Merge pull request #22 from evolutionaryscale/jaime/chore/version-bum…

…p-0.1.14

Bump version to 0.1.14

Merge pull request #20 from evolutionaryscale/jaime/fix/nvl-metadata-…

…aliasing

Fix internode metadata NVL buffer aliasing that corrupts total_tiles

Merge pull request #17 from evolutionaryscale/jaime/chore/quack-upgrade

Port StreamEP to quack main (CuTeDSL / CUTLASS 4.6.1)

Merge pull request #16 from evolutionaryscale/jaime/chore/version-bum…

…p-0.1.11

Bump version to 0.1.11

Merge pull request #14 from evolutionaryscale/jaime/chore/streamep-up…

…grade2

Bump version to 0.1.10

Merge pull request #5 from evolutionaryscale/jaime/chore/version-bump

chore: bump version 0.1.8 -> 0.1.9

comments: condense dense comments + drop dangling doc refs

Comment/docstring/help-only cleanup of the branch (no code changes):

- Drop dangling references: the markdowns/ docs and the section markers
  (§8.x, "hang doc", design.md §...) cited throughout don't exist in this
  repo, so the citations point nowhere for a reader. Kept the functional
  prose; removed the markers.
- Rewrite stale narrative: the data-ring header comments described the
  removed RDMA meta channel ("...is gone"); now describe the current
  data->head->tail layout directly. Fixed the dispatch_grads_stress
  --slack / --leave_free_gb help that still referenced the deleted
  rdma_channel_meta / kRdmaMetaRingDepth mechanism.
- Condense over-verbose blocks (SourceMeta ctor, arrival gate, baseline
  live-read race, tail-writeback race, and the stress-test argparse help)
  while preserving the load-bearing rationale (memory ordering, the
  never-0 gen-tag invariant, cross-gen writeback races, must-stay-after
  -wait_event constraints, false-accept-horizon bounds).

7 files, +71/-125. .so rebuilds clean; py_compile OK on all touched .py.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Merge pull request #3 from chanzuckerberg/jaime/fix/num-sms-autopick-…

…channel-mismatch

Resolve num_sms auto-pick before comm-buffer sizing (fix EP=16 IMA)