wassname
3ae1fd48c9
Record CorDA UAT result
2026-07-02 19:09:15 +08:00
wassname
0a12bc23aa
Track CorDA UAT retry
2026-07-02 06:07:30 +08:00
wassname
d2b07121e3
Clarify README steering showcase wording
2026-07-01 18:27:08 +08:00
wassname
152ffa87e1
Track CorDA steering branch UAT
2026-07-01 18:17:43 +08:00
wassname
f7d8e697d9
Rewrite README for pure Authority showcase
2026-07-01 07:50:47 +08:00
wassname
b8f812c0ac
Regenerate pure Authority showcase plots
2026-07-01 07:30:49 +08:00
wassname
0b3e28c9eb
Render authority method comparison plots
2026-07-01 06:25:49 +08:00
wassname
061ce8a4f6
Export MFQ2 sample logprobs for N bootstrap
2026-07-01 06:24:13 +08:00
wassname
d0278be144
Render PCA authority candidate maps
2026-07-01 06:10:58 +08:00
wassname
981d1c4ed9
Correct authority UAT to MFV path evidence
2026-07-01 06:04:16 +08:00
wassname
aec74319f6
Record mundane authority verifier results
2026-07-01 05:59:15 +08:00
wassname
0fdc101f5c
Record authority UAT queue blocker
2026-07-01 00:58:02 +08:00
wassname
0dce35c05f
Record mundane authority selection UAT
2026-07-01 00:31:01 +08:00
wassname
a7b49dd723
Record linear act authority verifier failure
2026-06-30 23:40:57 +08:00
wassname
a8933a98b4
Record directional ablation authority verifier failure
2026-06-30 23:10:53 +08:00
wassname
9eb32d0569
Record sspace authority verifier failure
2026-06-30 22:38:26 +08:00
wassname
9dd7a9a24a
Record pca authority verifier failure
2026-06-30 21:35:22 +08:00
wassname
23829d21bb
Record mean_diff authority verifier failure
2026-06-30 21:03:58 +08:00
wassname
df320d670b
Track pure authority method queue
2026-06-30 20:32:46 +08:00
wassname
2d8a736935
Record pure authority UAT gates
2026-06-30 19:52:03 +08:00
wassname
8241551ec1
Replan pure authority steering workflow
2026-06-30 19:02:34 +08:00
wassname
e9572ae00a
Fix authority pipeline to hold persona pair fixed
2026-06-30 18:59:39 +08:00
wassname
713b3003fd
Track authority method comparison queue
2026-06-30 18:43:50 +08:00
wassname
754da269ba
Record authority steer verification verdict
2026-06-30 18:15:48 +08:00
wassname
bbe59be7e3
Record MFQ2 path failure
2026-06-30 17:36:06 +08:00
wassname
81fda62ecf
Track unbuffered authority UAT run
2026-06-30 17:22:24 +08:00
wassname
6788e84738
Track authority fixed-C UAT run
2026-06-30 17:11:58 +08:00
wassname
4862e6f24f
Record authority score60 export
2026-06-30 16:29:34 +08:00
wassname
d32272790d
Track authority stage-b validation
2026-06-30 15:59:38 +08:00
wassname
0c43f1bc62
Tick authority validation subproofs
2026-06-30 15:33:32 +08:00
wassname
b1c09966a2
Record DeepInfra Qwen smoke
2026-06-30 15:09:27 +08:00
wassname
8dcaf30705
Record authority validator routing
2026-06-30 15:02:17 +08:00
wassname
b74c5ec9d3
Track authority validation workflow
2026-06-30 14:51:39 +08:00
wassname
ec0bc32aed
Track authority-only steer pipeline
2026-06-30 14:14:07 +08:00
wassname
9fc3852a8a
Require explicit showcase steering anchor
2026-06-30 14:08:16 +08:00
wassname
fe3b0aff2c
Gate showcase paths on contrast and margin
2026-06-30 14:01:59 +08:00
wassname
6fcdfea30f
Support sampled survey reads and MFV c-grid plots
2026-06-30 13:28:20 +08:00
wassname
8ab02adf63
docs: explain logprob readouts in README
2026-06-25 20:37:20 +08:00
wassname
5eb147871c
docs: record repo cleanup UAT
2026-06-25 20:22:13 +08:00
wassname
9abddaeac5
return pmass
2026-05-08 16:36:36 +08:00
wassname
0f8048d5d9
Implement N-token evaluation with guided rollouts
...
- Refactored evaluation logic in `src/tinymfv/eval.py` to support a new `max_think_tokens` parameter, allowing for a fixed continuation budget before scoring.
- Introduced `guided_rollout` function in `src/tinymfv/guided.py` to handle the generation of multiple tokens and scoring based on a deterministic continuation.
- Updated the CLI in `scripts/03_eval.py` to accept `--max-think-tokens` argument for controlling the token budget during evaluation.
- Created a new specification document `docs/spec/20260501_n_token_eval.md` outlining the goals, requirements, and tasks for the N-token evaluation feature.
- Simplified the record creation in `scripts/02_rewrite.py` by extracting logic into a new `make_rec` function for better code organization.
2026-05-01 21:44:14 +08:00
wassname