Repository navigation
Conversation
Pack suffix projections and readout while preserving every recurrent prefix adjoint. Retain the serial reference and independent parent comparisons, including short-prefix regressions. Signed-off-by: 0z5a <dezhen.lu@student.uni-tuebingen.de>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Serial recurrent-prefix replay repeats projection and readout work for every response. This adds an opt-in packed suffix path that batches those operations while preserving each suffix's causal history, every loop's prefix adjoints and first-response readout boundary. The serial reference remains the default.
Stacked on #11 (
73d5ff3); this PR contains only the packed-suffix iteration and its evidence.Validation: 448 CPU checks pass, six CUDA checks skip and three known missing-Triton factory checks are deselected. Across 648 same-tree and 216 independent-parent observations, all loss, complete-gradient and nonzero-update gates pass. Eighteen fresh comparison processes exit naturally.
Tiny FP32 CPU fixed-trace actor forward/backward/SGD, P256/D8/G32, compared with untouched packed-wave parent #10:
The report also compares the untouched serial-prefix parent #11 and retains every short-prefix/control regression, the initial numerical failures and the positional-capacity fixture failure. One-response controls regress; all clean P24/D4/G32 intervals against packed-wave include one. Huginn full-prefix reuse requires the same actual latent identity.
Results and reproduction include full oracle vectors, source/import checks, raw observations and SHA manifests. This is CPU component evidence; target CUDA, peak memory, production distributed RL and reward convergence remain unqualified.