Conversation
- bind policy, model, router, checkpoint, token, and kernel semantics - encode compact IDs in a checksummed fail-closed safetensors artifact - add pure route-drift diagnostics and a codec-only benchmark - keep capture, transport, replay, admission, and training unchanged Benchmark: NVIDIA L20, PyTorch 2.11.0+cu128, codec-only synthetic 8192x40x22 trace, 512 experts, median of 50 after 5 warmups - ID storage: 57671680 -> 14417920 bytes (0.250x, -75.0%)
- freeze a per-response trace grid behind exact token and behavior identities - reject policy, ordering, padding, boundary, and cache mutations - preserve the existing rollout path when trace artifacts are absent - add deterministic binding and validation benchmarks Benchmark: NVIDIA L20, PyTorch 2.11.0+cu128, synthetic B=2 G=2 P=512 R=1024 L=40 K=22 E=512, median of 5 binds and 50 validations after 5 warmups - CPU IDs: 7.215 MB artifact, 83.732 ms/bind, 0.871 ms/validation - GPU-resident IDs+weights+margin: 14.753 MB artifact, 103.433 ms/bind, 1.206 ms/validation
ViperEkura
force-pushed
the
main
branch
2 times, most recently
from
September 8, 2026 23:09
e7237f6 to
093c57e
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Depends on #64 (RouteTraceV0).
Exact-head confirmation (2026-09-05)
python -m pytest -p no:cacheprovider -q tests/atbea9ecf71f623d7b78aa2baed59646be871af19c: 749 passed, 57 skipped in 54.96s, container exit 0.This final run retained its log independently of SSH connectivity. It used
the CPU-only environment described below; GPU tests remain skipped.
Bind an immutable per-response RouteTraceV0 grid to the exact rollout policy,
tokens, masks, behavior logprobs, token boundaries and response order. Validate
bindings at scoring/cache boundaries and fail closed on mutation or mismatch.
RawRollout.route_trace_batch=Nonepreserves existing behavior.No model route capture, replay, dispatch, reward/objective or admission change
is included. This is attribution validation, not a transport implementation.
Revision and dependency
e1e533c88e74efab4f379d6ccbc830c2b3e698e2.bea9ecf71f623d7b78aa2baed59646be871af19c.89beef87e9f56ce921c3dd3bf0fee66041599071(
codex/draft-moe-route-trace-v0). Review the incremental commit after it;this main-based Draft includes that dependency in its diff.
per-response route attribution. No duplicate PR was found in the open PR list.
Validation
python -m pytest -p no:cacheprovider -q tests/moe tests/trainer/test_rollout_route_trace.py tests/trainer/test_rollout_route_trace_benchmark.py tests/trainer/test_rollout.py: 88 passed on replay headbcfcaaade7d46abddba20c5b8b9ab33d326f62c0.import/format and
git diff --checkpassed.attempt failed collection because the image lacked httpx2; after installing
it another SSH connection expired before returning a final result, so that
attempt is not reported as passing.
CPU image
astrai-test:cu128, Python 3.12.3, torch 2.11.0+cu128, pytest 8.3.5.No GPU was exposed. httpx2 2.12.0 is a disposable-container test dependency.
Historical benchmark (pre-rebase)
Synthetic B=2 G=2 P=512 R=1024 L=40 K=22 E=512 on L20 host:
CPU IDs 7.215 MB, 83.732 ms/bind and 0.871 ms/validation median;
GPU rich routes 14.753 MB, 103.433 and 1.206 ms. This starts from materialized
routes and excludes generation, transport, replay, scoring and training.
Review status
AI assistance was used. Keep Draft for human line-by-line review, dependency
review and CI. No current-head GPU or training-throughput claim is made.