Repository Issue Activity (beta)

#810 chat_rl: the sampled assistant_end token is masked out, so stopping is never reinforced

#825 SDPA fallback: sliding window doesn't reduce memory (full mask built regardless of window size)

#824 I made a smaller AI.

#698 RTX 3050TI Not detecting m.get_device()

#737 Use Official PyTorch FA3 Builds

#820 FA3 loader reports success on Blackwell (sm_120) then dies at first kernel launch

#756 Token smearing does not correctly handle chunked prefill / chunk inference (first token of a chunk misses its cross-chunk predecessor

#743 Question: Normalization blends positional information?

#284 Add support for TPU?

#804 Add support for OpenCL (e.g., for the integrated AMD GPU)?

#763 Stale fwe-train / fwe-val labels in scripts/tok_eval.py after the FineWeb-Edu → ClimbMix dataset switch

#532 base_train.py crashes with raw AssertionError on small batch sizes instead of auto-aligning or providing actionable error

#685 MPS BF16 support

#791 TODO: hash for d34 model

#590 [Bug] scripts/chat_sft.py produces loss: nan from step 00001 on small device-batch-size (≤8) due to fully-masked micro-batches

#772 Reproduce the performance of GPT2

#769 Go matrix

#427 base_eval.py: hellaswag gets progressively slower and leaks memory on small models (Mac Studio)

#752 Flash Attention module has some improvements for better performance and error handling

#749 Deterministic 3D Correspondence in raw CUDA: Solving 14-year Symmetry Ambiguity

#578 ZeroDivisionError in scaling law math when --target-param-data-ratio is set to 0

#679 why don't we make a simpler version of nanochat suitable for CPU and cheaper hardware

#678 feat(agents): identity layer in agents msg — tag sender in messages

#529 speedrun.sh updates

#637 confusion on mmlu dataset