README: state the real generality + scale evidence; retract "near-flat"#267
Conversation
Todd Baur (toadkicker)
requested review from
Nelson Spence (Fieldnote-Echo) and
Navi Bot (project-navi-bot)
as code owners
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 098b6ac863
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 (@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 (@codex) address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c9f784b4d5
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "Codex (@codex) review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "Codex (@codex) address that feedback".
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Nelson Spence (Fieldnote-Echo)
dismissed stale reviews from Navi Bot (project-navi-bot) and themself
via
96c6cca
Nelson Spence (Fieldnote-Echo)
force-pushed
the
readme/scale-and-generality-evidence
branch
from
c0fbd4c to
96c6cca
Compare
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Nelson Spence (Fieldnote-Echo)
requested review from
Nelson Spence (Fieldnote-Echo) and
Navi Bot (project-navi-bot)
Tighten README benchmark framing to the evidence available in this public branch. The README now describes the public Harrier-Q8 BEIR harness and the corpus sizes covered by those artifacts, while treating larger-corpus and alternate-encoder results as active research until public artifacts land. It also removes near-flat per-query cost language and clarifies the O(n) compressed-code scan story, small/medium corpus latency claims, and footprint wording. The reported two-stage path is sign/bitmap candidate generation followed by RankQuant rerank; it does not retain original floats for library-side reranking. Signed-off-by: Nelson Spence <nelson@projectnavi.ai>
Nelson Spence (Fieldnote-Echo)
force-pushed
the
readme/scale-and-generality-evidence
branch
from
96c6cca to
92efc66
Compare
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Nelson Spence (Fieldnote-Echo)
deleted the
readme/scale-and-generality-evidence
branch
3 participants
What
Tightens the README benchmark framing so public claims match the artifacts in this branch, and updates benchmark comments that still described the old flat-scaling story.
Why
The previous README mixed public release evidence with broader research notes, and it overstated the scale story. The public branch currently supports a narrower claim: Harrier-Q8 embeddings on public BEIR data, with the default
make benchmark-beircovering scifact and trec-covid and optional harness support for nfcorpus/fiqa.Changes
sign→rq2row accounts for both stage-1 sign codes and the RankQuant reranker at 384 B/vector. The public two-stage path is sign/bitmap candidate generation followed by RankQuant b=2 rerank; it does not retain or rescore against the original float corpus.Validation
cargo fmt --checkgit diff --checkpython -m py_compile benchmarks/beir/beir_plot.py