attn/range":6,"train/tensor_param_model_layers_58_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_92_mlp_up_proj_weight/max":0.1396484375,"train/tensor_param_model_layers_36_mlp_gate_proj_weight/range":0.2607421875,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_33_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_63_residual_post_mlp/mean":256,"train/tensor_act_model_layers_33_mlp_down_proj/min":14.3125,"train/tensor_act_model_layers_59_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_96_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_31_mlp_up_proj/range":7.1875,"train/tensor_act_model_layers_34_residual_post_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_77_mlp_gate_proj/mean":0.07177734375,"train/tensor_param_model_layers_41_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_43_mlp_gate_proj/norm":5361.52734375,"train/tensor_param_model_layers_7_mlp_gate_proj_weight/mean":-0.0002536773681640625,"train/tensor_param_model_layers_9_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_residual_pre_attn/max_abs":158,"train/tensor_act_model_layers_76_mlp_down_proj/max":16.375,"train/tensor_param_model_layers_17_mlp_up_proj_weight/min":-0.1162109375,"train/tensor_act_model_layers_87_residual_post_attn/max":350,"train/tensor_param_model_layers_82_mlp_gate_proj_weight/min":-0.13671875,"train/tensor_act_model_layers_6_residual_post_attn/mean":24,"train/tensor_grad_model_layers_4_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_15_mlp_up_proj_weight/std":0.0341796875,"train/tensor_act_model_layers_17_mlp_down_proj/mean":14.8125,"train/tensor_act_model_layers_90_mlp/norm":8192,"train/tensor_act_model_layers_1_residual_post_mlp/max":8.4375,"train/tensor_grad_model_layers_32_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_58_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_66_mlp_up_proj/min":-3.734375,"train/tensor_act_model_layers_93_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_22_mlp_up_proj/min":-4.125,"train/tensor_act_model_layers_98_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_mlp/mean":4,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_84_mlp_up_proj_weight/mean":0.001312255859375,"train/tensor_param_model_layers_21_mlp_up_proj_weight/range":0.25048828125,"train/tensor_param_model_layers_22_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_22_residual_post_mlp/min":90,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/std":0,"train/tensor_grad_model_layers_84_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_29_mlp_up_proj/range":6.71875,"train/tensor_grad_model_layers_40_mlp_up_proj_weight/norm":0,"_runtime":369,"train/tensor_param_model_layers_13_mlp_up_proj_weight/range":0.24951171875,"train/tensor_grad_model_layers_5_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_40_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_62_mlp_up_proj_weight/mean":-1.9311904907226562e-05,"train/tensor_act_model_layers_35_mlp_up_proj/norm":5330.75732421875,"train/tensor_act_model_layers_68_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_42_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_94_mlp_gate_proj/min":-4.34375,"train/tensor_act_model_layers_78_mlp_gate_proj/mean":-0.03662109375,"train/tensor_act_model_layers_37_mlp/min":4,"train/tensor_act_model_layers_65_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_63_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_25_mlp_up_proj/std":1.84375,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_62_mlp_down_proj/std":0.41796875,"train/tensor_param_model_layers_65_mlp_down_proj_weight/mean":-0.000400543212890625,"train/tensor_param_model_layers_45_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_grad_model_layers_71_mlp_down_proj_weight/norm":0,"train/tensor_act_model_layers_62_mlp_up_proj/max":3.796875,"train/tensor_act_model_layers_14_mlp_up_proj/max_abs":3.671875,"train/tensor_act_model_layers_62_residual_post_mlp/max_abs":254,"train/tensor_grad_model_layers_86_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_46_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_65_mlp_up_proj_weight/mean":0.00024127960205078125,"train/tensor_param_model_layers_95_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_residual_post_attn/std":0.8359375,"train/tensor_param_model_layers_92_mlp_down_proj_weight/range":0.2705078125,"train/tensor_act_model_layers_96_residual_post_attn/norm":786029.6875,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/range":0,"train/tensor_act_model_layers_65_residual_post_attn/max_abs":262,"train/tensor_act_model_layers_9_residual_post_mlp/std":0.373046875,"train/tensor_act_model_layers_46_mlp/norm":8192,"train/tensor_grad_model_layers_77_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_76_mlp_down_proj/min":14.1875,"train/tensor_act_model_layers_19_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_92_mlp_down_proj/std":0.5390625,"train/tensor_act_model_layers_87_residual_post_attn/min":344,"train/tensor_param_model_layers_93_mlp_gate_proj_weight/max_abs":0.1640625,"train/tensor_param_model_layers_22_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_97_mlp_up_proj/max_abs":4.34375,"train/tensor_act_model_layers_51_mlp/mean":4,"train/tensor_act_model_layers_33_mlp_down_proj/max":15.8125,"train/tensor_act_model_layers_71_mlp/range":0,"train/tensor_act_model_layers_34_mlp/range":0,"train/tensor_param_model_layers_87_mlp_up_proj_weight/max_abs":0.140625,"train/tensor_act_model_layers_21_residual_pre_attn/mean":84,"train/tensor_param_model_layers_21_mlp_down_proj_weight/range":0.23681640625,"train/tensor_grad_model_layers_57_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_21_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_56_mlp_up_proj_weight/max":0.134765625,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/range":0,"train/tensor_grad_model_layers_7_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_75_residual_pre_attn/max_abs":302,"train/tensor_grad_model_layers_83_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_26_mlp_down_proj/mean":14.875,"train/tensor_grad_model_layers_28_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_81_mlp_up_proj/max":3.765625,"train/tensor_act_model_layers_30_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_68_mlp/mean":4,"train/tensor_act_model_layers_33_residual_post_mlp/max":138,"train/tensor_param_model_layers_79_mlp_up_proj_weight/norm":7.21875,"train/tensor_act_model_layers_4_mlp_down_proj/std":0.48046875,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_36_mlp_up_proj/min":-3.53125,"train/tensor_act_model_layers_80_mlp/max_abs":4,"train/tensor_act_model_layers_63_residual_post_mlp/norm":523449.15625,"train/tensor_act_model_layers_50_mlp/max_abs":4,"train/tensor_act_model_layers_41_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_32_residual_post_mlp/max_abs":134,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_3_residual_post_mlp/max_abs":16.5,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/range":0.294921875,"train/tensor_act_model_layers_98_mlp_up_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_51_mlp_down_proj_weight/max":0.125,"train/tensor_act_model_layers_90_mlp/min":4,"train/tensor_act_model_layers_66_residual_pre_attn/std":0.7578125,"train/tensor_grad_model_layers_40_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_77_mlp/max_abs":4,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/range":0.24169921875,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/min":0,"train/tensor_param_model_layers_81_mlp_down_proj_weight/max":0.134765625,"train/tensor_param_model_layers_85_mlp_down_proj_weight/min":-0.1435546875,"train/tensor_grad_model_layers_87_mlp_gate_proj_weight/std":0,"train/tensor_param_model_layers_49_mlp_down_proj_weight/mean":-0.000759124755859375,"train/tensor_act_model_layers_43_mlp_gate_proj/max":3.734375,"train/tensor_act_model_layers_98_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_96_mlp_gate_proj/max":4.1875,"train/tensor_param_model_layers_76_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_48_mlp_down_proj/min":14.25,"train/tensor_param_model_layers_99_mlp_gate_proj_weight/norm":8.75,"train/tensor_act_model_layers_81_mlp_gate_proj/min":-3.65625,"train/tensor_grad_model_layers_9_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_24_mlp_down_proj/std":0.3203125,"train/tensor_grad_model_layers_47_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_15_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_65_mlp_down_proj_weight/mean":0,"train/tensor_act_model_layers_48_mlp_down_proj/max":16.125,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_82_residual_post_attn/std":0.7578125,"train/tensor_param_model_layers_72_mlp_down_proj_weight/mean":0.000568389892578125,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_67_mlp_gate_proj_weight/max":0.12890625,"train/tensor_act_model_layers_54_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_28_residual_pre_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_53_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/range":0,"train/tensor_act_model_layers_28_mlp/max":4,"train/tensor_act_model_layers_46_mlp_up_proj/max_abs":3.828125,"train/tensor_act_model_layers_73_mlp/mean":4,"train/tensor_act_model_layers_72_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_66_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_20_residual_post_mlp/std":0.58203125,"train/tensor_act_model_layers_37_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_92_mlp_up_proj/max":4.0625,"train/tensor_act_model_layers_85_residual_pre_attn/range":6,"train/tensor_act_model_layers_46_mlp_up_proj/mean":-0.1787109375,"train/tensor_act_model_layers_60_mlp/mean":4,"train/tensor_grad_model_layers_48_mlp_up_proj_weight/range":0,"train/tensor_param_model_layers_50_mlp_gate_proj_weight/range":0.255859375,"train/tensor_param_model_layers_98_mlp_gate_proj_weight/norm":8.6875,"train/tensor_act_model_layers_79_mlp_gate_proj/std":1.8125,"train/tensor_param_model_layers_87_mlp_down_proj_weight/range":0.263671875,"train/tensor_act_model_layers_51_mlp_gate_proj/max":3.375,"train/tensor_param_model_layers_22_mlp_down_proj_weight/max":0.125,"train/tensor_param_model_layers_29_mlp_down_proj_weight/range":0.24658203125,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_82_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_6_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_57_mlp_down_proj_weight/min":-0.125,"train/tensor_act_model_layers_34_mlp/mean":4,"train/tensor_act_model_layers_73_mlp_down_proj/std":0.384765625,"train/tensor_act_model_layers_36_residual_pre_attn/max":146,"train/tensor_act_model_layers_46_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_param_model_layers_6_mlp_up_proj_weight/range":0.23779296875,"train/tensor_grad_model_layers_53_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_90_mlp_down_proj/max_abs":17.125,"train/tensor_act_model_layers_42_residual_post_attn/max_abs":170,"train/tensor_act_model_layers_46_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_16_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_90_residual_post_attn/min":356,"train/tensor_param_model_layers_4_mlp_up_proj_weight/std":0.031494140625,"train/tensor_param_model_layers_43_mlp_gate_proj_weight/std":0.034423828125,"train/tensor_param_model_layers_10_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_17_mlp_down_proj_weight/max_abs":0.1220703125,"train/tensor_act_model_layers_31_mlp_gate_proj/max":3.6875,"train/tensor_param_model_layers_42_mlp_up_proj_weight/range":0.24365234375,"train/tensor_act_model_layers_80_mlp_down_proj/max":16.875,"train/tensor_act_model_layers_9_residual_post_mlp/range":3,"train/tensor_act_model_layers_76_residual_post_mlp/frac_near_user_limit":1,"train/tensor_grad_model_layers_26_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/std":0,"train/tensor_act_model_layers_83_residual_post_attn/norm":679533.9375,"train/tensor_act_model_layers_0_mlp_up_proj/max":5.625,"train/tensor_grad_model_layers_54_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_7_residual_post_mlp/norm":65482.046875,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_72_mlp_gate_proj_weight/std":0.037841796875,"train/tensor_param_model_layers_37_mlp_down_proj_weight/range":0.2509765625,"train/tensor_act_model_layers_19_residual_pre_attn/std":0.58203125,"train/tensor_act_model_layers_61_mlp_down_proj/max_abs":16.125,"train/tensor_act_model_layers_75_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_99_mlp_down_proj_weight/norm":8.75,"train/tensor_act_model_layers_52_mlp_up_proj/mean":0.125,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/max_abs":0.125,"train/tensor_act_model_layers_94_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_17_mlp/range":0,"train/tensor_act_model_layers_37_mlp_gate_proj/norm":5404.76953125,"train/tensor_act_model_layers_29_mlp_gate_proj/max":3.46875,"train/tensor_param_model_layers_62_mlp_gate_proj_weight/std":0.03564453125,"train/tensor_act_model_layers_24_residual_post_attn/range":4,"train/tensor_grad_model_layers_86_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_82_residual_post_mlp/max":334,"train/tensor_act_model_layers_99_mlp/norm":8192,"train/tensor_act_model_layers_92_mlp_gate_proj/max_abs":4.21875,"train/tensor_param_model_layers_32_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_57_mlp_up_proj_weight/max_abs":0.130859375,"train/tensor_param_model_layers_84_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_62_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_60_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_80_mlp_down_proj/range":2.625,"train/tensor_param_model_layers_90_mlp_down_proj_weight/std":0.044189453125,"train/tensor_act_model_layers_47_residual_post_mlp/frac_near_user_limit":1,"train/tensor_act_model_layers_13_residual_post_attn/range":3,"train/tensor_act_model_layers_64_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_39_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_9_mlp_up_proj_weight/range":0.24169921875,"train/tensor_param_model_layers_66_mlp_down_proj_weight/range":0.2587890625,"train/tensor_act_model_layers_90_residual_post_mlp/min":360,"train/tensor_param_model_layers_15_mlp_gate_proj_weight/min":-0.1279296875,"train/tensor_act_model_layers_23_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_22_mlp_down_proj/std":0.388671875,"train/tensor_param_model_layers_63_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_98_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_8_mlp_gate_proj_weight/range":0.24560546875,"train/tensor_act_model_layers_19_residual_post_attn/max_abs":78,"train/tensor_grad_model_layers_44_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_11_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_38_mlp_up_proj_weight/min":-0.1259765625,"train/tensor_act_model_layers_8_mlp_down_proj/std":0.3671875,"train/tensor_act_model_layers_49_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/norm":7.65625,"train/tensor_param_model_layers_85_mlp_up_proj_weight/range":0.275390625,"train/tensor_act_model_layers_83_residual_post_mlp/max":338,"train/tensor_act_model_layers_75_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_56_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_13_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_92_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_9_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_mlp_down_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_45_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_76_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_62_residual_pre_attn/frac_near_user_limit":1,"train/tensor_param_model_layers_43_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_33_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_86_residual_post_attn/std":0.7578125,"train/tensor_act_model_layers_83_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_46_mlp_down_proj_weight/min":-0.12451171875,"train/tensor_grad_model_layers_8_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_63_mlp_up_proj_weight/mean":0,"train/tensor_act_model_layers_33_mlp_down_proj/range":1.5,"train/tensor_grad_model_layers_7_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_27_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_15_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_3_mlp_up_proj_weight/mean":-0.000606536865234375,"train/tensor_grad_model_layers_39_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_63_mlp_gate_proj_weight/mean":0,"train/tensor_param_model_layers_5_mlp_down_proj_weight/max":0.1201171875,"train/tensor_act_model_layers_71_mlp_up_proj/max":3.5625,"train/tensor_param_model_layers_63_mlp_up_proj_weight/max":0.125,"train/tensor_act_model_layers_65_mlp_up_proj/mean":0.0311279296875,"train/tensor_param_model_layers_79_mlp_up_proj_weight/max_abs":0.1494140625,"train/tensor_act_model_layers_90_residual_post_attn/max":362,"train/tensor_act_model_layers_8_residual_post_mlp/min":34.5,"train/tensor_act_model_layers_63_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_4_mlp_down_proj/max_abs":16.5,"train/tensor_grad_model_layers_9_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_57_residual_post_mlp/norm":474629.03125,"train/tensor_act_model_layers_80_mlp_down_proj/min":14.25,"train/tensor_act_model_layers_69_residual_post_mlp/std":0.7578125,"train/tensor_act_model_layers_58_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_36_mlp_up_proj_weight/norm":6.3125,"train/tensor_grad_model_layers_71_mlp_gate_proj_weight/max":0,"train/tensor_grad_model_layers_52_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_49_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_60_mlp_gate_proj/min":-3.578125,"train/tensor_grad_model_layers_81_mlp_gate_proj_weight/min":0,"train/tensor_param_model_layers_78_mlp_gate_proj_weight/min":-0.130859375,"train/tensor_param_model_layers_33_mlp_down_proj_weight/max":0.1259765625,"train/tensor_act_model_layers_84_mlp_gate_proj/min":-4,"train/tensor_act_model_layers_18_mlp_gate_proj/range":7.140625,"train/tensor_param_model_layers_76_mlp_up_proj_weight/norm":7.0625,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/min":-0.1259765625,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_51_mlp/norm":8192,"train/tensor_act_model_layers_19_residual_post_mlp/range":4,"train/tensor_act_model_layers_86_mlp_gate_proj/max":3.875,"train/tensor_act_model_layers_23_residual_post_attn/range":4,"train/tensor_act_model_layers_6_mlp_up_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_37_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_15_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_6_mlp_gate_proj/max_abs":3.453125,"train/tensor_act_model_layers_35_mlp_up_proj/std":1.8359375,"train/tensor_grad_model_layers_48_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_93_mlp_up_proj/mean":0.1884765625,"train/tensor_act_model_layers_64_mlp_gate_proj/max_abs":4.03125,"train/tensor_act_model_layers_27_residual_post_mlp/range":4,"train/tensor_act_model_layers_82_residual_post_mlp/mean":332,"train/tensor_grad_model_layers_38_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_98_mlp_gate_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_99_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_30_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_14_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_18_mlp_up_proj/std":1.8203125,"train/tensor_act_model_layers_30_mlp_down_proj/norm":30402.298828125,"train/tensor_grad_model_layers_67_mlp_down_proj_weight/max_abs":0,"train/tensor_act_model_layers_76_mlp/frac_near_user_limit":0,"train/tensor_grad_model_layers_25_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_residual_post_mlp/mean":396,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_2_mlp_down_proj_weight/max":0,"train/tensor_grad_model_layers_48_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_12_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_54_residual_pre_attn/range":5,"train/tensor_param_model_layers_76_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_24_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_16_mlp/std":0,"train/tensor_param_model_layers_60_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_34_residual_pre_attn/max_abs":138,"train/tensor_act_model_layers_81_residual_post_mlp/max_abs":330,"train/tensor_grad_model_layers_43_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_42_residual_post_mlp/range":5,"train/tensor_act_model_layers_82_residual_post_attn/norm":671341.9375,"train/tensor_grad_model_layers_16_mlp_gate_proj_weight/norm":0,"train/tensor_grad_model_layers_71_mlp_up_proj_weight/mean":0,"train/tensor_grad_model_layers_76_mlp_up_proj_weight/range":0,"train/tensor_grad_model_layers_21_mlp_gate_proj_weight/max":0,"train/tensor_param_model_layers_38_mlp_gate_proj_weight/range":0.25,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_10_mlp_up_proj_weight/std":0.033935546875,"train/tensor_grad_model_layers_96_mlp_gate_proj_weight/norm":0,"train/tensor_act_model_layers_86_residual_pre_attn/norm":704109.875,"train/tensor_param_model_layers_19_mlp_up_proj_weight/max_abs":0.1220703125,"train/tensor_grad_model_layers_80_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_4_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_53_mlp/std":0,"train/tensor_grad_model_layers_68_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_38_mlp_up_proj/std":1.8671875,"train/tensor_act_model_layers_14_residual_post_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_19_mlp_gate_proj/max":3.640625,"train/tensor_act_model_layers_13_residual_pre_attn/std":0.4296875,"train/tensor_param_model_layers_98_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_70_mlp_up_proj_weight/max_abs":0.1337890625,"train/tensor_grad_model_layers_62_mlp_up_proj_weight/std":0,"train/tensor_grad_model_layers_45_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_60_mlp_down_proj_weight/range":0.248046875,"train/tensor_act_model_layers_42_residual_pre_attn/max":170,"train/tensor_param_model_layers_7_mlp_up_proj_weight/norm":6.03125,"train/tensor_act_model_layers_68_residual_pre_attn/std":0.7578125,"train/tensor_param_model_layers_46_mlp_up_proj_weight/mean":-0.00139617919921875,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/max":0.12255859375,"train/tensor_param_model_layers_90_mlp_up_proj_weight/mean":-0.000514984130859375,"train/tensor_act_model_layers_85_mlp/mean":4,"train/tensor_act_model_layers_71_mlp_gate_proj/range":7.171875,"train/tensor_act_model_layers_8_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_param_model_layers_13_mlp_down_proj_weight/max_abs":0.12353515625,"train/tensor_act_model_layers_58_residual_post_mlp/range":5,"train/tensor_param_model_layers_66_mlp_down_proj_weight/max":0.1298828125,"train/tensor_act_model_layers_7_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_75_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_94_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_35_mlp_gate_proj_weight/min":-0.12060546875,"train/tensor_act_model_layers_67_mlp_gate_proj/std":1.84375,"train/tensor_param_model_layers_93_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_95_mlp_down_proj_weight/norm":8.5625,"train/tensor_act_model_layers_41_mlp_gate_proj/max":3.453125,"train/tensor_act_model_layers_37_mlp_up_proj/std":1.890625,"train/tensor_act_model_layers_40_mlp_up_proj/max":3.234375,"train/tensor_act_model_layers_69_mlp_up_proj/std":1.859375,"train/tensor_act_model_layers_86_residual_post_mlp/mean":348,"train/tensor_act_model_layers_95_residual_post_mlp/max":386,"train/tensor_grad_model_layers_18_mlp_down_proj_weight/range":0,"train/tensor_param_model_layers_96_mlp_gate_proj_weight/max_abs":0.1513671875,"train/tensor_grad_model_layers_31_mlp_gate_proj_weight/mean":0,"train/tensor_act_model_layers_58_residual_post_attn/max_abs":234,"train/tensor_param_model_layers_20_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_84_mlp/range":0,"train/tensor_act_model_layers_77_mlp/min":4,"train/tensor_act_model_layers_8_residual_post_attn/min":30.375,"train/tensor_grad_model_layers_50_mlp_gate_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_98_residual_pre_attn/norm":802413.6875,"train/tensor_grad_model_layers_50_mlp_up_proj_weight/min":0,"train/tensor_act_model_layers_88_residual_post_mlp/max":358,"train/tensor_grad_model_layers_88_mlp_down_proj_weight/min":0,"train/tensor_grad_model_layers_59_mlp_up_proj_weight/std":0,"train/tensor_act_model_layers_59_mlp_down_proj/frac_near_user_limit":1,"train/tensor_param_model_layers_47_mlp_up_proj_weight/range":0.25,"train/tensor_param_model_layers_54_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_param_model_layers_87_mlp_gate_proj_weight/max":0.138671875,"train/tensor_act_model_layers_1_residual_post_mlp/min":7.5,"train/tensor_act_model_layers_82_residual_pre_attn/norm":671341.9375,"train/tensor_act_model_layers_28_mlp_gate_proj/min":-3.5,"train/tensor_grad_model_layers_21_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_43_mlp/frac_near_user_limit":0,"train/tensor_param_model_layers_74_mlp_gate_proj_weight/mean":0.0003185272216796875,"train/tensor_param_model_layers_85_mlp_up_proj_weight/mean":0.001312255859375,"train/tensor_act_model_layers_27_mlp/max":4,"train/tensor_grad_model_layers_0_mlp_gate_proj_weight/min":-364,"train/tensor_act_model_layers_26_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_act_model_layers_4_residual_post_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_87_mlp/std":0,"train/tensor_act_model_layers_38_mlp/frac_near_user_limit":0,"train/tensor_act_model_layers_55_residual_post_mlp/frac_near_user_limit":1,"train/tensor_param_model_layers_42_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_0_residual_post_mlp/mean":4,"train/tensor_param_model_layers_86_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_15_mlp_down_proj_weight/min":-0.12158203125,"train/tensor_act_model_layers_54_residual_post_attn/max_abs":218,"train/tensor_param_model_layers_20_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_27_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_74_mlp_up_proj/norm":5275.708984375,"train/tensor_act_model_layers_65_mlp/norm":8192,"train/tensor_act_model_layers_52_mlp_gate_proj/frac_near_user_limit":0,"train/tensor_grad_model_layers_62_mlp_gate_proj_weight/std":0,"train/tensor_grad_model_layers_93_mlp_gate_proj_weight/range":0,"train/tensor_grad_model_layers_34_mlp_down_proj_weight/max_abs":0,"train/tensor_grad_model_layers_23_mlp_down_proj_weight/min":0,"train/tensor_act_model_layers_14_residual_post_mlp/range":3,"train/tensor_grad_model_layers_24_mlp_up_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_0_residual_pre_attn/max_abs":0.275390625,"train/tensor_act_model_layers_92_residual_pre_attn/frac_near_user_limit":1,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/norm":0,"train/tensor_act_model_layers_74_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_72_mlp_down_proj/range":2.875,"train/tensor_act_model_layers_57_mlp_down_proj/range":1.8125,"train/tensor_grad_model_layers_44_mlp_down_proj_weight/max":0,"train/tensor_param_model_layers_81_mlp_gate_proj_weight/max_abs":0.1591796875,"train/tensor_act_model_layers_35_mlp_down_proj/range":2.25,"train/tensor_param_model_layers_89_mlp_gate_proj_weight/norm":8.1875,"train/tensor_act_model_layers_71_mlp_down_proj/mean":14.9375,"train/tensor_param_model_layers_56_mlp_gate_proj_weight/max_abs":0.1298828125,"train/tensor_grad_model_layers_89_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_43_residual_post_mlp/min":173,"train/tensor_act_model_layers_52_mlp_gate_proj/norm":5451.28369140625,"train/tensor_act_model_layers_73_residual_post_attn/frac_near_user_limit":1,"train/tensor_act_model_layers_62_residual_post_mlp/norm":515525.21875,"train/tensor_act_model_layers_81_mlp_gate_proj/mean":0.13671875,"train/tensor_act_model_layers_99_mlp/mean":4,"train/tensor_param_model_layers_22_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_16_mlp_down_proj_weight/range":0.24365234375,"train/tensor_act_model_layers_9_residual_pre_attn/mean":36,"train/tensor_act_model_layers_64_residual_post_attn/range":6,"train/tensor_act_model_layers_61_mlp/mean":4,"train/tensor_act_model_layers_21_mlp_down_proj/std":0.41015625,"train/tensor_act_model_layers_24_residual_post_attn/min":94,"train/tensor_act_model_layers_27_mlp_down_proj/mean":14.875,"train/tensor_param_model_layers_73_mlp_gate_proj_weight/max_abs":0.1279296875,"train/tensor_act_model_layers_67_mlp_up_proj/frac_near_user_limit":0,"train/tensor_act_model_layers_62_mlp_up_proj/mean":-0.002288818359375,"train/tensor_act_model_layers_17_mlp/norm":8192,"train/tensor_param_model_layers_9_mlp_up_proj_weight/max":0.1220703125,"train/tensor_grad_model_layers_68_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_86_residual_post_mlp/norm":712301.875,"train/tensor_act_model_layers_70_residual_post_mlp/min":280,"train/tensor_act_model_layers_87_mlp_gate_proj/std":1.8515625,"train/tensor_act_model_layers_89_mlp/std":0,"train/tensor_param_model_layers_56_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_53_mlp_down_proj/frac_near_user_limit":1,"train/tensor_act_model_layers_73_residual_pre_attn/mean":292,"train/tensor_act_model_layers_89_residual_pre_attn/std":0.7578125,"train/tensor_act_model_layers_13_mlp/frac_near_dtype_limit":0,"train/tensor_act_model_layers_40_residual_pre_attn/mean":160,"train/tensor_act_model_layers_58_mlp/max_abs":4,"train/tensor_act_model_layers_58_mlp_up_proj/max_abs":4.1875,"train/tensor_act_model_layers_19_mlp_down_proj/range":1.75,"train/tensor_grad_model_layers_1_mlp_down_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_45_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_46_mlp_gate_proj/min":-3.59375,"train/tensor_act_model_layers_84_residual_pre_attn/frac_near_dtype_limit":0,"train/tensor_act_model_layers_26_mlp_down_proj/range":2.1875,"train/tensor_act_model_layers_9_residual_post_attn/max":37.5,"train/tensor_param_model_layers_51_mlp_up_proj_weight/range":0.24267578125,"train/tensor_param_model_layers_28_mlp_up_proj_weight/max":0.11767578125,"train/tensor_param_model_layers_88_mlp_gate_proj_weight/max":0.1474609375,"train/tensor_grad_model_layers_94_mlp_up_proj_weight/norm":0,"train/tensor_param_model_layers_56_mlp_down_proj_weight/max_abs":0.1298828125,"train/tensor_param_model_layers_51_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_61_mlp_up_proj/max_abs":3.875,"train/tensor_grad_model_layers_36_mlp_up_proj_weight/std":0,"train/tensor_param_model_layers_58_mlp_gate_proj_weight/min":-0.125,"train/tensor_param_model_layers_34_mlp_down_proj_weight/max":0.12255859375,"train/tensor_act_model_layers_93_mlp_down_proj/norm":30717.166015625,"train/tensor_grad_model_layers_30_mlp_up_proj_weight/max_abs":0,"train/tensor_act_model_layers_39_residual_post_attn/mean":156,"train/tensor_param_model_layers_19_mlp_gate_proj_weight/max":0.125,"train/tensor_grad_model_layers_95_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_75_mlp_up_proj_weight/range":0,"train/tensor_act_model_layers_52_mlp_up_proj/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_3_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_86_mlp_up_proj/range":7.84375,"train/tensor_param_model_layers_65_mlp_gate_proj_weight/range":0.24951171875,"train/tensor_param_model_layers_61_mlp_gate_proj_weight/max":0.12890625,"train/tensor_grad_model_layers_36_mlp_gate_proj_weight/max_abs":0,"train/tensor_act_model_layers_79_mlp_up_proj/mean":-0.060302734375,"train/tensor_act_model_layers_77_residual_post_mlp/max_abs":314,"train/tensor_param_model_layers_42_mlp_up_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_82_mlp_up_proj/max":3.578125,"train/tensor_act_model_layers_0_residual_pre_attn/norm":187.48826599121094,"train/tensor_act_model_layers_82_residual_post_attn/range":6,"train/tensor_param_model_layers_26_mlp_down_proj_weight/range":0.24951171875,"train/tensor_act_model_layers_91_residual_post_mlp/min":364,"train/tensor_act_model_layers_1_residual_post_mlp/norm":16381.2919921875,"train/tensor_act_model_layers_93_residual_pre_attn/max":374,"train/tensor_act_model_layers_93_mlp_gate_proj/range":8.375,"train/tensor_act_model_layers_45_residual_post_attn/mean":180,"train/tensor_param_model_layers_11_mlp_gate_proj_weight/range":0.2392578125,"train/tensor_act_model_layers_99_mlp/min":4,"train/tensor_act_model_layers_29_mlp_gate_proj/norm":5375.65966796875,"train/tensor_grad_model_layers_26_mlp_gate_proj_weight/mean":0,"train/tensor_grad_model_layers_55_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_87_mlp_down_proj_weight/max":0.1328125,"train/tensor_param_model_layers_91_mlp_gate_proj_weight/std":0.0458984375,"train/tensor_act_model_layers_71_residual_post_mlp/min":284,"train/tensor_grad_model_layers_98_mlp_gate_proj_weight/norm":0,"train/tensor_param_model_layers_6_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_param_model_layers_44_mlp_gate_proj_weight/std":0.034912109375,"train/tensor_act_model_layers_1_residual_pre_attn/min":3.546875,"train/tensor_act_model_layers_9_mlp_down_proj/mean":14.9375,"train/tensor_act_model_layers_82_mlp_gate_proj/std":1.8203125,"train/tensor_act_model_layers_35_residual_post_mlp/max":146,"train/tensor_grad_model_layers_69_mlp_gate_proj_weight/max":0,"train/tensor_act_model_layers_69_mlp_down_proj/std":0.380859375,"train/tensor_act_model_layers_57_residual_pre_attn/norm":466404.96875,"train/tensor_grad_model_layers_47_mlp_up_proj_weight/norm":0,"train/tensor_grad_model_layers_53_mlp_down_proj_weight/std":0,"train/tensor_act_model_layers_62_residual_post_mlp/mean":252,"train/tensor_act_model_layers_1_residual_post_attn/range":0.921875,"train/tensor_act_model_layers_95_residual_pre_attn/max_abs":382,"train/tensor_param_model_layers_97_mlp_gate_proj_weight/min":-0.1572265625,"train/tensor_act_model_layers_6_mlp_up_proj/max_abs":3.5,"train/tensor_act_model_layers_1_mlp/range":0,"train/tensor_act_model_layers_24_residual_post_attn/max":98,"train/tensor_act_model_layers_19_mlp_up_proj/range":7.3125,"train/tensor_act_model_layers_87_mlp_gate_proj/max_abs":4.15625,"train/tensor_act_model_layers_55_mlp_up_proj/mean":0.09521484375,"train/tensor_act_model_layers_26_mlp_gate_proj/std":1.8203125,"train/tensor_act_model_layers_99_mlp_down_proj/std":0.734375,"train/tensor_act_model_layers_9_mlp_gate_proj/mean":0.3125,"train/tensor_act_model_layers_13_mlp_up_proj/min":-3.78125,"train/tensor_act_model_layers_94_mlp_up_proj/min":-4.28125,"train/tensor_param_model_layers_26_mlp_down_proj_weight/max_abs":0.125,"train/tensor_grad_model_layers_9_mlp_down_proj_weight/max":0,"train/tensor_act_model_layers_57_residual_post_attn/range":5,"train/tensor_act_model_layers_85_mlp_gate_proj/mean":0.037109375,"train/tensor_grad_model_layers_60_mlp_gate_proj_weight/frac_near_user_limit":0,"train/tensor_act_model_layers_5_mlp/max_abs":4,"train/tensor_grad_model_layers_46_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_act_model_layers_51_mlp_gate_proj/mean":0.13671875,"train/tensor_param_model_layers_48_mlp_down_proj_weight/std":0.03466796875,"train/tensor_grad_model_layers_46_mlp_up_proj_weight/frac_near_user_limit":0,"train/tensor_grad_model_layers_82_mlp_gate_proj_weight/min":0,"train/tensor_act_model_layers_41_mlp_up_proj/range":6.859375,"train/tensor_grad_model_layers_34_mlp_up_proj_weight/min":0,"train/tensor_grad_model_layers_32_mlp_down_proj_weight/frac_near_dtype_limit":0,"train/tensor_grad_model_layers_93_mlp_down_proj_weight/max":0} \ No newline at end of file diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-core.log b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-core.log index 29193ff216a2dfc14b37fa63f5dd3013ce8da0ee..59acf96f4c751e74e9217cb996ef31c503e3590c 100644 --- a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-core.log +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-core.log @@ -68,3 +68,10 @@ {"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} {"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-internal.log b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-internal.log index 0872ecab6e4a2037597cff9742e8e0a2318e6237..c987e84302d55a45e10f1ecdcf813a1e4b2f94c4 100644 --- a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-internal.log +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug-internal.log @@ -37,3 +37,37 @@ {"time":"2026-08-14T21:24:40.427030871Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} {"time":"2026-08-14T21:24:54.460542104Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":2,"events_offset":19,"events_lines":2,"console_offset":35,"console_lines":9} {"time":"2026-08-14T21:24:55.385800453Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:09.459223532Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":2,"events_offset":21,"events_lines":2,"console_offset":41,"console_lines":1} +{"time":"2026-08-14T21:25:10.338668442Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:24.454676952Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":2,"events_offset":23,"events_lines":2,"console_offset":41,"console_lines":1} +{"time":"2026-08-14T21:25:25.245490507Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:39.467654857Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":26,"history_lines":2,"events_offset":25,"events_lines":2,"console_offset":44,"console_lines":11} +{"time":"2026-08-14T21:25:40.389907102Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:25:54.463154808Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":28,"history_lines":2,"events_offset":27,"events_lines":2,"console_offset":51,"console_lines":1} +{"time":"2026-08-14T21:25:55.24970084Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:26:09.464095349Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":30,"history_lines":2,"events_offset":29,"events_lines":2,"console_offset":55,"console_lines":9} +{"time":"2026-08-14T21:26:10.391736659Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:26:24.458991962Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":32,"history_lines":3,"events_offset":31,"events_lines":2,"console_offset":61,"console_lines":1} +{"time":"2026-08-14T21:26:25.382996244Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:26:39.461578547Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":35,"history_lines":3,"events_offset":33,"events_lines":2,"console_offset":64,"console_lines":11} +{"time":"2026-08-14T21:26:40.376873392Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:26:54.461092479Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":38,"history_lines":2,"events_offset":35,"events_lines":2,"console_offset":71,"console_lines":1} +{"time":"2026-08-14T21:26:55.499948157Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:27:09.464018745Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":40,"history_lines":3,"events_offset":37,"events_lines":2,"console_offset":75,"console_lines":10} +{"time":"2026-08-14T21:27:10.348585635Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:27:24.46119055Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":43,"history_lines":3,"events_offset":39,"events_lines":2,"console_offset":81,"console_lines":1} +{"time":"2026-08-14T21:27:25.337127716Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:27:39.459568913Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":46,"history_lines":2,"events_offset":41,"events_lines":2,"console_offset":85,"console_lines":10} +{"time":"2026-08-14T21:27:40.420961456Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:27:54.460221415Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":48,"history_lines":2,"events_offset":43,"events_lines":2,"console_offset":91,"console_lines":1} +{"time":"2026-08-14T21:27:55.378108654Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:09.454435714Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":50,"history_lines":2,"events_offset":45,"events_lines":2,"console_offset":91,"console_lines":1} +{"time":"2026-08-14T21:28:10.354369244Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:19.516181311Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"} +{"time":"2026-08-14T21:28:19.529178819Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":52,"history_lines":1,"events_offset":47,"events_lines":1,"console_offset":95,"console_lines":16,"uploaded_len":3,"complete":true,"exit_code":0} +{"time":"2026-08-14T21:28:20.359553197Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:20.360958219Z","level":"INFO","msg":"handler: operation stats","stats":{}} +{"time":"2026-08-14T21:28:20.40038758Z","level":"INFO","msg":"stream: finishing up"} +{"time":"2026-08-14T21:28:20.400431303Z","level":"INFO","msg":"handler: closed"} +{"time":"2026-08-14T21:28:20.400559672Z","level":"INFO","msg":"sender: closed"} +{"time":"2026-08-14T21:28:20.400565215Z","level":"INFO","msg":"stream: all finished"} diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug.log b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug.log index ed65004b55b7662adc72c954df1af62c76a902fa..de9a6f9f563dc0de6533c3c8ad666db6620763ea 100644 --- a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug.log +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/logs/debug.log @@ -18,3 +18,8 @@ config: {'_wandb': {}} 2026-08-14 21:22:09,429 INFO MainThread:936813 [wandb_run.py:_config_callback():1346] config_cb None None {'transformers_version': '5.16.0.dev0', 'architectures': None, 'output_hidden_states': False, 'return_dict': True, 'dtype': None, 'chunk_size_feed_forward': 0, 'is_encoder_decoder': False, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'problem_type': None, 'vocab_size': 4096, 'hidden_size': 128, 'intermediate_size': 256, 'num_hidden_layers': 100, 'num_attention_heads': 4, 'num_key_value_heads': 4, 'hidden_act': 'silu', 'max_position_embeddings': 512, 'initializer_range': 0.02, 'rms_norm_eps': 1e-06, 'use_cache': False, 'pad_token_id': 0, 'bos_token_id': 1, 'eos_token_id': 2, 'pretraining_tp': 1, 'tie_word_embeddings': True, 'rope_parameters': {'rope_theta': 10000.0, 'rope_type': 'default'}, 'attention_bias': False, 'attention_dropout': 0.0, 'mlp_bias': False, 'head_dim': 32, '_name_or_path': '', 'tokenizer_name': 'w-ahmad/tiny-stories-tokenizer', 'mlp_type': 'glu', 'activation': 'waleed10', 'waleed_beta': 4, 'model_type': 'tiny_llama', 'output_attentions': False, 'output_dir': 'out/glu-waleed10-100L_trash_run', 'per_device_train_batch_size': 64, 'num_train_epochs': 1, 'max_steps': 3500, 'learning_rate': 9e-05, 'lr_scheduler_type': 'constant_with_warmup', 'lr_scheduler_kwargs': None, 'warmup_steps': 50, 'optim': 'adamw_torch_fused', 'optim_args': None, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'optim_target_modules': None, 'gradient_accumulation_steps': 1, 'average_tokens_across_devices': True, 'max_grad_norm': 1.0, 'label_smoothing_factor': 0.0, 'bf16': True, 'fp16': False, 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': None, 'gradient_checkpointing': False, 'gradient_checkpointing_kwargs': None, 'torch_compile': False, 'torch_compile_backend': None, 'torch_compile_mode': None, 'use_liger_kernel': False, 'liger_kernel_config': None, 'neftune_noise_alpha': None, 'torch_empty_cache_steps': None, 'auto_find_batch_size': False, 'logging_strategy': 'steps', 'logging_steps': 20, 'logging_first_step': False, 'log_on_each_node': True, 'logging_nan_inf_filter': True, 'include_num_input_tokens_seen': 'no', 'log_level': 'passive', 'log_level_replica': 'warning', 'disable_tqdm': False, 'report_to': ['wandb'], 'run_name': 'LM-glu-waleed10-100L-16.9M-20260814-212207', 'project': 'huggingface', 'trackio_space_id': None, 'trackio_bucket_id': None, 'trackio_static_space_id': None, 'eval_strategy': 'steps', 'eval_steps': 2498, 'eval_delay': 0, 'per_device_eval_batch_size': 1024, 'prediction_loss_only': False, 'eval_on_start': False, 'eval_do_concat_batches': True, 'eval_use_gather_object': False, 'eval_accumulation_steps': None, 'include_for_metrics': [], 'batch_eval_metrics': False, 'save_only_model': False, 'save_strategy': 'steps', 'save_steps': 50, 'save_on_each_node': False, 'save_total_limit': None, 'enable_jit_checkpoint': False, 'push_to_hub': False, 'hub_token': '', 'hub_private_repo': None, 'hub_model_id': 'w-ahmad/6L-glu-waleed10-100L', 'hub_strategy': 'every_save', 'hub_always_push': False, 'hub_revision': None, 'load_best_model_at_end': False, 'metric_for_best_model': None, 'greater_is_better': None, 'ignore_data_skip': False, 'restore_callback_states_from_checkpoint': False, 'full_determinism': False, 'seed': 42, 'data_seed': 42, 'use_cpu': False, 'accelerator_config': {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}, 'parallelism_config': None, 'dataloader_drop_last': False, 'dataloader_num_workers': 0, 'dataloader_pin_memory': True, 'dataloader_persistent_workers': False, 'dataloader_prefetch_factor': None, 'dataloader_multiprocessing_context': None, 'dataloader_in_order': True, 'remove_unused_columns': False, 'label_names': None, 'train_sampling_strategy': 'random', 'length_column_name': 'length', 'ddp_find_unused_parameters': None, 'ddp_bucket_cap_mb': None, 'ddp_broadcast_buffers': None, 'ddp_static_graph': None, 'ddp_backend': None, 'ddp_timeout': 1800, 'fsdp': None, 'fsdp_config': None, 'deepspeed': None, 'debug': [], 'skip_memory_metrics': True, 'do_train': False, 'do_eval': True, 'do_predict': False, 'resume_from_checkpoint': None, 'local_rank': -1} 2026-08-14 21:22:09,434 INFO MainThread:936813 [wandb_config.py:__setitem__():155] [no run ID] config set model/num_parameters = 16934016 - > 2026-08-14 21:22:09,434 INFO MainThread:936813 [wandb_run.py:_config_callback():1346] config_cb model/num_parameters 16934016 None +2026-08-14 21:28:18,726 INFO MainThread:936813 [wandb_run.py:_finish():2383] finishing run deepnevro-deepnevro/research7/n6fjpg9o +2026-08-14 21:28:18,726 INFO MainThread:936813 [wandb_run.py:_atexit_cleanup():2588] got exitcode: 0 +2026-08-14 21:28:18,727 INFO MainThread:936813 [wandb_run.py:_restore():2570] restore +2026-08-14 21:28:18,727 INFO MainThread:936813 [wandb_run.py:_restore():2576] restore done +2026-08-14 21:28:20,399 INFO MainThread:936813 [wandb_run.py:_footer_sync_info():3993] logging synced files diff --git a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/run-n6fjpg9o.wandb b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/run-n6fjpg9o.wandb index dfe328bbb82d26eb2dd59ea3da1ee3f03fbe4253..6ee3a628f481df18f18ff2b40dcfc888d21cc08d 100644 --- a/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/run-n6fjpg9o.wandb +++ b/zain/Activation/wandb/run-20260814_212208-n6fjpg9o/run-n6fjpg9o.wandb @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:e0184c598b27cc4154a7f934240743bb9d6d7505a18ca9a86ea5a3eeebfdaf07 -size 22315008 +oid sha256:270aeec28f07f1e18ac8153551f79ad41dae1453006422476b3eba2307f8afbc +size 51292143 diff --git a/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/files/config.yaml b/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/files/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..849cff8a24f56e4c23e6591e5aafba14433ae2cc --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/files/config.yaml @@ -0,0 +1,438 @@ +_name_or_path: + value: "" +_wandb: + value: + cli_version: 0.28.1 + e: + yu21vcw1fqgtt12osjsu76tw34s21eqd: + args: + - --config + - /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline.yaml + - --variants + - glu-waleed10 + - glu-silu-waleed10 + - mlp-waleed10 + - mlp-silu-waleed10 + - glu-waleed + - glu-situglu_low + - glu-waleedglu_low + codePath: sweep.py + codePathLocal: sweep.py + cpu_count: 112 + cpu_count_logical: 224 + cudaVersion: "12.4" + disk: + /: + total: "1560765693952" + used: "706489905152" + email: deepnevro@gmail.com + executable: /mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python + git: + commit: 3fb47c6943d6927eafc39bd399a3d3a520bae74a + remote: https://github.com/w-ahmad1a10/Activation.git + gpu: NVIDIA H100 80GB HBM3 + gpu_count: 8 + gpu_nvidia: + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-39c684a5-fde6-83d7-1663-0859795881ae + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-132944c4-b689-2b5f-89a4-d730401677ab + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bc6c3e3c-9b90-09ca-c034-774961847c54 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea + host: deeplens-k3s-node1 + memory: + total: "2164089937920" + os: Linux-5.15.0-126-generic-x86_64-with-glibc2.35 + program: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py + python: CPython 3.11.15 + root: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation + startedAt: "2026-08-14T21:28:14.182237Z" + writerId: yu21vcw1fqgtt12osjsu76tw34s21eqd + m: + - "1": train/global_step + "6": + - 3 + "7": [] + - "2": '*' + "5": 1 + "6": + - 1 + "7": [] + python_version: 3.11.15 + t: + "1": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "2": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "3": + - 2 + - 7 + - 13 + - 19 + - 66 + "4": 3.11.15 + "5": 0.28.1 + "6": 5.16.0.dev0 + "9": + "1": transformers_trainer + "12": 0.28.1 + "13": linux-x86_64 +accelerator_config: + value: + dispatch_batches: null + even_batches: true + gradient_accumulation_kwargs: null + non_blocking: false + split_batches: false + use_seedable_sampler: true +activation: + value: waleed10 +adam_beta1: + value: 0.9 +adam_beta2: + value: 0.999 +adam_epsilon: + value: 1e-08 +architectures: + value: null +attention_bias: + value: false +attention_dropout: + value: 0 +auto_find_batch_size: + value: false +average_tokens_across_devices: + value: true +batch_eval_metrics: + value: false +bf16: + value: true +bf16_full_eval: + value: false +bos_token_id: + value: 1 +chunk_size_feed_forward: + value: 0 +data_seed: + value: 42 +dataloader_drop_last: + value: false +dataloader_in_order: + value: true +dataloader_multiprocessing_context: + value: null +dataloader_num_workers: + value: 0 +dataloader_persistent_workers: + value: false +dataloader_pin_memory: + value: true +dataloader_prefetch_factor: + value: null +ddp_backend: + value: null +ddp_broadcast_buffers: + value: null +ddp_bucket_cap_mb: + value: null +ddp_find_unused_parameters: + value: null +ddp_static_graph: + value: null +ddp_timeout: + value: 1800 +debug: + value: [] +deepspeed: + value: null +disable_tqdm: + value: false +do_eval: + value: true +do_predict: + value: false +do_train: + value: false +dtype: + value: null +enable_jit_checkpoint: + value: false +eos_token_id: + value: 2 +eval_accumulation_steps: + value: null +eval_delay: + value: 0 +eval_do_concat_batches: + value: true +eval_on_start: + value: false +eval_steps: + value: 60000 +eval_strategy: + value: steps +eval_use_gather_object: + value: false +fp16: + value: false +fp16_full_eval: + value: false +fsdp: + value: null +fsdp_config: + value: null +full_determinism: + value: false +gradient_accumulation_steps: + value: 1 +gradient_checkpointing: + value: false +gradient_checkpointing_kwargs: + value: null +greater_is_better: + value: null +head_dim: + value: 32 +hidden_act: + value: silu +hidden_size: + value: 128 +hub_always_push: + value: false +hub_model_id: + value: w-ahmad/A-glu-waleed10-21L +hub_private_repo: + value: null +hub_revision: + value: null +hub_strategy: + value: every_save +hub_token: + value: +id2label: + value: + "0": LABEL_0 + "1": LABEL_1 +ignore_data_skip: + value: false +include_for_metrics: + value: [] +include_num_input_tokens_seen: + value: "no" +initializer_range: + value: 0.02 +intermediate_size: + value: 256 +is_encoder_decoder: + value: false +label_names: + value: null +label_smoothing_factor: + value: 0 +label2id: + value: + LABEL_0: 0 + LABEL_1: 1 +learning_rate: + value: 0.001 +length_column_name: + value: length +liger_kernel_config: + value: null +load_best_model_at_end: + value: false +local_rank: + value: -1 +log_level: + value: passive +log_level_replica: + value: warning +log_on_each_node: + value: true +logging_first_step: + value: false +logging_nan_inf_filter: + value: true +logging_steps: + value: 20 +logging_strategy: + value: steps +lr_scheduler_kwargs: + value: null +lr_scheduler_type: + value: constant +max_grad_norm: + value: 1 +max_position_embeddings: + value: 512 +max_steps: + value: 1000 +metric_for_best_model: + value: null +mlp_bias: + value: false +mlp_type: + value: glu +model/num_parameters: + value: 3970432 +model_type: + value: tiny_llama +neftune_noise_alpha: + value: null +num_attention_heads: + value: 4 +num_hidden_layers: + value: 21 +num_key_value_heads: + value: 4 +num_train_epochs: + value: 1 +optim: + value: adamw_torch +optim_args: + value: null +optim_target_modules: + value: null +output_attentions: + value: false +output_dir: + value: out/glu-waleed10-21L_run +output_hidden_states: + value: false +pad_token_id: + value: 0 +parallelism_config: + value: null +per_device_eval_batch_size: + value: 512 +per_device_train_batch_size: + value: 512 +prediction_loss_only: + value: false +pretraining_tp: + value: 1 +problem_type: + value: null +project: + value: huggingface +push_to_hub: + value: false +remove_unused_columns: + value: false +report_to: + value: + - wandb +restore_callback_states_from_checkpoint: + value: false +resume_from_checkpoint: + value: null +return_dict: + value: true +rms_norm_eps: + value: 1e-06 +rope_parameters: + value: + rope_theta: 10000 + rope_type: default +run_name: + value: LM-glu-waleed10-21L-4.0M-20260814-212813 +save_on_each_node: + value: false +save_only_model: + value: false +save_steps: + value: 100 +save_strategy: + value: steps +save_total_limit: + value: null +seed: + value: 42 +skip_memory_metrics: + value: true +tf32: + value: null +tie_word_embeddings: + value: true +tokenizer_name: + value: w-ahmad/tiny-stories-tokenizer +torch_compile: + value: false +torch_compile_backend: + value: null +torch_compile_mode: + value: null +torch_empty_cache_steps: + value: null +trackio_bucket_id: + value: null +trackio_space_id: + value: null +trackio_static_space_id: + value: null +train_sampling_strategy: + value: random +transformers_version: + value: 5.16.0.dev0 +use_cache: + value: false +use_cpu: + value: false +use_liger_kernel: + value: false +vocab_size: + value: 4096 +waleed_beta: + value: 10 +warmup_steps: + value: 0 +weight_decay: + value: 0.01 diff --git a/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/files/output.log b/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/files/output.log new file mode 100644 index 0000000000000000000000000000000000000000..15125abfc9b8abe0b7739d3f0b81a3325d0faae3 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/files/output.log @@ -0,0 +1,51 @@ + 0%| | 0/1000 [00:00', 'hub_private_repo': None, 'hub_model_id': 'w-ahmad/A-glu-waleed10-21L', 'hub_strategy': 'every_save', 'hub_always_push': False, 'hub_revision': None, 'load_best_model_at_end': False, 'metric_for_best_model': None, 'greater_is_better': None, 'ignore_data_skip': False, 'restore_callback_states_from_checkpoint': False, 'full_determinism': False, 'seed': 42, 'data_seed': 42, 'use_cpu': False, 'accelerator_config': {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}, 'parallelism_config': None, 'dataloader_drop_last': False, 'dataloader_num_workers': 0, 'dataloader_pin_memory': True, 'dataloader_persistent_workers': False, 'dataloader_prefetch_factor': None, 'dataloader_multiprocessing_context': None, 'dataloader_in_order': True, 'remove_unused_columns': False, 'label_names': None, 'train_sampling_strategy': 'random', 'length_column_name': 'length', 'ddp_find_unused_parameters': None, 'ddp_bucket_cap_mb': None, 'ddp_broadcast_buffers': None, 'ddp_static_graph': None, 'ddp_backend': None, 'ddp_timeout': 1800, 'fsdp': None, 'fsdp_config': None, 'deepspeed': None, 'debug': [], 'skip_memory_metrics': True, 'do_train': False, 'do_eval': True, 'do_predict': False, 'resume_from_checkpoint': None, 'local_rank': -1} +2026-08-14 21:28:15,019 INFO MainThread:1193850 [wandb_config.py:__setitem__():155] [no run ID] config set model/num_parameters = 3970432 - > +2026-08-14 21:28:15,019 INFO MainThread:1193850 [wandb_run.py:_config_callback():1346] config_cb model/num_parameters 3970432 None +2026-08-14 21:28:16,533 INFO MainThread:1193850 [wandb_run.py:_finish():2383] finishing run deepnevro-deepnevro/research1/vtp6pf3m +2026-08-14 21:28:16,534 INFO MainThread:1193850 [wandb_run.py:_atexit_cleanup():2588] got exitcode: 0 +2026-08-14 21:28:16,535 INFO MainThread:1193850 [wandb_run.py:_restore():2570] restore +2026-08-14 21:28:16,535 INFO MainThread:1193850 [wandb_run.py:_restore():2576] restore done +2026-08-14 21:28:17,132 INFO MainThread:1193850 [wandb_run.py:_footer_sync_info():3993] logging synced files diff --git a/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/run-vtp6pf3m.wandb b/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/run-vtp6pf3m.wandb new file mode 100644 index 0000000000000000000000000000000000000000..7164fadebaf289c44473ee9492320cfc4725eaa3 Binary files /dev/null and b/zain/Activation/wandb/run-20260814_212814-vtp6pf3m/run-vtp6pf3m.wandb differ diff --git a/zain/Activation/wandb/run-20260814_212818-fj94kr9p/files/config.yaml b/zain/Activation/wandb/run-20260814_212818-fj94kr9p/files/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ddce4c513057d44efccb637c5e6553c0f27404c5 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212818-fj94kr9p/files/config.yaml @@ -0,0 +1,438 @@ +_name_or_path: + value: "" +_wandb: + value: + cli_version: 0.28.1 + e: + xio5js2delb7osbi95ndowruwry5o7mi: + args: + - --config + - /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline.yaml + - --variants + - glu-waleed10 + - glu-silu-waleed10 + - mlp-waleed10 + - mlp-silu-waleed10 + - glu-waleed + - glu-situglu_low + - glu-waleedglu_low + codePath: sweep.py + codePathLocal: sweep.py + cpu_count: 112 + cpu_count_logical: 224 + cudaVersion: "12.4" + disk: + /: + total: "1560765693952" + used: "706489917440" + email: deepnevro@gmail.com + executable: /mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python + git: + commit: 3fb47c6943d6927eafc39bd399a3d3a520bae74a + remote: https://github.com/w-ahmad1a10/Activation.git + gpu: NVIDIA H100 80GB HBM3 + gpu_count: 8 + gpu_nvidia: + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-39c684a5-fde6-83d7-1663-0859795881ae + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-132944c4-b689-2b5f-89a4-d730401677ab + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bc6c3e3c-9b90-09ca-c034-774961847c54 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea + host: deeplens-k3s-node1 + memory: + total: "2164089937920" + os: Linux-5.15.0-126-generic-x86_64-with-glibc2.35 + program: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py + python: CPython 3.11.15 + root: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation + startedAt: "2026-08-14T21:28:18.170018Z" + writerId: xio5js2delb7osbi95ndowruwry5o7mi + m: + - "1": train/global_step + "6": + - 3 + "7": [] + - "2": '*' + "5": 1 + "6": + - 1 + "7": [] + python_version: 3.11.15 + t: + "1": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "2": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "3": + - 2 + - 7 + - 13 + - 19 + - 66 + "4": 3.11.15 + "5": 0.28.1 + "6": 5.16.0.dev0 + "9": + "1": transformers_trainer + "12": 0.28.1 + "13": linux-x86_64 +accelerator_config: + value: + dispatch_batches: null + even_batches: true + gradient_accumulation_kwargs: null + non_blocking: false + split_batches: false + use_seedable_sampler: true +activation: + value: silu-waleed10 +adam_beta1: + value: 0.9 +adam_beta2: + value: 0.999 +adam_epsilon: + value: 1e-08 +architectures: + value: null +attention_bias: + value: false +attention_dropout: + value: 0 +auto_find_batch_size: + value: false +average_tokens_across_devices: + value: true +batch_eval_metrics: + value: false +bf16: + value: true +bf16_full_eval: + value: false +bos_token_id: + value: 1 +chunk_size_feed_forward: + value: 0 +data_seed: + value: 42 +dataloader_drop_last: + value: false +dataloader_in_order: + value: true +dataloader_multiprocessing_context: + value: null +dataloader_num_workers: + value: 0 +dataloader_persistent_workers: + value: false +dataloader_pin_memory: + value: true +dataloader_prefetch_factor: + value: null +ddp_backend: + value: null +ddp_broadcast_buffers: + value: null +ddp_bucket_cap_mb: + value: null +ddp_find_unused_parameters: + value: null +ddp_static_graph: + value: null +ddp_timeout: + value: 1800 +debug: + value: [] +deepspeed: + value: null +disable_tqdm: + value: false +do_eval: + value: true +do_predict: + value: false +do_train: + value: false +dtype: + value: null +enable_jit_checkpoint: + value: false +eos_token_id: + value: 2 +eval_accumulation_steps: + value: null +eval_delay: + value: 0 +eval_do_concat_batches: + value: true +eval_on_start: + value: false +eval_steps: + value: 60000 +eval_strategy: + value: steps +eval_use_gather_object: + value: false +fp16: + value: false +fp16_full_eval: + value: false +fsdp: + value: null +fsdp_config: + value: null +full_determinism: + value: false +gradient_accumulation_steps: + value: 1 +gradient_checkpointing: + value: false +gradient_checkpointing_kwargs: + value: null +greater_is_better: + value: null +head_dim: + value: 32 +hidden_act: + value: silu +hidden_size: + value: 128 +hub_always_push: + value: false +hub_model_id: + value: w-ahmad/A-glu-silu-waleed10-21L +hub_private_repo: + value: null +hub_revision: + value: null +hub_strategy: + value: every_save +hub_token: + value: +id2label: + value: + "0": LABEL_0 + "1": LABEL_1 +ignore_data_skip: + value: false +include_for_metrics: + value: [] +include_num_input_tokens_seen: + value: "no" +initializer_range: + value: 0.02 +intermediate_size: + value: 256 +is_encoder_decoder: + value: false +label_names: + value: null +label_smoothing_factor: + value: 0 +label2id: + value: + LABEL_0: 0 + LABEL_1: 1 +learning_rate: + value: 0.001 +length_column_name: + value: length +liger_kernel_config: + value: null +load_best_model_at_end: + value: false +local_rank: + value: -1 +log_level: + value: passive +log_level_replica: + value: warning +log_on_each_node: + value: true +logging_first_step: + value: false +logging_nan_inf_filter: + value: true +logging_steps: + value: 20 +logging_strategy: + value: steps +lr_scheduler_kwargs: + value: null +lr_scheduler_type: + value: constant +max_grad_norm: + value: 1 +max_position_embeddings: + value: 512 +max_steps: + value: 1000 +metric_for_best_model: + value: null +mlp_bias: + value: false +mlp_type: + value: glu +model/num_parameters: + value: 3970432 +model_type: + value: tiny_llama +neftune_noise_alpha: + value: null +num_attention_heads: + value: 4 +num_hidden_layers: + value: 21 +num_key_value_heads: + value: 4 +num_train_epochs: + value: 1 +optim: + value: adamw_torch +optim_args: + value: null +optim_target_modules: + value: null +output_attentions: + value: false +output_dir: + value: out/glu-silu-waleed10-21L_run +output_hidden_states: + value: false +pad_token_id: + value: 0 +parallelism_config: + value: null +per_device_eval_batch_size: + value: 512 +per_device_train_batch_size: + value: 512 +prediction_loss_only: + value: false +pretraining_tp: + value: 1 +problem_type: + value: null +project: + value: huggingface +push_to_hub: + value: false +remove_unused_columns: + value: false +report_to: + value: + - wandb +restore_callback_states_from_checkpoint: + value: false +resume_from_checkpoint: + value: null +return_dict: + value: true +rms_norm_eps: + value: 1e-06 +rope_parameters: + value: + rope_theta: 10000 + rope_type: default +run_name: + value: LM-glu-silu-waleed10-21L-4.0M-20260814-212817 +save_on_each_node: + value: false +save_only_model: + value: false +save_steps: + value: 100 +save_strategy: + value: steps +save_total_limit: + value: null +seed: + value: 42 +skip_memory_metrics: + value: true +tf32: + value: null +tie_word_embeddings: + value: true +tokenizer_name: + value: w-ahmad/tiny-stories-tokenizer +torch_compile: + value: false +torch_compile_backend: + value: null +torch_compile_mode: + value: null +torch_empty_cache_steps: + value: null +trackio_bucket_id: + value: null +trackio_space_id: + value: null +trackio_static_space_id: + value: null +train_sampling_strategy: + value: random +transformers_version: + value: 5.16.0.dev0 +use_cache: + value: false +use_cpu: + value: false +use_liger_kernel: + value: false +vocab_size: + value: 4096 +waleed_beta: + value: 10 +warmup_steps: + value: 0 +weight_decay: + value: 0.01 diff --git a/zain/Activation/wandb/run-20260814_212818-fj94kr9p/files/output.log b/zain/Activation/wandb/run-20260814_212818-fj94kr9p/files/output.log new file mode 100644 index 0000000000000000000000000000000000000000..010c0ece82d38f3fbba4f5827dfa15d2153c1e6f --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212818-fj94kr9p/files/output.log @@ -0,0 +1,49 @@ + 0%| | 0/1000 [00:00', 'hub_private_repo': None, 'hub_model_id': 'w-ahmad/A-glu-silu-waleed10-21L', 'hub_strategy': 'every_save', 'hub_always_push': False, 'hub_revision': None, 'load_best_model_at_end': False, 'metric_for_best_model': None, 'greater_is_better': None, 'ignore_data_skip': False, 'restore_callback_states_from_checkpoint': False, 'full_determinism': False, 'seed': 42, 'data_seed': 42, 'use_cpu': False, 'accelerator_config': {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}, 'parallelism_config': None, 'dataloader_drop_last': False, 'dataloader_num_workers': 0, 'dataloader_pin_memory': True, 'dataloader_persistent_workers': False, 'dataloader_prefetch_factor': None, 'dataloader_multiprocessing_context': None, 'dataloader_in_order': True, 'remove_unused_columns': False, 'label_names': None, 'train_sampling_strategy': 'random', 'length_column_name': 'length', 'ddp_find_unused_parameters': None, 'ddp_bucket_cap_mb': None, 'ddp_broadcast_buffers': None, 'ddp_static_graph': None, 'ddp_backend': None, 'ddp_timeout': 1800, 'fsdp': None, 'fsdp_config': None, 'deepspeed': None, 'debug': [], 'skip_memory_metrics': True, 'do_train': False, 'do_eval': True, 'do_predict': False, 'resume_from_checkpoint': None, 'local_rank': -1} +2026-08-14 21:28:18,754 INFO MainThread:1193850 [wandb_config.py:__setitem__():155] [no run ID] config set model/num_parameters = 3970432 - > +2026-08-14 21:28:18,754 INFO MainThread:1193850 [wandb_run.py:_config_callback():1346] config_cb model/num_parameters 3970432 None +2026-08-14 21:28:19,673 INFO MainThread:1193850 [wandb_run.py:_finish():2383] finishing run deepnevro-deepnevro/research1/fj94kr9p +2026-08-14 21:28:19,674 INFO MainThread:1193850 [wandb_run.py:_atexit_cleanup():2588] got exitcode: 0 +2026-08-14 21:28:19,674 INFO MainThread:1193850 [wandb_run.py:_restore():2570] restore +2026-08-14 21:28:19,674 INFO MainThread:1193850 [wandb_run.py:_restore():2576] restore done +2026-08-14 21:28:20,223 INFO MainThread:1193850 [wandb_run.py:_footer_sync_info():3993] logging synced files diff --git a/zain/Activation/wandb/run-20260814_212818-fj94kr9p/run-fj94kr9p.wandb b/zain/Activation/wandb/run-20260814_212818-fj94kr9p/run-fj94kr9p.wandb new file mode 100644 index 0000000000000000000000000000000000000000..298bc5546617a7915a1c89f31aa79180fd5c250c Binary files /dev/null and b/zain/Activation/wandb/run-20260814_212818-fj94kr9p/run-fj94kr9p.wandb differ diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/config.yaml b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7ea3a1bd98bd73c76027fc833743e4d08bbb70cd --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/config.yaml @@ -0,0 +1,438 @@ +_name_or_path: + value: "" +_wandb: + value: + cli_version: 0.28.1 + e: + 2drnyjxeagicq7ia3qgncq8z361n6gif: + args: + - --config + - /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline.yaml + - --variants + - glu-waleed10 + - glu-silu-waleed10 + - mlp-waleed10 + - mlp-silu-waleed10 + - glu-waleed + - glu-situglu_low + - glu-waleedglu_low + codePath: sweep.py + codePathLocal: sweep.py + cpu_count: 112 + cpu_count_logical: 224 + cudaVersion: "12.4" + disk: + /: + total: "1560765693952" + used: "706489917440" + email: deepnevro@gmail.com + executable: /mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python + git: + commit: 3fb47c6943d6927eafc39bd399a3d3a520bae74a + remote: https://github.com/w-ahmad1a10/Activation.git + gpu: NVIDIA H100 80GB HBM3 + gpu_count: 8 + gpu_nvidia: + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-39c684a5-fde6-83d7-1663-0859795881ae + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-132944c4-b689-2b5f-89a4-d730401677ab + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-bc6c3e3c-9b90-09ca-c034-774961847c54 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9 + - architecture: Hopper + cudaCores: 16896 + memoryTotal: "85520809984" + name: NVIDIA H100 80GB HBM3 + uuid: GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea + host: deeplens-k3s-node1 + memory: + total: "2164089937920" + os: Linux-5.15.0-126-generic-x86_64-with-glibc2.35 + program: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py + python: CPython 3.11.15 + root: /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation + startedAt: "2026-08-14T21:28:20.753237Z" + writerId: 2drnyjxeagicq7ia3qgncq8z361n6gif + m: + - "1": train/global_step + "6": + - 3 + "7": [] + - "2": '*' + "5": 1 + "6": + - 1 + "7": [] + python_version: 3.11.15 + t: + "1": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "2": + - 1 + - 5 + - 11 + - 41 + - 49 + - 51 + - 53 + - 71 + "3": + - 2 + - 7 + - 13 + - 19 + - 66 + "4": 3.11.15 + "5": 0.28.1 + "6": 5.16.0.dev0 + "9": + "1": transformers_trainer + "12": 0.28.1 + "13": linux-x86_64 +accelerator_config: + value: + dispatch_batches: null + even_batches: true + gradient_accumulation_kwargs: null + non_blocking: false + split_batches: false + use_seedable_sampler: true +activation: + value: waleed10 +adam_beta1: + value: 0.9 +adam_beta2: + value: 0.999 +adam_epsilon: + value: 1e-08 +architectures: + value: null +attention_bias: + value: false +attention_dropout: + value: 0 +auto_find_batch_size: + value: false +average_tokens_across_devices: + value: true +batch_eval_metrics: + value: false +bf16: + value: true +bf16_full_eval: + value: false +bos_token_id: + value: 1 +chunk_size_feed_forward: + value: 0 +data_seed: + value: 42 +dataloader_drop_last: + value: false +dataloader_in_order: + value: true +dataloader_multiprocessing_context: + value: null +dataloader_num_workers: + value: 0 +dataloader_persistent_workers: + value: false +dataloader_pin_memory: + value: true +dataloader_prefetch_factor: + value: null +ddp_backend: + value: null +ddp_broadcast_buffers: + value: null +ddp_bucket_cap_mb: + value: null +ddp_find_unused_parameters: + value: null +ddp_static_graph: + value: null +ddp_timeout: + value: 1800 +debug: + value: [] +deepspeed: + value: null +disable_tqdm: + value: false +do_eval: + value: true +do_predict: + value: false +do_train: + value: false +dtype: + value: null +enable_jit_checkpoint: + value: false +eos_token_id: + value: 2 +eval_accumulation_steps: + value: null +eval_delay: + value: 0 +eval_do_concat_batches: + value: true +eval_on_start: + value: false +eval_steps: + value: 60000 +eval_strategy: + value: steps +eval_use_gather_object: + value: false +fp16: + value: false +fp16_full_eval: + value: false +fsdp: + value: null +fsdp_config: + value: null +full_determinism: + value: false +gradient_accumulation_steps: + value: 1 +gradient_checkpointing: + value: false +gradient_checkpointing_kwargs: + value: null +greater_is_better: + value: null +head_dim: + value: 32 +hidden_act: + value: silu +hidden_size: + value: 128 +hub_always_push: + value: false +hub_model_id: + value: w-ahmad/A-mlp-waleed10-21L +hub_private_repo: + value: null +hub_revision: + value: null +hub_strategy: + value: every_save +hub_token: + value: +id2label: + value: + "0": LABEL_0 + "1": LABEL_1 +ignore_data_skip: + value: false +include_for_metrics: + value: [] +include_num_input_tokens_seen: + value: "no" +initializer_range: + value: 0.02 +intermediate_size: + value: 256 +is_encoder_decoder: + value: false +label_names: + value: null +label_smoothing_factor: + value: 0 +label2id: + value: + LABEL_0: 0 + LABEL_1: 1 +learning_rate: + value: 0.001 +length_column_name: + value: length +liger_kernel_config: + value: null +load_best_model_at_end: + value: false +local_rank: + value: -1 +log_level: + value: passive +log_level_replica: + value: warning +log_on_each_node: + value: true +logging_first_step: + value: false +logging_nan_inf_filter: + value: true +logging_steps: + value: 20 +logging_strategy: + value: steps +lr_scheduler_kwargs: + value: null +lr_scheduler_type: + value: constant +max_grad_norm: + value: 1 +max_position_embeddings: + value: 512 +max_steps: + value: 1000 +metric_for_best_model: + value: null +mlp_bias: + value: false +mlp_type: + value: mlp +model/num_parameters: + value: 3970432 +model_type: + value: tiny_llama +neftune_noise_alpha: + value: null +num_attention_heads: + value: 4 +num_hidden_layers: + value: 21 +num_key_value_heads: + value: 4 +num_train_epochs: + value: 1 +optim: + value: adamw_torch +optim_args: + value: null +optim_target_modules: + value: null +output_attentions: + value: false +output_dir: + value: out/mlp-waleed10-21L_run +output_hidden_states: + value: false +pad_token_id: + value: 0 +parallelism_config: + value: null +per_device_eval_batch_size: + value: 512 +per_device_train_batch_size: + value: 512 +prediction_loss_only: + value: false +pretraining_tp: + value: 1 +problem_type: + value: null +project: + value: huggingface +push_to_hub: + value: false +remove_unused_columns: + value: false +report_to: + value: + - wandb +restore_callback_states_from_checkpoint: + value: false +resume_from_checkpoint: + value: null +return_dict: + value: true +rms_norm_eps: + value: 1e-06 +rope_parameters: + value: + rope_theta: 10000 + rope_type: default +run_name: + value: LM-mlp-waleed10-21L-4.0M-20260814-212820 +save_on_each_node: + value: false +save_only_model: + value: false +save_steps: + value: 100 +save_strategy: + value: steps +save_total_limit: + value: null +seed: + value: 42 +skip_memory_metrics: + value: true +tf32: + value: null +tie_word_embeddings: + value: true +tokenizer_name: + value: w-ahmad/tiny-stories-tokenizer +torch_compile: + value: false +torch_compile_backend: + value: null +torch_compile_mode: + value: null +torch_empty_cache_steps: + value: null +trackio_bucket_id: + value: null +trackio_space_id: + value: null +trackio_static_space_id: + value: null +train_sampling_strategy: + value: random +transformers_version: + value: 5.16.0.dev0 +use_cache: + value: false +use_cpu: + value: false +use_liger_kernel: + value: false +vocab_size: + value: 4096 +waleed_beta: + value: 10 +warmup_steps: + value: 0 +weight_decay: + value: 0.01 diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/output.log b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/output.log new file mode 100644 index 0000000000000000000000000000000000000000..001fee24383aaadac928e7dcde3d93917e4a2edd --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/output.log @@ -0,0 +1,3 @@ + 5%|▍ | 46/1000 [00:39<15:11, 1.05it/s] +{'loss': '7.273', 'grad_norm': '1.055', 'learning_rate': '0.001', 'epoch': '0.01078', 'train/total_time_seconds': '6.615', 'train/time_per_step_avg': '0.3307', 'train/epoch_time_elapsed': '14.85', 'train/estimated_remaining_minutes': '5.402'} +{'loss': '5.986', 'grad_norm': '0.8242', 'learning_rate': '0.001', 'epoch': '0.02156', 'train/total_time_seconds': '18.31', 'train/time_per_step_avg': '0.4577', 'train/epoch_time_elapsed': '34.23', 'train/estimated_remaining_minutes': '7.323'} diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/requirements.txt b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/requirements.txt new file mode 100644 index 0000000000000000000000000000000000000000..f36b19895172d9637438a445ceacb71d09dc5d02 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/requirements.txt @@ -0,0 +1,149 @@ +asttokens==3.0.1 +comm==0.2.3 +debugpy==1.8.21 +decorator==5.3.1 +executing==2.2.1 +nest-asyncio==1.6.0 +parso==0.8.7 +platformdirs==4.11.0 +psutil==7.2.2 +ptyprocess==0.7.0 +pure_eval==0.2.3 +Pygments==2.20.0 +pyzmq==27.1.0 +setuptools==83.0.0 +six==1.17.0 +tornado==6.5.7 +traitlets==5.15.0 +fsspec==2026.4.0 +wcwidth==0.8.2 +ipython_pygments_lexers==1.1.1 +jedi==0.20.0 +jupyter_core==5.9.1 +matplotlib-inline==0.2.2 +pexpect==4.9.0 +prompt_toolkit==3.0.53 +python-dateutil==2.9.0.post0 +stack_data==0.6.3 +wheel==0.47.0 +jupyter_client==8.9.1 +pip==26.1.2 +ipython==9.15.0 +ipykernel==7.2.0 +threadpoolctl==3.6.0 +pyparsing==3.3.2 +typing_extensions==4.15.0 +Jinja2==3.1.6 +narwhals==2.24.0 +kiwisolver==1.5.0 +joblib==1.5.3 +fonttools==4.63.0 +cycler==0.12.1 +scipy==1.17.1 +pandas==3.0.5 +contourpy==1.3.3 +scikit-learn==1.9.0 +matplotlib==3.11.1 +urllib3==2.7.0 +tqdm==4.70.0 +idna==3.18 +charset-normalizer==3.4.9 +certifi==2026.7.22 +requests==2.34.2 +seaborn==0.13.2 +uv==0.12.0 +shellingham==1.5.4 +mpmath==1.3.0 +attrs==26.1.0 +hf-xet==1.5.2 +nvidia-nccl-cu12==2.21.5 +MarkupSafe==3.0.3 +regex==2026.7.19 +importlib_metadata==9.0.0 +httpcore==1.0.9 +annotated-doc==0.0.5 +multidict==6.7.1 +aiohttp==3.14.3 +aiosignal==1.4.0 +xxhash==3.8.1 +aiohappyeyeballs==2.7.1 +mdurl==0.1.2 +cuda-toolkit==13.0.3.0 +networkx==3.6.1 +PyYAML==6.0.3 +nvidia-cufile==1.15.1.6 +typer==0.27.0 +torchaudio==2.6.0+cu124 +rich==15.0.0 +nvidia-cufft-cu12==11.2.1.3 +h11==0.16.0 +dill==0.4.1 +cuda-pathfinder==1.6.0 +filelock==3.29.0 +nvidia-nvtx-cu12==12.4.127 +httpx==0.28.1 +anyio==4.14.2 +numpy==2.4.4 +yarl==1.24.5 +click==8.4.2 +triton==3.2.0 +frozenlist==1.8.0 +zipp==4.1.0 +propcache==0.5.2 +tokenizers==0.22.2 +markdown-it-py==4.2.0 +nvidia-cuda-runtime==13.0.96 +cuda-bindings==13.3.1 +nvidia-cuda-cupti==13.0.85 +torch==2.6.0+cu124 +multiprocess==0.70.19 +pillow==12.2.0 +transformers==5.16.0.dev0 +wandb==0.28.1 +nvidia-curand==10.4.0.35 +sympy==1.13.1 +nvidia-cusparse==12.6.3.3 +nvidia-cuda-nvrtc==13.0.88 +typing-inspection==0.4.2 +nvidia-cusolver==12.0.4.66 +nvidia-cufft==12.0.0.61 +nvidia-cudnn-cu13==9.20.0.48 +nvidia-cublas==13.1.1.3 +pyarrow==25.0.0 +evaluate==0.4.6 +diffusers==0.39.0 +pydantic==2.13.4 +annotated-types==0.8.0 +protobuf==7.35.1 +sentry-sdk==2.66.1 +einops==0.8.2 +packaging==26.2 +nvidia-nvjitlink-cu12==12.4.127 +nvidia-curand-cu12==10.3.5.147 +nvidia-cusparselt-cu12==0.6.2 +nvidia-cusparse-cu12==12.3.1.170 +nvidia-cuda-runtime-cu12==12.4.127 +torchvision==0.21.0+cu124 +nvidia-cuda-nvrtc-cu12==12.4.127 +nvidia-cuda-cupti-cu12==12.4.127 +nvidia-cusolver-cu12==11.6.1.9 +nvidia-cublas-cu12==12.4.5.8 +nvidia-cudnn-cu12==9.1.0.70 +huggingface_hub==1.26.0 +datasets==5.0.1 +safetensors==0.8.0 +accelerate==1.14.0 +pydantic_core==2.46.4 +ninja==1.13.0 +autocommand==2.2.2 +backports.tarfile==1.2.0 +importlib_metadata==8.7.1 +jaraco.text==4.0.0 +jaraco.context==6.1.0 +jaraco.functools==4.4.0 +more-itertools==10.8.0 +packaging==26.0 +platformdirs==4.4.0 +tomli==2.4.0 +wheel==0.46.3 +zipp==3.23.0 diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/wandb-metadata.json b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/wandb-metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..e512a1e6013cf31c9ff7fe0b2a47b9b9153b4cfe --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/wandb-metadata.json @@ -0,0 +1,101 @@ +{ + "os": "Linux-5.15.0-126-generic-x86_64-with-glibc2.35", + "python": "CPython 3.11.15", + "startedAt": "2026-08-14T21:28:20.753237Z", + "args": [ + "--config", + "/mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline.yaml", + "--variants", + "glu-waleed10", + "glu-silu-waleed10", + "mlp-waleed10", + "mlp-silu-waleed10", + "glu-waleed", + "glu-situglu_low", + "glu-waleedglu_low" + ], + "program": "/mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py", + "codePath": "sweep.py", + "codePathLocal": "sweep.py", + "git": { + "remote": "https://github.com/w-ahmad1a10/Activation.git", + "commit": "3fb47c6943d6927eafc39bd399a3d3a520bae74a" + }, + "email": "deepnevro@gmail.com", + "root": "/mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation", + "host": "deeplens-k3s-node1", + "executable": "/mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python", + "cpu_count": 112, + "cpu_count_logical": 224, + "gpu": "NVIDIA H100 80GB HBM3", + "gpu_count": 8, + "disk": { + "/": { + "total": "1560765693952", + "used": "706489917440" + } + }, + "memory": { + "total": "2164089937920" + }, + "gpu_nvidia": [ + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-39c684a5-fde6-83d7-1663-0859795881ae" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-132944c4-b689-2b5f-89a4-d730401677ab" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-bc6c3e3c-9b90-09ca-c034-774961847c54" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea" + } + ], + "cudaVersion": "12.4", + "writerId": "2drnyjxeagicq7ia3qgncq8z361n6gif" +} \ No newline at end of file diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/wandb-summary.json b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/wandb-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..040152603cfefce5934b8a590e3fd9237d44a926 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/files/wandb-summary.json @@ -0,0 +1 @@ +{"train/learning_rate":0.001,"_runtime":40,"train/loss":5.986497497558593,"train/epoch":0.0215633423180593,"_timestamp":1.786742935626837e+09,"_wandb":{"runtime":40},"train/global_step":40,"train/grad_norm":0.82421875,"_step":1} \ No newline at end of file diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-core.log b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-core.log new file mode 100644 index 0000000000000000000000000000000000000000..6d17cf7f6524b2c7f77ef9f100bffb8a6d3b4975 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-core.log @@ -0,0 +1,32 @@ +{"time":"2026-08-14T21:28:13.875214015Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmp61t5kns7/port-1193850.txt","pid":1193850,"detached":false,"idle-timeout":600000000000,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false} +{"time":"2026-08-14T21:28:13.875684603Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":1193850} +{"time":"2026-08-14T21:28:13.875676582Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-1193850-1195287-1939645545/socket","Net":"unix"}} +{"time":"2026-08-14T21:28:14.052767875Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"} +{"time":"2026-08-14T21:28:14.184824068Z","level":"INFO","msg":"handleInformInit: received","streamId":"vtp6pf3m","id":"1(@)"} +{"time":"2026-08-14T21:28:14.445813383Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"vtp6pf3m","id":"1(@)"} +{"time":"2026-08-14T21:28:16.535398722Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"ppjgf4bx6kkl"} +{"time":"2026-08-14T21:28:17.132108791Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"ppjgf4bx6kkl"} +{"time":"2026-08-14T21:28:17.133244824Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"vtp6pf3m","id":"1(@)"} +{"time":"2026-08-14T21:28:17.13384059Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"vtp6pf3m","id":"1(@)"} +{"time":"2026-08-14T21:28:18.176700551Z","level":"INFO","msg":"handleInformInit: received","streamId":"fj94kr9p","id":"1(@)"} +{"time":"2026-08-14T21:28:18.417597198Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"fj94kr9p","id":"1(@)"} +{"time":"2026-08-14T21:28:19.674866991Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"e4d4etq11mlq"} +{"time":"2026-08-14T21:28:20.222556488Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"e4d4etq11mlq"} +{"time":"2026-08-14T21:28:20.223783577Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"fj94kr9p","id":"1(@)"} +{"time":"2026-08-14T21:28:20.224344691Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"fj94kr9p","id":"1(@)"} +{"time":"2026-08-14T21:28:20.755487912Z","level":"INFO","msg":"handleInformInit: received","streamId":"ucwxn3gs","id":"1(@)"} +{"time":"2026-08-14T21:28:20.996759448Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"ucwxn3gs","id":"1(@)"} +{"time":"2026-08-14T21:28:26.491790165Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"4jw8kj1lwl6y"} +{"time":"2026-08-14T21:29:01.918410879Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"4jw8kj1lwl6y"} +{"time":"2026-08-14T21:29:02.578939706Z","level":"INFO","msg":"connection: cancelling request","id":"1(@)","requestId":"4jw8kj1lwl6y"} +{"time":"2026-08-14T21:29:02.580571999Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"ucwxn3gs","id":"1(@)"} +{"time":"2026-08-14T21:29:02.581104066Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"ucwxn3gs","id":"1(@)"} +{"time":"2026-08-14T21:29:02.642980411Z","level":"INFO","msg":"handleInformTeardown: server teardown initiated","id":"1(@)"} +{"time":"2026-08-14T21:29:02.643036973Z","level":"INFO","msg":"handleInformTeardown: server shutdown complete","id":"1(@)"} +{"time":"2026-08-14T21:29:02.643048401Z","level":"INFO","msg":"connection: closing","id":"1(@)"} +{"time":"2026-08-14T21:29:02.643056857Z","level":"INFO","msg":"server: is shutting down"} +{"time":"2026-08-14T21:29:02.643101792Z","level":"INFO","msg":"connection: closed successfully","id":"1(@)"} +{"time":"2026-08-14T21:29:02.643077696Z","level":"INFO","msg":"processOutgoingData: finished","id":"1(@)"} +{"time":"2026-08-14T21:29:02.643124976Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"1(@)"} +{"time":"2026-08-14T21:29:02.643304099Z","level":"INFO","msg":"server: listener closed","addr":{"Name":"/tmp/wandb-1193850-1195287-1939645545/socket","Net":"unix"}} +{"time":"2026-08-14T21:29:02.643377148Z","level":"INFO","msg":"server: all connections closed"} diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-internal.log b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-internal.log new file mode 100644 index 0000000000000000000000000000000000000000..fa9ccebe26b3cb10de863772453b8968717b8696 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-internal.log @@ -0,0 +1,23 @@ +{"time":"2026-08-14T21:28:20.75566175Z","level":"INFO","msg":"wandb-core"} +{"time":"2026-08-14T21:28:20.755796491Z","level":"INFO","msg":"stream: starting","core version":"0.28.1"} +{"time":"2026-08-14T21:28:20.996598765Z","level":"INFO","msg":"stream: created new stream","id":"ucwxn3gs"} +{"time":"2026-08-14T21:28:20.996682158Z","level":"INFO","msg":"handler: started"} +{"time":"2026-08-14T21:28:20.996753346Z","level":"INFO","msg":"stream: started"} +{"time":"2026-08-14T21:28:20.996761846Z","level":"INFO","msg":"writer: started","stream_id":"ucwxn3gs"} +{"time":"2026-08-14T21:28:20.996781612Z","level":"INFO","msg":"sender: started"} +{"time":"2026-08-14T21:28:21.40242725Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1} +{"time":"2026-08-14T21:28:21.496872671Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:36.40321225Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":0,"events_lines":1,"console_offset":0,"console_lines":2,"uploaded_len":2} +{"time":"2026-08-14T21:28:36.521484735Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:51.402995777Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":1,"events_lines":2,"console_offset":0,"console_lines":1} +{"time":"2026-08-14T21:28:51.549478998Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:29:02.334952178Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"} +{"time":"2026-08-14T21:29:02.335224575Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":1,"events_offset":3,"events_lines":1,"console_offset":0,"console_lines":1,"uploaded_len":3} +{"time":"2026-08-14T21:29:02.45700121Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:29:02.45717743Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":2,"console_lines":1,"complete":true,"exit_code":0} +{"time":"2026-08-14T21:29:02.57605594Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:29:02.577378457Z","level":"INFO","msg":"handler: operation stats","stats":{}} +{"time":"2026-08-14T21:29:02.580598023Z","level":"INFO","msg":"stream: finishing up"} +{"time":"2026-08-14T21:29:02.580616646Z","level":"INFO","msg":"handler: closed"} +{"time":"2026-08-14T21:29:02.580706505Z","level":"INFO","msg":"sender: closed"} +{"time":"2026-08-14T21:29:02.580713607Z","level":"INFO","msg":"stream: all finished"} diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug.log b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug.log new file mode 100644 index 0000000000000000000000000000000000000000..347b07f0c254758ba816ed481c7a3a3a25a1a478 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug.log @@ -0,0 +1,24 @@ +2026-08-14 21:28:20,754 INFO MainThread:1193850 [wandb_init.py:setup_run_log_directory():729] Logging user logs to /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug.log +2026-08-14 21:28:20,754 INFO MainThread:1193850 [wandb_init.py:setup_run_log_directory():730] Logging internal logs to /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/logs/debug-internal.log +2026-08-14 21:28:20,754 INFO MainThread:1193850 [wandb_init.py:init():772] calling init triggers +2026-08-14 21:28:20,754 INFO MainThread:1193850 [wandb_init.py:init():777] wandb.init called with sweep_config: {} +config: {'_wandb': {}} +2026-08-14 21:28:20,754 INFO MainThread:1193850 [wandb_init.py:init():820] starting backend +2026-08-14 21:28:20,754 INFO MainThread:1193850 [wandb_init.py:init():835] sending inform_init request +2026-08-14 21:28:20,997 INFO MainThread:1193850 [wandb_init.py:init():840] backend started and connected +2026-08-14 21:28:20,998 INFO MainThread:1193850 [wandb_init.py:init():910] updated telemetry +2026-08-14 21:28:21,005 INFO MainThread:1193850 [wandb_init.py:init():933] communicating run to backend with 90.0 second timeout +2026-08-14 21:28:21,320 INFO MainThread:1193850 [wandb_init.py:init():978] starting run threads in backend +2026-08-14 21:28:21,396 INFO MainThread:1193850 [wandb_run.py:_console_start():2621] atexit reg +2026-08-14 21:28:21,396 INFO MainThread:1193850 [wandb_run.py:_redirect():2471] redirect: wrap_raw +2026-08-14 21:28:21,396 INFO MainThread:1193850 [wandb_run.py:_redirect():2540] Wrapping output streams. +2026-08-14 21:28:21,396 INFO MainThread:1193850 [wandb_run.py:_redirect():2563] Redirects installed. +2026-08-14 21:28:21,398 INFO MainThread:1193850 [wandb_init.py:init():1016] run started, returning control to user process +2026-08-14 21:28:21,398 INFO MainThread:1193850 [wandb_run.py:_config_callback():1346] config_cb None None {'transformers_version': '5.16.0.dev0', 'architectures': None, 'output_hidden_states': False, 'return_dict': True, 'dtype': None, 'chunk_size_feed_forward': 0, 'is_encoder_decoder': False, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'problem_type': None, 'vocab_size': 4096, 'hidden_size': 128, 'intermediate_size': 256, 'num_hidden_layers': 21, 'num_attention_heads': 4, 'num_key_value_heads': 4, 'hidden_act': 'silu', 'max_position_embeddings': 512, 'initializer_range': 0.02, 'rms_norm_eps': 1e-06, 'use_cache': False, 'pad_token_id': 0, 'bos_token_id': 1, 'eos_token_id': 2, 'pretraining_tp': 1, 'tie_word_embeddings': True, 'rope_parameters': {'rope_theta': 10000.0, 'rope_type': 'default'}, 'attention_bias': False, 'attention_dropout': 0.0, 'mlp_bias': False, 'head_dim': 32, '_name_or_path': '', 'tokenizer_name': 'w-ahmad/tiny-stories-tokenizer', 'mlp_type': 'mlp', 'activation': 'waleed10', 'waleed_beta': 10.0, 'model_type': 'tiny_llama', 'output_attentions': False, 'output_dir': 'out/mlp-waleed10-21L_run', 'per_device_train_batch_size': 512, 'num_train_epochs': 1, 'max_steps': 1000, 'learning_rate': 0.001, 'lr_scheduler_type': 'constant', 'lr_scheduler_kwargs': None, 'warmup_steps': 0, 'optim': 'adamw_torch', 'optim_args': None, 'weight_decay': 0.01, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'optim_target_modules': None, 'gradient_accumulation_steps': 1, 'average_tokens_across_devices': True, 'max_grad_norm': 1.0, 'label_smoothing_factor': 0.0, 'bf16': True, 'fp16': False, 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': None, 'gradient_checkpointing': False, 'gradient_checkpointing_kwargs': None, 'torch_compile': False, 'torch_compile_backend': None, 'torch_compile_mode': None, 'use_liger_kernel': False, 'liger_kernel_config': None, 'neftune_noise_alpha': None, 'torch_empty_cache_steps': None, 'auto_find_batch_size': False, 'logging_strategy': 'steps', 'logging_steps': 20, 'logging_first_step': False, 'log_on_each_node': True, 'logging_nan_inf_filter': True, 'include_num_input_tokens_seen': 'no', 'log_level': 'passive', 'log_level_replica': 'warning', 'disable_tqdm': False, 'report_to': ['wandb'], 'run_name': 'LM-mlp-waleed10-21L-4.0M-20260814-212820', 'project': 'huggingface', 'trackio_space_id': None, 'trackio_bucket_id': None, 'trackio_static_space_id': None, 'eval_strategy': 'steps', 'eval_steps': 60000, 'eval_delay': 0, 'per_device_eval_batch_size': 512, 'prediction_loss_only': False, 'eval_on_start': False, 'eval_do_concat_batches': True, 'eval_use_gather_object': False, 'eval_accumulation_steps': None, 'include_for_metrics': [], 'batch_eval_metrics': False, 'save_only_model': False, 'save_strategy': 'steps', 'save_steps': 100, 'save_on_each_node': False, 'save_total_limit': None, 'enable_jit_checkpoint': False, 'push_to_hub': False, 'hub_token': '', 'hub_private_repo': None, 'hub_model_id': 'w-ahmad/A-mlp-waleed10-21L', 'hub_strategy': 'every_save', 'hub_always_push': False, 'hub_revision': None, 'load_best_model_at_end': False, 'metric_for_best_model': None, 'greater_is_better': None, 'ignore_data_skip': False, 'restore_callback_states_from_checkpoint': False, 'full_determinism': False, 'seed': 42, 'data_seed': 42, 'use_cpu': False, 'accelerator_config': {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}, 'parallelism_config': None, 'dataloader_drop_last': False, 'dataloader_num_workers': 0, 'dataloader_pin_memory': True, 'dataloader_persistent_workers': False, 'dataloader_prefetch_factor': None, 'dataloader_multiprocessing_context': None, 'dataloader_in_order': True, 'remove_unused_columns': False, 'label_names': None, 'train_sampling_strategy': 'random', 'length_column_name': 'length', 'ddp_find_unused_parameters': None, 'ddp_bucket_cap_mb': None, 'ddp_broadcast_buffers': None, 'ddp_static_graph': None, 'ddp_backend': None, 'ddp_timeout': 1800, 'fsdp': None, 'fsdp_config': None, 'deepspeed': None, 'debug': [], 'skip_memory_metrics': True, 'do_train': False, 'do_eval': True, 'do_predict': False, 'resume_from_checkpoint': None, 'local_rank': -1} +2026-08-14 21:28:21,401 INFO MainThread:1193850 [wandb_config.py:__setitem__():155] [no run ID] config set model/num_parameters = 3970432 - > +2026-08-14 21:28:21,401 INFO MainThread:1193850 [wandb_run.py:_config_callback():1346] config_cb model/num_parameters 3970432 None +2026-08-14 21:29:01,916 INFO MainThread:1193850 [wandb_run.py:_finish():2383] finishing run deepnevro-deepnevro/research1/ucwxn3gs +2026-08-14 21:29:01,917 INFO MainThread:1193850 [wandb_run.py:_atexit_cleanup():2588] got exitcode: 0 +2026-08-14 21:29:01,918 INFO MainThread:1193850 [wandb_run.py:_restore():2570] restore +2026-08-14 21:29:01,918 INFO MainThread:1193850 [wandb_run.py:_restore():2576] restore done +2026-08-14 21:29:02,580 INFO MainThread:1193850 [wandb_run.py:_footer_sync_info():3993] logging synced files diff --git a/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/run-ucwxn3gs.wandb b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/run-ucwxn3gs.wandb new file mode 100644 index 0000000000000000000000000000000000000000..f168c5482bd527ba5ff7f27b12d41c57a95870f5 Binary files /dev/null and b/zain/Activation/wandb/run-20260814_212820-ucwxn3gs/run-ucwxn3gs.wandb differ diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/output.log b/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/output.log new file mode 100644 index 0000000000000000000000000000000000000000..e1a7a38cdb67627d66e9fb4c93e8ec6d39b26769 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/output.log @@ -0,0 +1,26 @@ +[Override] Activated with value = -10.0 + 74%|███████▍ | 2600/3500 [00:37<00:32, 27.47it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '2.126', 'grad_norm': '2.097e+07', 'learning_rate': '3.42e-05', 'epoch': '0.1699', 'train/total_time_seconds': '8.147', 'train/time_per_step_avg': '0.4073', 'train/epoch_time_elapsed': '9.358', 'train/estimated_remaining_minutes': '0.0528', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '185.7', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.0009422', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09082', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2852', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5859', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '186.9', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '8.774e-05', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3457', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.332', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3457', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6777', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '1157', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '0.00235', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '0.4004', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '2.141', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-2.141', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '2.031', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '4.172', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '1122', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '-1.419e-05', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '0.3867', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '2.047', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-1.945', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '2.047', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '3.992', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '158.6', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '0.0004482', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '0.07764', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '0.4082', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '-0.4082', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '0.3984', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '0.8066', 'train/tensor_act_model_layers_0_mlp/norm': '158.6', 'train/tensor_act_model_layers_0_mlp/mean': '0.0004501', 'train/tensor_act_model_layers_0_mlp/std': '0.07764', 'train/tensor_act_model_layers_0_mlp/max_abs': '0.4062', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '-0.4062', 'train/tensor_act_model_layers_0_mlp/max': '0.3965', 'train/tensor_act_model_layers_0_mlp/range': '0.8027', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '262.8', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '0.0005379', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.1279', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs +{'loss': '2.141', 'grad_norm': '2.268e+07', 'learning_rate': '7.02e-05', 'epoch': '0.1712', 'train/total_time_seconds': '14.71', 'train/time_per_step_avg': '0.3678', 'train/epoch_time_elapsed': '16.94', 'train/estimated_remaining_minutes': '0.09266'} +{'loss': '3.505', 'grad_norm': '1.169e+08', 'learning_rate': '9e-05', 'epoch': '0.1726', 'train/total_time_seconds': '22.64', 'train/time_per_step_avg': '0.3773', 'train/epoch_time_elapsed': '25.98', 'train/estimated_remaining_minutes': '0.1385', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '185.8', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.0008545', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09082', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2852', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5859', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '187', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '-9.489e-05', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.332', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.332', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.2988', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6309', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '1203', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '0.004608', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '0.416', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '2.25', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-2.25', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '2.141', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '4.391', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '1166', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '0.001984', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '0.4023', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '2.016', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-2.016', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '1.992', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '4.008', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '179.7', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '0.02026', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '0.08545', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '0.5312', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '-0.4004', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '0.5312', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '0.9316', 'train/tensor_act_model_layers_0_mlp/norm': '179.7', 'train/tensor_act_model_layers_0_mlp/mean': '0.02026', 'train/tensor_act_model_layers_0_mlp/std': '0.08545', 'train/tensor_act_model_layers_0_mlp/max_abs': '0.5273', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '-0.3984', 'train/tensor_act_model_layers_0_mlp/max': '0.5273', 'train/tensor_act_model_layers_0_mlp/range': '0.9258', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '279.2', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '0.02014', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.1348', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '0.6562', 't +{'loss': '6.925', 'grad_norm': '2.674e+08', 'learning_rate': '9e-05', 'epoch': '0.1739', 'train/total_time_seconds': '27.23', 'train/time_per_step_avg': '0.3404', 'train/epoch_time_elapsed': '31.61', 'train/estimated_remaining_minutes': '0.1618'} +{'loss': '7.5', 'grad_norm': '5.306e+08', 'learning_rate': '9e-05', 'epoch': '0.1753', 'train/total_time_seconds': '31.8', 'train/time_per_step_avg': '0.318', 'train/epoch_time_elapsed': '37.15', 'train/estimated_remaining_minutes': '0.1834'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 15.65it/s] + 77%|███████▋ | 2700/3500 [01:06<03:40, 3.63it/s][transformers] TinyLlamaForCausalLM has generative capabilities, as `prepare_inputs_for_generation` is explicitly defined. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions. +{'loss': '7.739', 'grad_norm': '8.053e+08', 'learning_rate': '9e-05', 'epoch': '0.1766', 'train/total_time_seconds': '36.81', 'train/time_per_step_avg': '0.2866', 'train/epoch_time_elapsed': '43.48', 'train/estimated_remaining_minutes': '0.206', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '185.8', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.0008774', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09082', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2852', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5859', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '187', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '-9.108e-05', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.332', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.332', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3203', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6523', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '1339', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '0.009583', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '0.4629', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '2.172', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-2.094', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '2.172', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '4.266', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '1305', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '0.004028', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '0.4512', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '2.188', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-2.172', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '2.188', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '4.359', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '274.7', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '0.05664', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '0.1216', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '0.8711', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '-0.4453', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '0.8711', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '1.316', 'train/tensor_act_model_layers_0_mlp/norm': '274.1', 'train/tensor_act_model_layers_0_mlp/mean': '0.05664', 'train/tensor_act_model_layers_0_mlp/std': '0.1211', 'train/tensor_act_model_layers_0_mlp/max_abs': '0.8594', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '-0.4434', 'train/tensor_act_model_layers_0_mlp/max': '0.8594', 'train/tensor_act_model_layers_0_mlp/range': '1.303', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '354.1', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '0.0564', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.1631', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '0.9883', 'trai +{'loss': '7.862', 'grad_norm': '9.018e+08', 'learning_rate': '9e-05', 'epoch': '0.178', 'train/total_time_seconds': '41.34', 'train/time_per_step_avg': '0.2663', 'train/epoch_time_elapsed': '49.03', 'train/estimated_remaining_minutes': '0.2245'} +{'loss': '7.944', 'grad_norm': '7.885e+08', 'learning_rate': '9e-05', 'epoch': '0.1793', 'train/total_time_seconds': '46.34', 'train/time_per_step_avg': '0.237', 'train/epoch_time_elapsed': '55.13', 'train/estimated_remaining_minutes': '0.2439', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '186', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.001007', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09082', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2852', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5859', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '187.3', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '3.982e-05', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3457', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3105', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3457', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6562', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '1380', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '0.01282', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '0.4766', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '2.266', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-2.109', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '2.266', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '4.375', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '1344', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '0.004211', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '0.4648', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '2.328', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-2.141', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '2.328', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '4.469', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '309.4', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '0.06787', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '0.1348', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '0.9492', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '-0.5078', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '0.9492', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '1.457', 'train/tensor_act_model_layers_0_mlp/norm': '308.4', 'train/tensor_act_model_layers_0_mlp/mean': '0.06787', 'train/tensor_act_model_layers_0_mlp/std': '0.1348', 'train/tensor_act_model_layers_0_mlp/max_abs': '0.9336', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '-0.5039', 'train/tensor_act_model_layers_0_mlp/max': '0.9336', 'train/tensor_act_model_layers_0_mlp/range': '1.438', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '383.1', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '0.06787', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.1748', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '1.055', 'train +{'loss': '7.991', 'grad_norm': '6.208e+08', 'learning_rate': '9e-05', 'epoch': '0.1807', 'train/total_time_seconds': '50.93', 'train/time_per_step_avg': '0.237', 'train/epoch_time_elapsed': '60.79', 'train/estimated_remaining_minutes': '0.2597'} +{'loss': '8.019', 'grad_norm': '4.844e+08', 'learning_rate': '9e-05', 'epoch': '0.182', 'train/total_time_seconds': '55.52', 'train/time_per_step_avg': '0.2373', 'train/epoch_time_elapsed': '66.44', 'train/estimated_remaining_minutes': '0.2742'} + - If you're using `trust_remote_code=True`, you can get rid of this warning by loading the model with an auto class. See https://huggingface.co/docs/transformers/en/model_doc/auto#auto-classes + - If you are the owner of the model architecture code, please modify your model class such that it inherits from `GenerationMixin` (after `PreTrainedModel`, otherwise you'll get an exception). + - If you are not the owner of the model architecture class, please contact the model code owner to update it. +Writing model shards: 100%|██████████| 1/1 [00:00<00:00, 13.82it/s] + 80%|███████▉ | 2797/3500 [01:34<03:16, 3.57it/s], ?it/s] +{'loss': '8.051', 'grad_norm': '3.817e+08', 'learning_rate': '9e-05', 'epoch': '0.1834', 'train/total_time_seconds': '60.56', 'train/time_per_step_avg': '0.2375', 'train/epoch_time_elapsed': '72.75', 'train/estimated_remaining_minutes': '0.2894', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '185.9', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.0009193', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09082', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2852', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5859', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '187.1', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '-1.162e-05', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3555', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3555', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.3047', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6602', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '1488', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '0.0188', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '0.5117', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '3.141', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-3.141', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '2.406', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '5.547', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '1454', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '0.006104', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '0.5039', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '2.469', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-2.469', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '2.469', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '4.938', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '492', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '0.125', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '0.2051', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '1.398', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '-0.7461', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '1.398', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '2.145', 'train/tensor_act_model_layers_0_mlp/norm': '487.8', 'train/tensor_act_model_layers_0_mlp/mean': '0.1245', 'train/tensor_act_model_layers_0_mlp/std': '0.2031', 'train/tensor_act_model_layers_0_mlp/max_abs': '1.344', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '-0.7383', 'train/tensor_act_model_layers_0_mlp/max': '1.344', 'train/tensor_act_model_layers_0_mlp/range': '2.082', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '542.1', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '0.1245', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.2334', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '1.406', 'train/tenso +{'loss': '8.067', 'grad_norm': '3.02e+08', 'learning_rate': '9e-05', 'epoch': '0.1847', 'train/total_time_seconds': '65.12', 'train/time_per_step_avg': '0.2378', 'train/epoch_time_elapsed': '78.35', 'train/estimated_remaining_minutes': '0.301'} +{'loss': '8.077', 'grad_norm': '2.317e+08', 'learning_rate': '9e-05', 'epoch': '0.186', 'train/total_time_seconds': '70.12', 'train/time_per_step_avg': '0.2378', 'train/epoch_time_elapsed': '84.44', 'train/estimated_remaining_minutes': '0.3133', 'train/tensor_act_model_layers_0_residual_pre_attn/norm': '186.1', 'train/tensor_act_model_layers_0_residual_pre_attn/mean': '0.0008926', 'train/tensor_act_model_layers_0_residual_pre_attn/std': '0.09082', 'train/tensor_act_model_layers_0_residual_pre_attn/max_abs': '0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_pre_attn/min': '-0.3008', 'train/tensor_act_model_layers_0_residual_pre_attn/max': '0.2852', 'train/tensor_act_model_layers_0_residual_pre_attn/range': '0.5859', 'train/tensor_act_model_layers_0_residual_post_attn/norm': '187.2', 'train/tensor_act_model_layers_0_residual_post_attn/mean': '-3.07e-06', 'train/tensor_act_model_layers_0_residual_post_attn/std': '0.09131', 'train/tensor_act_model_layers_0_residual_post_attn/max_abs': '0.3262', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_residual_post_attn/min': '-0.3262', 'train/tensor_act_model_layers_0_residual_post_attn/max': '0.2949', 'train/tensor_act_model_layers_0_residual_post_attn/range': '0.6211', 'train/tensor_act_model_layers_0_mlp_gate_proj/norm': '1897', 'train/tensor_act_model_layers_0_mlp_gate_proj/mean': '0.03638', 'train/tensor_act_model_layers_0_mlp_gate_proj/std': '0.6523', 'train/tensor_act_model_layers_0_mlp_gate_proj/max_abs': '4.438', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_gate_proj/min': '-4.438', 'train/tensor_act_model_layers_0_mlp_gate_proj/max': '3.609', 'train/tensor_act_model_layers_0_mlp_gate_proj/range': '8.047', 'train/tensor_act_model_layers_0_mlp_up_proj/norm': '1877', 'train/tensor_act_model_layers_0_mlp_up_proj/mean': '0.01044', 'train/tensor_act_model_layers_0_mlp_up_proj/std': '0.6484', 'train/tensor_act_model_layers_0_mlp_up_proj/max_abs': '3.797', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_up_proj/min': '-3.797', 'train/tensor_act_model_layers_0_mlp_up_proj/max': '3.422', 'train/tensor_act_model_layers_0_mlp_up_proj/range': '7.219', 'train/tensor_act_model_layers_0_mlp_down_proj/norm': '2102', 'train/tensor_act_model_layers_0_mlp_down_proj/mean': '0.6133', 'train/tensor_act_model_layers_0_mlp_down_proj/std': '0.8242', 'train/tensor_act_model_layers_0_mlp_down_proj/max_abs': '4.688', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp_down_proj/min': '-0.8867', 'train/tensor_act_model_layers_0_mlp_down_proj/max': '4.688', 'train/tensor_act_model_layers_0_mlp_down_proj/range': '5.574', 'train/tensor_act_model_layers_0_mlp/norm': '1878', 'train/tensor_act_model_layers_0_mlp/mean': '0.5703', 'train/tensor_act_model_layers_0_mlp/std': '0.7188', 'train/tensor_act_model_layers_0_mlp/max_abs': '3.297', 'train/tensor_act_model_layers_0_mlp/frac_near_dtype_limit': '0', 'train/tensor_act_model_layers_0_mlp/frac_near_user_limit': '0', 'train/tensor_act_model_layers_0_mlp/min': '-0.8711', 'train/tensor_act_model_layers_0_mlp/max': '3.297', 'train/tensor_act_model_layers_0_mlp/range': '4.168', 'train/tensor_act_model_layers_0_residual_post_mlp/norm': '1897', 'train/tensor_act_model_layers_0_residual_post_mlp/mean': '0.5703', 'train/tensor_act_model_layers_0_residual_post_mlp/std': '0.7305', 'train/tensor_act_model_layers_0_residual_post_mlp/max_abs': '3.422', 'train/tensor_ +{'loss': '8.085', 'grad_norm': '1.646e+08', 'learning_rate': '9e-05', 'epoch': '0.1874', 'train/total_time_seconds': '74.72', 'train/time_per_step_avg': '0.2379', 'train/epoch_time_elapsed': '90.08', 'train/estimated_remaining_minutes': '0.3225'} diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/requirements.txt b/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/requirements.txt new file mode 100644 index 0000000000000000000000000000000000000000..f36b19895172d9637438a445ceacb71d09dc5d02 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/requirements.txt @@ -0,0 +1,149 @@ +asttokens==3.0.1 +comm==0.2.3 +debugpy==1.8.21 +decorator==5.3.1 +executing==2.2.1 +nest-asyncio==1.6.0 +parso==0.8.7 +platformdirs==4.11.0 +psutil==7.2.2 +ptyprocess==0.7.0 +pure_eval==0.2.3 +Pygments==2.20.0 +pyzmq==27.1.0 +setuptools==83.0.0 +six==1.17.0 +tornado==6.5.7 +traitlets==5.15.0 +fsspec==2026.4.0 +wcwidth==0.8.2 +ipython_pygments_lexers==1.1.1 +jedi==0.20.0 +jupyter_core==5.9.1 +matplotlib-inline==0.2.2 +pexpect==4.9.0 +prompt_toolkit==3.0.53 +python-dateutil==2.9.0.post0 +stack_data==0.6.3 +wheel==0.47.0 +jupyter_client==8.9.1 +pip==26.1.2 +ipython==9.15.0 +ipykernel==7.2.0 +threadpoolctl==3.6.0 +pyparsing==3.3.2 +typing_extensions==4.15.0 +Jinja2==3.1.6 +narwhals==2.24.0 +kiwisolver==1.5.0 +joblib==1.5.3 +fonttools==4.63.0 +cycler==0.12.1 +scipy==1.17.1 +pandas==3.0.5 +contourpy==1.3.3 +scikit-learn==1.9.0 +matplotlib==3.11.1 +urllib3==2.7.0 +tqdm==4.70.0 +idna==3.18 +charset-normalizer==3.4.9 +certifi==2026.7.22 +requests==2.34.2 +seaborn==0.13.2 +uv==0.12.0 +shellingham==1.5.4 +mpmath==1.3.0 +attrs==26.1.0 +hf-xet==1.5.2 +nvidia-nccl-cu12==2.21.5 +MarkupSafe==3.0.3 +regex==2026.7.19 +importlib_metadata==9.0.0 +httpcore==1.0.9 +annotated-doc==0.0.5 +multidict==6.7.1 +aiohttp==3.14.3 +aiosignal==1.4.0 +xxhash==3.8.1 +aiohappyeyeballs==2.7.1 +mdurl==0.1.2 +cuda-toolkit==13.0.3.0 +networkx==3.6.1 +PyYAML==6.0.3 +nvidia-cufile==1.15.1.6 +typer==0.27.0 +torchaudio==2.6.0+cu124 +rich==15.0.0 +nvidia-cufft-cu12==11.2.1.3 +h11==0.16.0 +dill==0.4.1 +cuda-pathfinder==1.6.0 +filelock==3.29.0 +nvidia-nvtx-cu12==12.4.127 +httpx==0.28.1 +anyio==4.14.2 +numpy==2.4.4 +yarl==1.24.5 +click==8.4.2 +triton==3.2.0 +frozenlist==1.8.0 +zipp==4.1.0 +propcache==0.5.2 +tokenizers==0.22.2 +markdown-it-py==4.2.0 +nvidia-cuda-runtime==13.0.96 +cuda-bindings==13.3.1 +nvidia-cuda-cupti==13.0.85 +torch==2.6.0+cu124 +multiprocess==0.70.19 +pillow==12.2.0 +transformers==5.16.0.dev0 +wandb==0.28.1 +nvidia-curand==10.4.0.35 +sympy==1.13.1 +nvidia-cusparse==12.6.3.3 +nvidia-cuda-nvrtc==13.0.88 +typing-inspection==0.4.2 +nvidia-cusolver==12.0.4.66 +nvidia-cufft==12.0.0.61 +nvidia-cudnn-cu13==9.20.0.48 +nvidia-cublas==13.1.1.3 +pyarrow==25.0.0 +evaluate==0.4.6 +diffusers==0.39.0 +pydantic==2.13.4 +annotated-types==0.8.0 +protobuf==7.35.1 +sentry-sdk==2.66.1 +einops==0.8.2 +packaging==26.2 +nvidia-nvjitlink-cu12==12.4.127 +nvidia-curand-cu12==10.3.5.147 +nvidia-cusparselt-cu12==0.6.2 +nvidia-cusparse-cu12==12.3.1.170 +nvidia-cuda-runtime-cu12==12.4.127 +torchvision==0.21.0+cu124 +nvidia-cuda-nvrtc-cu12==12.4.127 +nvidia-cuda-cupti-cu12==12.4.127 +nvidia-cusolver-cu12==11.6.1.9 +nvidia-cublas-cu12==12.4.5.8 +nvidia-cudnn-cu12==9.1.0.70 +huggingface_hub==1.26.0 +datasets==5.0.1 +safetensors==0.8.0 +accelerate==1.14.0 +pydantic_core==2.46.4 +ninja==1.13.0 +autocommand==2.2.2 +backports.tarfile==1.2.0 +importlib_metadata==8.7.1 +jaraco.text==4.0.0 +jaraco.context==6.1.0 +jaraco.functools==4.4.0 +more-itertools==10.8.0 +packaging==26.0 +platformdirs==4.4.0 +tomli==2.4.0 +wheel==0.46.3 +zipp==3.23.0 diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/wandb-metadata.json b/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/wandb-metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..3ce82630e9f209b06b3065ff8ba5ff41af6e8057 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/files/wandb-metadata.json @@ -0,0 +1,99 @@ +{ + "os": "Linux-5.15.0-126-generic-x86_64-with-glibc2.35", + "python": "CPython 3.11.15", + "startedAt": "2026-08-14T21:28:35.342440Z", + "args": [ + "--config", + "/mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/configs/baseline100l.yaml", + "--variants", + "glu-relu", + "glu-gelu", + "glu-sigmoid", + "glu-waleed10", + "glu-silu-waleed10" + ], + "program": "/mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/sweep.py", + "codePath": "sweep.py", + "codePathLocal": "sweep.py", + "git": { + "remote": "https://github.com/w-ahmad1a10/Activation.git", + "commit": "3fb47c6943d6927eafc39bd399a3d3a520bae74a" + }, + "email": "deepnevro@gmail.com", + "root": "/mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation", + "host": "deeplens-k3s-node1", + "executable": "/mnt/data/zainulabideen/zain-exp/notebooks/my_env/bin/python", + "cpu_count": 112, + "cpu_count_logical": 224, + "gpu": "NVIDIA H100 80GB HBM3", + "gpu_count": 8, + "disk": { + "/": { + "total": "1560765693952", + "used": "709055246336" + } + }, + "memory": { + "total": "2164089937920" + }, + "gpu_nvidia": [ + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-39c684a5-fde6-83d7-1663-0859795881ae" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-68012e5a-38b6-b643-0ca6-62fb66720bf3" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-132944c4-b689-2b5f-89a4-d730401677ab" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-2df386cc-6d26-d0e2-7a2d-a057b0d95864" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-bfa16575-1d94-1aa2-4537-2c93433f42ef" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-bc6c3e3c-9b90-09ca-c034-774961847c54" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-00a441e1-7c95-e7d6-4c35-43d6b291aea9" + }, + { + "name": "NVIDIA H100 80GB HBM3", + "memoryTotal": "85520809984", + "cudaCores": 16896, + "architecture": "Hopper", + "uuid": "GPU-1c4d29a2-4647-6fce-d8fc-0c5ecfbbd6ea" + } + ], + "cudaVersion": "12.4", + "writerId": "pf2h3cdzqvkrfjeqx7v0rn4hp1f52wde" +} \ No newline at end of file diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-core.log b/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-core.log new file mode 100644 index 0000000000000000000000000000000000000000..59acf96f4c751e74e9217cb996ef31c503e3590c --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-core.log @@ -0,0 +1,77 @@ +{"time":"2026-08-14T19:38:51.636058907Z","level":"INFO","msg":"main: starting server","port-filename":"/tmp/tmp54729boe/port-1375539.txt","pid":1375539,"detached":false,"idle-timeout":600000000000,"log-level":0,"disable-analytics":false,"shutdown-on-parent-exit":false,"enable-dcgm-profiling":false} +{"time":"2026-08-14T19:38:51.639970993Z","level":"INFO","msg":"server: will exit if parent process dies","ppid":1375539} +{"time":"2026-08-14T19:38:51.639940209Z","level":"INFO","msg":"server: accepting connections","addr":{"Name":"/tmp/wandb-1375539-170701-251437063/socket","Net":"unix"}} +{"time":"2026-08-14T19:38:51.812900736Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"1(@)"} +{"time":"2026-08-14T19:39:07.276688434Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"2(@)"} +{"time":"2026-08-14T19:39:07.370077387Z","level":"INFO","msg":"handleInformInit: received","streamId":"kk1xbqht","id":"2(@)"} +{"time":"2026-08-14T19:39:07.640206016Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"kk1xbqht","id":"2(@)"} +{"time":"2026-08-14T19:39:13.074249733Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"zj3lg1k64oen"} +{"time":"2026-08-14T19:51:49.732364673Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"zj3lg1k64oen"} +{"time":"2026-08-14T19:51:50.223490459Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"zj3lg1k64oen"} +{"time":"2026-08-14T19:51:50.225275352Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"kk1xbqht","id":"2(@)"} +{"time":"2026-08-14T19:51:50.225791347Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"kk1xbqht","id":"2(@)"} +{"time":"2026-08-14T19:51:51.140595835Z","level":"INFO","msg":"handleInformInit: received","streamId":"ix6wcbwv","id":"2(@)"} +{"time":"2026-08-14T19:51:51.407779103Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"ix6wcbwv","id":"2(@)"} +{"time":"2026-08-14T19:51:56.793135206Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"m2u1nqrk4r4j"} +{"time":"2026-08-14T20:05:01.389026301Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"m2u1nqrk4r4j"} +{"time":"2026-08-14T20:05:01.891785946Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"m2u1nqrk4r4j"} +{"time":"2026-08-14T20:05:01.893356932Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"ix6wcbwv","id":"2(@)"} +{"time":"2026-08-14T20:05:01.894067235Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"ix6wcbwv","id":"2(@)"} +{"time":"2026-08-14T20:05:03.038763184Z","level":"INFO","msg":"handleInformInit: received","streamId":"bdhnr27j","id":"2(@)"} +{"time":"2026-08-14T20:05:03.299131712Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"bdhnr27j","id":"2(@)"} +{"time":"2026-08-14T20:05:08.644597551Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"u5vxhmn1tauv"} +{"time":"2026-08-14T20:21:26.598738734Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"u5vxhmn1tauv"} +{"time":"2026-08-14T20:21:27.243676424Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"u5vxhmn1tauv"} +{"time":"2026-08-14T20:21:27.245262581Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"bdhnr27j","id":"2(@)"} +{"time":"2026-08-14T20:21:27.245859065Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"bdhnr27j","id":"2(@)"} +{"time":"2026-08-14T20:21:28.155191217Z","level":"INFO","msg":"handleInformInit: received","streamId":"twgo1e89","id":"2(@)"} +{"time":"2026-08-14T20:21:28.414618766Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"twgo1e89","id":"2(@)"} +{"time":"2026-08-14T20:21:33.76572887Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"smlkpmpk76rp"} +{"time":"2026-08-14T20:38:01.401128169Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"smlkpmpk76rp"} +{"time":"2026-08-14T20:38:02.166866415Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"smlkpmpk76rp"} +{"time":"2026-08-14T20:38:02.168373578Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"twgo1e89","id":"2(@)"} +{"time":"2026-08-14T20:38:02.169204491Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"twgo1e89","id":"2(@)"} +{"time":"2026-08-14T20:38:03.474631884Z","level":"INFO","msg":"handleInformInit: received","streamId":"7p5140bq","id":"2(@)"} +{"time":"2026-08-14T20:38:03.739404113Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"7p5140bq","id":"2(@)"} +{"time":"2026-08-14T20:38:09.1668666Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"9gr96sam7ydd"} +{"time":"2026-08-14T20:54:40.077063626Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"9gr96sam7ydd"} +{"time":"2026-08-14T20:54:40.77979146Z","level":"INFO","msg":"connection: cancelling request","id":"2(@)","requestId":"9gr96sam7ydd"} +{"time":"2026-08-14T20:54:40.781496498Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"7p5140bq","id":"2(@)"} +{"time":"2026-08-14T20:54:40.782279361Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"7p5140bq","id":"2(@)"} +{"time":"2026-08-14T20:54:42.400855251Z","level":"INFO","msg":"connection: closing","id":"2(@)"} +{"time":"2026-08-14T20:54:42.40094641Z","level":"INFO","msg":"connection: closed successfully","id":"2(@)"} +{"time":"2026-08-14T20:54:42.400864953Z","level":"INFO","msg":"processOutgoingData: finished","id":"2(@)"} +{"time":"2026-08-14T20:54:42.400960827Z","level":"INFO","msg":"connection: ManageConnectionData: connection closed","id":"2(@)"} +{"time":"2026-08-14T21:01:09.497626422Z","level":"INFO","msg":"connection: ManageConnectionData: new connection created","id":"3(@)"} +{"time":"2026-08-14T21:01:09.714752804Z","level":"INFO","msg":"handleInformInit: received","streamId":"qauftiuh","id":"3(@)"} +{"time":"2026-08-14T21:01:09.994191533Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"qauftiuh","id":"3(@)"} +{"time":"2026-08-14T21:01:15.507216386Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"wd49dhcig506"} +{"time":"2026-08-14T21:07:53.011770412Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"wd49dhcig506"} +{"time":"2026-08-14T21:07:54.661515353Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"wd49dhcig506"} +{"time":"2026-08-14T21:07:54.926288764Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"qauftiuh","id":"3(@)"} +{"time":"2026-08-14T21:07:54.927423677Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"qauftiuh","id":"3(@)"} +{"time":"2026-08-14T21:08:09.037499672Z","level":"INFO","msg":"handleInformInit: received","streamId":"a6574zx8","id":"3(@)"} +{"time":"2026-08-14T21:08:09.296770198Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"a6574zx8","id":"3(@)"} +{"time":"2026-08-14T21:08:09.806890437Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"bsqpd0c29zk9"} +{"time":"2026-08-14T21:08:14.888124439Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"ilk9hnwtzxxz"} +{"time":"2026-08-14T21:14:51.274587185Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"ilk9hnwtzxxz"} +{"time":"2026-08-14T21:14:52.85240857Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"ilk9hnwtzxxz"} +{"time":"2026-08-14T21:14:52.873645397Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"a6574zx8","id":"3(@)"} +{"time":"2026-08-14T21:14:52.874311484Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"a6574zx8","id":"3(@)"} +{"time":"2026-08-14T21:15:07.169254211Z","level":"INFO","msg":"handleInformInit: received","streamId":"54wnyexz","id":"3(@)"} +{"time":"2026-08-14T21:15:07.435699541Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"54wnyexz","id":"3(@)"} +{"time":"2026-08-14T21:15:12.853675736Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"fzqy9cymfofj"} +{"time":"2026-08-14T21:21:52.842687654Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"fzqy9cymfofj"} +{"time":"2026-08-14T21:21:54.009681636Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"fzqy9cymfofj"} +{"time":"2026-08-14T21:21:54.257319168Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"54wnyexz","id":"3(@)"} +{"time":"2026-08-14T21:21:54.257934554Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"54wnyexz","id":"3(@)"} +{"time":"2026-08-14T21:22:08.718181031Z","level":"INFO","msg":"handleInformInit: received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:22:08.976795469Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:22:14.472169125Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:18.727248774Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.378913364Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"qlc6t75fdq0u"} +{"time":"2026-08-14T21:28:20.400345521Z","level":"INFO","msg":"handleInformFinish: finish message received","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:20.400969576Z","level":"INFO","msg":"handleInformFinish: stream closed","streamId":"n6fjpg9o","id":"3(@)"} +{"time":"2026-08-14T21:28:35.344777056Z","level":"INFO","msg":"handleInformInit: received","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:35.603024157Z","level":"INFO","msg":"handleInformInit: stream started","streamId":"gqskne0x","id":"3(@)"} +{"time":"2026-08-14T21:28:41.078893683Z","level":"INFO","msg":"connection: cancelling request","id":"3(@)","requestId":"yxbo55z35lse"} diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-internal.log b/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-internal.log new file mode 100644 index 0000000000000000000000000000000000000000..70b55dc30b3bb82a035e6dee6e50a99f3b0291fd --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-internal.log @@ -0,0 +1,47 @@ +{"time":"2026-08-14T21:28:35.344927311Z","level":"INFO","msg":"wandb-core"} +{"time":"2026-08-14T21:28:35.34506265Z","level":"INFO","msg":"stream: starting","core version":"0.28.1"} +{"time":"2026-08-14T21:28:35.602855263Z","level":"INFO","msg":"stream: created new stream","id":"gqskne0x"} +{"time":"2026-08-14T21:28:35.602933409Z","level":"INFO","msg":"handler: started"} +{"time":"2026-08-14T21:28:35.603016258Z","level":"INFO","msg":"stream: started"} +{"time":"2026-08-14T21:28:35.603043817Z","level":"INFO","msg":"writer: started","stream_id":"gqskne0x"} +{"time":"2026-08-14T21:28:35.60305765Z","level":"INFO","msg":"sender: started"} +{"time":"2026-08-14T21:28:36.090276496Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1} +{"time":"2026-08-14T21:28:36.211652827Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:51.09126529Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":0,"events_lines":1,"console_offset":0,"console_lines":3,"uploaded_len":2} +{"time":"2026-08-14T21:28:51.261717283Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:28:53.048390257Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":3104} +{"time":"2026-08-14T21:28:53.048417504Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":1} +{"time":"2026-08-14T21:28:53.050441969Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":3665} +{"time":"2026-08-14T21:28:53.050479366Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":1} +{"time":"2026-08-14T21:28:53.051308024Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":3981} +{"time":"2026-08-14T21:28:53.052248742Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":269} +{"time":"2026-08-14T21:28:53.052348345Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":4293} +{"time":"2026-08-14T21:28:53.053710973Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":419} +{"time":"2026-08-14T21:28:53.054973484Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":5282} +{"time":"2026-08-14T21:28:53.05654974Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":576} +{"time":"2026-08-14T21:28:53.057424589Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":6166} +{"time":"2026-08-14T21:28:53.058161238Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":225} +{"time":"2026-08-14T21:28:53.059182778Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":6765} +{"time":"2026-08-14T21:28:53.059372384Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":41} +{"time":"2026-08-14T21:28:53.059719473Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":6917} +{"time":"2026-08-14T21:28:53.059850702Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":21} +{"time":"2026-08-14T21:28:53.061202591Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":7342} +{"time":"2026-08-14T21:28:53.061276823Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":12} +{"time":"2026-08-14T21:28:53.068366857Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":9009} +{"time":"2026-08-14T21:28:53.068403516Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":1} +{"time":"2026-08-14T21:28:53.070211602Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":9602} +{"time":"2026-08-14T21:28:53.071216814Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":294} +{"time":"2026-08-14T21:28:53.071521294Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":10000} +{"time":"2026-08-14T21:28:53.072106936Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":190} +{"time":"2026-08-14T21:28:53.075408887Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":11406} +{"time":"2026-08-14T21:28:53.076614643Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":333} +{"time":"2026-08-14T21:29:06.110603465Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":2,"events_offset":1,"events_lines":2,"console_offset":1,"console_lines":1} +{"time":"2026-08-14T21:29:06.83621196Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:29:21.108433332Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":3,"history_lines":3,"events_offset":3,"events_lines":2,"console_offset":1,"console_lines":1} +{"time":"2026-08-14T21:29:21.999015678Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:29:36.110784946Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":2,"events_offset":5,"events_lines":2,"console_offset":3,"console_lines":12} +{"time":"2026-08-14T21:29:36.946170207Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:29:51.109458297Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":8,"history_lines":3,"events_offset":7,"events_lines":2,"console_offset":11,"console_lines":1} +{"time":"2026-08-14T21:29:52.226871571Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} +{"time":"2026-08-14T21:30:06.110393089Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":11,"history_lines":2,"events_offset":9,"events_lines":2,"console_offset":15,"console_lines":10} +{"time":"2026-08-14T21:30:07.570174219Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"} diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug.log b/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug.log new file mode 100644 index 0000000000000000000000000000000000000000..b43ac43447df0773671fa75307878dd4d6210cf9 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug.log @@ -0,0 +1,20 @@ +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:setup_run_log_directory():729] Logging user logs to /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug.log +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:setup_run_log_directory():730] Logging internal logs to /mnt/data/zainulabideen/zain-exp/notebooks/zain/Activation/wandb/run-20260814_212835-gqskne0x/logs/debug-internal.log +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:init():772] calling init triggers +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:init():777] wandb.init called with sweep_config: {} +config: {'_wandb': {}} +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:init():820] starting backend +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:init():826] Connected to an existing wandb-core service via WANDB_SERVICE +2026-08-14 21:28:35,343 INFO MainThread:936813 [wandb_init.py:init():835] sending inform_init request +2026-08-14 21:28:35,603 INFO MainThread:936813 [wandb_init.py:init():840] backend started and connected +2026-08-14 21:28:35,605 INFO MainThread:936813 [wandb_init.py:init():910] updated telemetry +2026-08-14 21:28:35,614 INFO MainThread:936813 [wandb_init.py:init():933] communicating run to backend with 90.0 second timeout +2026-08-14 21:28:35,996 INFO MainThread:936813 [wandb_init.py:init():978] starting run threads in backend +2026-08-14 21:28:36,075 INFO MainThread:936813 [wandb_run.py:_console_start():2621] atexit reg +2026-08-14 21:28:36,075 INFO MainThread:936813 [wandb_run.py:_redirect():2471] redirect: wrap_raw +2026-08-14 21:28:36,076 INFO MainThread:936813 [wandb_run.py:_redirect():2540] Wrapping output streams. +2026-08-14 21:28:36,076 INFO MainThread:936813 [wandb_run.py:_redirect():2563] Redirects installed. +2026-08-14 21:28:36,077 INFO MainThread:936813 [wandb_init.py:init():1016] run started, returning control to user process +2026-08-14 21:28:36,078 INFO MainThread:936813 [wandb_run.py:_config_callback():1346] config_cb None None {'transformers_version': '5.16.0.dev0', 'architectures': None, 'output_hidden_states': False, 'return_dict': True, 'dtype': None, 'chunk_size_feed_forward': 0, 'is_encoder_decoder': False, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'problem_type': None, 'vocab_size': 4096, 'hidden_size': 128, 'intermediate_size': 256, 'num_hidden_layers': 100, 'num_attention_heads': 4, 'num_key_value_heads': 4, 'hidden_act': 'silu', 'max_position_embeddings': 512, 'initializer_range': 0.02, 'rms_norm_eps': 1e-06, 'use_cache': False, 'pad_token_id': 0, 'bos_token_id': 1, 'eos_token_id': 2, 'pretraining_tp': 1, 'tie_word_embeddings': True, 'rope_parameters': {'rope_theta': 10000.0, 'rope_type': 'default'}, 'attention_bias': False, 'attention_dropout': 0.0, 'mlp_bias': False, 'head_dim': 32, '_name_or_path': '', 'tokenizer_name': 'w-ahmad/tiny-stories-tokenizer', 'mlp_type': 'glu', 'activation': 'silu-waleed10', 'waleed_beta': 4, 'model_type': 'tiny_llama', 'output_attentions': False, 'output_dir': 'out/glu-silu-waleed10-100L_trash_run', 'per_device_train_batch_size': 64, 'num_train_epochs': 1, 'max_steps': 3500, 'learning_rate': 9e-05, 'lr_scheduler_type': 'constant_with_warmup', 'lr_scheduler_kwargs': None, 'warmup_steps': 50, 'optim': 'adamw_torch_fused', 'optim_args': None, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'optim_target_modules': None, 'gradient_accumulation_steps': 1, 'average_tokens_across_devices': True, 'max_grad_norm': 1.0, 'label_smoothing_factor': 0.0, 'bf16': True, 'fp16': False, 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': None, 'gradient_checkpointing': False, 'gradient_checkpointing_kwargs': None, 'torch_compile': False, 'torch_compile_backend': None, 'torch_compile_mode': None, 'use_liger_kernel': False, 'liger_kernel_config': None, 'neftune_noise_alpha': None, 'torch_empty_cache_steps': None, 'auto_find_batch_size': False, 'logging_strategy': 'steps', 'logging_steps': 20, 'logging_first_step': False, 'log_on_each_node': True, 'logging_nan_inf_filter': True, 'include_num_input_tokens_seen': 'no', 'log_level': 'passive', 'log_level_replica': 'warning', 'disable_tqdm': False, 'report_to': ['wandb'], 'run_name': 'LM-glu-silu-waleed10-100L-16.9M-20260814-212834', 'project': 'huggingface', 'trackio_space_id': None, 'trackio_bucket_id': None, 'trackio_static_space_id': None, 'eval_strategy': 'steps', 'eval_steps': 2498, 'eval_delay': 0, 'per_device_eval_batch_size': 1024, 'prediction_loss_only': False, 'eval_on_start': False, 'eval_do_concat_batches': True, 'eval_use_gather_object': False, 'eval_accumulation_steps': None, 'include_for_metrics': [], 'batch_eval_metrics': False, 'save_only_model': False, 'save_strategy': 'steps', 'save_steps': 50, 'save_on_each_node': False, 'save_total_limit': None, 'enable_jit_checkpoint': False, 'push_to_hub': False, 'hub_token': '', 'hub_private_repo': None, 'hub_model_id': 'w-ahmad/6L-glu-silu-waleed10-100L', 'hub_strategy': 'every_save', 'hub_always_push': False, 'hub_revision': None, 'load_best_model_at_end': False, 'metric_for_best_model': None, 'greater_is_better': None, 'ignore_data_skip': False, 'restore_callback_states_from_checkpoint': False, 'full_determinism': False, 'seed': 42, 'data_seed': 42, 'use_cpu': False, 'accelerator_config': {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}, 'parallelism_config': None, 'dataloader_drop_last': False, 'dataloader_num_workers': 0, 'dataloader_pin_memory': True, 'dataloader_persistent_workers': False, 'dataloader_prefetch_factor': None, 'dataloader_multiprocessing_context': None, 'dataloader_in_order': True, 'remove_unused_columns': False, 'label_names': None, 'train_sampling_strategy': 'random', 'length_column_name': 'length', 'ddp_find_unused_parameters': None, 'ddp_bucket_cap_mb': None, 'ddp_broadcast_buffers': None, 'ddp_static_graph': None, 'ddp_backend': None, 'ddp_timeout': 1800, 'fsdp': None, 'fsdp_config': None, 'deepspeed': None, 'debug': [], 'skip_memory_metrics': True, 'do_train': False, 'do_eval': True, 'do_predict': False, 'resume_from_checkpoint': None, 'local_rank': -1} +2026-08-14 21:28:36,083 INFO MainThread:936813 [wandb_config.py:__setitem__():155] [no run ID] config set model/num_parameters = 16934016 - > +2026-08-14 21:28:36,083 INFO MainThread:936813 [wandb_run.py:_config_callback():1346] config_cb model/num_parameters 16934016 None diff --git a/zain/Activation/wandb/run-20260814_212835-gqskne0x/run-gqskne0x.wandb b/zain/Activation/wandb/run-20260814_212835-gqskne0x/run-gqskne0x.wandb new file mode 100644 index 0000000000000000000000000000000000000000..0383343d7a849462ff40ea01d7bcea2173ff1a55 --- /dev/null +++ b/zain/Activation/wandb/run-20260814_212835-gqskne0x/run-gqskne0x.wandb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:642685d115792bd6ab89f27f6b3b3941e2759dd1c0b094ae917478acfdb1e3ac +size 14843904