tacit

tacit synced commits to refs/pull/328/merge at tacit/nanochat from mirror 2026-02-11 13:09:54 +00:00

77829af59b Merge 02b22a5a13 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

1ec0a34779 at 28 and above we start to need batch size 8

ff46300720 tune miniseries just a bit, fairly cosmetic, keep to even depths where the math works out nicely in model sizing

Compare 7 commits »

tacit synced commits to refs/pull/442/merge at tacit/nanochat from mirror 2026-02-11 13:09:54 +00:00

979846811c Merge 3bf06802e6 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/151/merge at tacit/nanochat from mirror 2026-02-11 13:09:53 +00:00

ed9dc84caa Merge 7f3154f025 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

1ec0a34779 at 28 and above we start to need batch size 8

ff46300720 tune miniseries just a bit, fairly cosmetic, keep to even depths where the math works out nicely in model sizing

Compare 7 commits »

tacit synced commits to refs/pull/141/merge at tacit/nanochat from mirror 2026-02-11 13:09:52 +00:00

210c16221f Merge 65865df300 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 3 commits »

tacit synced commits to refs/pull/492/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

bd48f2b18d Merge 7ac837cff8 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/501/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

bbe1093ac6 Merge 005daea668 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/516/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

9061480975 Merge 00932d1955 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to master at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

2f09686724 clarify that this is bf16 mfu we're talking about

tacit synced commits to refs/pull/512/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

0efaa80db3 Merge 2e0fda1893 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/519/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

ba12bf47a4 Merge c4a183dfef into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/437/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

764b02621a Merge 8cfa0451f4 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/513/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

0c73e53aaa Merge 23acb17f17 into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/520/merge at tacit/nanochat from mirror 2026-02-11 04:59:51 +00:00

919f2a6484 Merge 1bf1fdaa0d into 2f09686724

2f09686724 clarify that this is bf16 mfu we're talking about

Compare 2 commits »

tacit synced commits to refs/pull/519/merge at tacit/nanochat from mirror 2026-02-10 20:50:00 +00:00

df77f21819 Merge c4a183dfef into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »

tacit synced commits to refs/pull/520/merge at tacit/nanochat from mirror 2026-02-10 20:50:00 +00:00

ac324a46ea Merge 1bf1fdaa0d into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »

tacit synced commits to refs/pull/513/merge at tacit/nanochat from mirror 2026-02-10 20:49:59 +00:00

254b2f6875 Merge 23acb17f17 into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »

tacit synced commits to refs/pull/511/merge at tacit/nanochat from mirror 2026-02-10 20:49:59 +00:00

e778b2e412 Merge 5ba77a31b9 into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »

tacit synced commits to refs/pull/515/merge at tacit/nanochat from mirror 2026-02-10 20:49:59 +00:00

dbd1e7e70b Merge 2ae28292aa into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »

tacit synced commits to refs/pull/512/merge at tacit/nanochat from mirror 2026-02-10 20:49:59 +00:00

4ea77e8ae2 Merge 2e0fda1893 into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »

tacit synced commits to refs/pull/516/merge at tacit/nanochat from mirror 2026-02-10 20:49:59 +00:00

8abd890d6d Merge 00932d1955 into e569b59f92

e569b59f92 delete torchao dependency, create our own exact API-matched version of Float8Linear, document it very well. for some poorly understood reason, the performance is not only ~identical but actually runs 3% faster. despite of it being significantly simpler and much less code. i don't fully understand why/how atm

Compare 2 commits »