docs: DeepSeek-V4.1-Flash learnings + TODO.impl work ledger - #95
Merged
Merged
Conversation
The report's applicability analysis, now with our own measurements: speculative-decode acceptance on the shipped tier pair, the r6/r7 per-domain teacher slice (routing hypothesis negative - r7 wins every domain), and the corrected lite-rung reading (run-009 is trained; the untested variable is init source). TODO.impl/01-10: the remaining implementation work, one file per workstream with priorities and acceptance criteria. Corrections: TODO.arabic/10 + TODO.hebrew/11 describe example retrieval, but Engram is a hash-addressed n-gram lookup table (rebuilt note); TODO.arabic/19 carries the single-pass mHC shift; TODO.arabic/18 updated to the implemented headwise Muon state.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The DeepSeek-V4.1-Flash technical report, read end-to-end and mapped to our stack. Most of it (KV-cache compression at 552B/1M-context scale) does not apply; five levers and two recipe validations do, and two of them are already measured on our artifacts.
Learnings memo (
TODO.qwen-next/10), including our measurements:TODO.impl/: the resulting work ledger, one file per workstream (01-10), priorities P0-P3, acceptance criteria per file.