tacit

0 Followers · 0 Following

• Joined on 2024-05-31

Repositories

3

Projects Packages Public Activity Starred Repositories

tacit synced and deleted reference refs/tags/refs/pull/90/merge at tacit/nanochat from mirror 2025-12-09 13:52:17 +00:00

tacit synced commits to refs/pull/151/merge at tacit/nanochat from mirror 2025-12-09 13:52:17 +00:00

d14d5630dd Merge 7f3154f025 into d5759400f9

d5759400f9 fixing two typos in comments

e72c3299df fix random.seed() footgun bug for SpellingBee data generation

7931e0903a rename checkpoint_dir to checkpoints_dir for consistency.

849d95ae1f remove unnecessary check to make the logic in CausalSelfAttention.forward() clearer

Compare 28 commits »

tacit synced commits to refs/pull/312/merge at tacit/nanochat from mirror 2025-12-09 05:42:24 +00:00

012160b252 Merge 26b0941f75 into 39cccc527f

39cccc527f small bugfix make mid_train script work even with a tiny number of iterations

8b1cecaa95 Apply suggestion from @svlandeg for nicer looking comparison

58f3e84e01 clean up train/val loader in sft for consistency with mid/base

1b2a675c88 Improve KV cache code readability

Compare 20 commits »

tacit synced commits to refs/pull/348/merge at tacit/nanochat from mirror 2025-12-09 05:42:24 +00:00

22a9631444 Merge e50896dcdc into 7931e0903a

7931e0903a rename checkpoint_dir to checkpoints_dir for consistency.

849d95ae1f remove unnecessary check to make the logic in CausalSelfAttention.forward() clearer

39cccc527f small bugfix make mid_train script work even with a tiny number of iterations

8b1cecaa95 Apply suggestion from @svlandeg for nicer looking comparison

Compare 24 commits »

tacit synced commits to refs/pull/324/merge at tacit/nanochat from mirror 2025-12-09 05:42:24 +00:00

0bab2629ba Merge 861cbce2e9 into d5759400f9

d5759400f9 fixing two typos in comments

e72c3299df fix random.seed() footgun bug for SpellingBee data generation

7931e0903a rename checkpoint_dir to checkpoints_dir for consistency.

849d95ae1f remove unnecessary check to make the logic in CausalSelfAttention.forward() clearer

Compare 28 commits »

tacit synced and deleted reference refs/tags/refs/pull/351/merge at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

tacit synced and deleted reference refs/tags/refs/pull/358/merge at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

tacit synced commits to refs/pull/309/head at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

8b1cecaa95 Apply suggestion from @svlandeg for nicer looking comparison

tacit synced commits to refs/pull/256/merge at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

aab1839bad Merge 337d9649c6 into cbf30c842c

cbf30c842c apply float32 cast before logits softcapping so the tanh is in fp32. torch compile fuses this correctly with no extra memory costs.

90442de35f fix bug where any rank has to be able to create checkpoint_dir if saving optim

2fd0440355 fix: missing val_bpb on resume

16788eed3c fix(model): apply float32 cast before logits softcapping

Compare 6 commits »

tacit synced commits to master at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

d5759400f9 fixing two typos in comments

e72c3299df fix random.seed() footgun bug for SpellingBee data generation

7931e0903a rename checkpoint_dir to checkpoints_dir for consistency.

849d95ae1f remove unnecessary check to make the logic in CausalSelfAttention.forward() clearer

39cccc527f small bugfix make mid_train script work even with a tiny number of iterations

Compare 24 commits »

tacit synced commits to refs/pull/161/merge at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

b425f90dae Merge 0a1059d571 into d5759400f9

d5759400f9 fixing two typos in comments

e72c3299df fix random.seed() footgun bug for SpellingBee data generation

7931e0903a rename checkpoint_dir to checkpoints_dir for consistency.

849d95ae1f remove unnecessary check to make the logic in CausalSelfAttention.forward() clearer

Compare 28 commits »

tacit synced and deleted reference refs/tags/refs/pull/361/merge at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

tacit synced commits to refs/pull/311/merge at tacit/nanochat from mirror 2025-12-09 05:42:23 +00:00

0c0df08f1e Merge 81a958350a into cbf30c842c

cbf30c842c apply float32 cast before logits softcapping so the tanh is in fp32. torch compile fuses this correctly with no extra memory costs.

90442de35f fix bug where any rank has to be able to create checkpoint_dir if saving optim

2fd0440355 fix: missing val_bpb on resume

16788eed3c fix(model): apply float32 cast before logits softcapping

Compare 6 commits »

tacit synced and deleted reference refs/tags/refs/pull/326/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00

tacit synced and deleted reference refs/tags/refs/pull/327/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00

tacit synced and deleted reference refs/tags/refs/pull/342/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00

tacit synced and deleted reference refs/tags/refs/pull/325/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00

tacit synced and deleted reference refs/tags/refs/pull/317/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00

tacit synced and deleted reference refs/tags/refs/pull/345/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00

tacit synced and deleted reference refs/tags/refs/pull/306/merge at tacit/nanochat from mirror 2025-12-09 05:42:22 +00:00