Repository navigation
Conversation
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 10962336ca
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 172ab58fdb
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Summary
This stacked PR proves the ModelMap ontology end to end with automated validation and a deliberately small, primary-source-grounded seed corpus.
Validation
tools/validate.pysourcerecordsclaim_refsmainSeed corpus
The initial graph establishes three proof paths:
The technical edges are backed by atomic
confirmed_primaryclaims sourced to the MQA paper, GQA paper, DeepSeekMoE paper, DeepSeek-V2 report, and DeepSeek-V3 report.Ontology refinement
Adds
conceptual_predecessor_ofto the relationship schema so conceptual technique lineage remains distinct from checkpoint/parameter lineage.Scope discipline
This PR intentionally does not assert a direct GQA → MLA lineage edge. They belong to the broader history of KV-efficiency techniques, but chronology or shared motivation alone is not enough to establish ancestry under ModelMap's evidence policy.
Stack
Base:
atlas-foundation-v0.1(PR #1)Head:
atlas-seed-v0.1