Repository navigation
feat(240m): Session 8-9 eval, certification and blend tooling pre-eval - #17
Merged
Merged
Conversation
uncoalesced
commented
Nov 7, 2026
Owner
- S9 certification runner (adds --out-dir and --no-vault-handoff so re-runs on other checkpoints do not overwrite the golden S9 reports)
- Hermes specialized 1000-case suite builder, uncapped runner and scorer; runner watchdog now times decoding only (token-by-token prefill used up to 7 of 10 s on long-context prompts) and logs prefill/decode seconds
- S8 DPO prompt build/verify/top-up scripts, S6/S8 blend sweep, converse.py, identity/honesty/benign-refusal/16k long-context evals, 900-point grader
- NPU exporter: 32-step multi-provider parity check
- Model card draft; licence policy copy
- .gitignore: keep licensed/unverified data and held-out eval keys out of the public repo
- S9 certification runner (adds --out-dir and --no-vault-handoff so re-runs on other checkpoints do not overwrite the golden S9 reports) - Hermes specialized 1000-case suite builder, uncapped runner and scorer; runner watchdog now times decoding only (token-by-token prefill used up to 7 of 10 s on long-context prompts) and logs prefill/decode seconds - S8 DPO prompt build/verify/top-up scripts, S6/S8 blend sweep, converse.py, identity/honesty/benign-refusal/16k long-context evals, 900-point grader - NPU exporter: 32-step multi-provider parity check - Model card draft; licence policy copy - .gitignore: keep licensed/unverified data and held-out eval keys out of the public repo
Hermes-1000 strict blind grading (all checkpoints ~2%), strict 60-probe re-grade, controls, S9 gates on the blend, and problems found in the evaluation instruments.
Per-session BPB, answering and behaviour tables, root causes (SFT tool-call data, ineffective S8 DPO, test-set selection), staged 85-90% gate targets and the S10 training/eval design.
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.