-
Notifications
You must be signed in to change notification settings - Fork 11.7k
Pull requests: karpathy/autoresearch
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Bump PyTorch to 2.11.0 (fixes silent expandable_segments bug in 2.9.1)
#588
opened May 11, 2026 by
srosro
Loading…
4 tasks done
feat(mps): Apple Silicon MPS port — Flash Attention 3 → PyTorch SDPA
#582
opened May 10, 2026 by
Rensp
Loading…
3 tasks
fix(train): size rotary cache to config.sequence_len, not 10x
#557
opened Apr 30, 2026 by
lonexreb
Loading…
2 of 3 tasks
fix(train): warmup off-by-one in steady_state_mfu accumulation
#556
opened Apr 30, 2026 by
lonexreb
Loading…
3 tasks done
fix(tokenizer): support multi-token prepend and tighten scope (#348)
#553
opened Apr 30, 2026 by
lonexreb
Loading…
8 tasks done
fix(prepare): use raw bytes for BPB to avoid U+FFFD inflation (#384)
#552
opened Apr 30, 2026 by
lonexreb
Loading…
1 of 3 tasks
fix(prepare): detect truncated shard downloads via Content-Length
#549
opened Apr 29, 2026 by
lonexreb
Loading…
1 of 3 tasks
fix(prepare): load token_bytes.pt with weights_only=True
#548
opened Apr 29, 2026 by
lonexreb
Loading…
1 of 3 tasks
fix: GPU-aware MFU with Blackwell (sm_100/sm_120) coverage
#547
opened Apr 29, 2026 by
lonexreb
Loading…
1 of 4 tasks
Minor cleanups: README typo and redundant .tmp check in prepare.py
#542
opened Apr 24, 2026 by
tempoo04
Loading…
feat: robust SDPA fallback, headless logging, and checkpointing optimizations
#507
opened Apr 10, 2026 by
yangcongcong-coding
Loading…
fix: remove unsafe exec() in prepare.py
#506
opened Apr 9, 2026 by
orbisai0security
Loading…
3 tasks done
PyTorch 2.11 + FlexAttention + SDPA (+ MFU for more GPUs)
#504
opened Apr 8, 2026 by
ademeure
Loading…
Fix #460: Enhanced artifact cleanup workflow with robust error handling
#503
opened Apr 8, 2026 by
realcarsonterry
Loading…
feat: add CLI analysis tool for experiment results
#475
opened Apr 3, 2026 by
ravyg
Loading…
5 tasks done
fix(analysis): handle crash-first baselines and empty keep sets
#469
opened Apr 2, 2026 by
afurm
Loading…
Previous Next
ProTip!
Follow long discussions with comments:>50.