jsahn

jsahn pushed to main at jsahn/QueryExpantion

2026-10-23 02:51:13 +00:00

b07d5041a0 Official SweRank driver: retrieval source/output options; chain Embed-Large -> LLM-Large

jsahn pushed to main at jsahn/QueryExpantion

2026-10-23 02:43:40 +00:00

4d40b96394 SweRankEmbed-Large retrieval on the DGX: corpus export (same docs as Small), deduplicated last-token embedding, top-100 per bug

jsahn pushed to main at jsahn/QueryExpantion

2026-10-22 02:01:44 +00:00

e8d8dc693a Pace MiniMax by weekly usage: pacer polls the Token Plan usage API; primary may use its whole week evenly, backup at most 30%; all three clients wait while no key is within budget

jsahn pushed to main at jsahn/QueryExpantion

2026-10-22 01:48:22 +00:00

6db2141d91 Backup key weekly reset ~10-26 08:50

jsahn pushed to main at jsahn/QueryExpantion

2026-10-22 01:47:35 +00:00

8ced74b644 Record the backup key's weekly limit (exhausted 10-22 08:50)

jsahn pushed to main at jsahn/QueryExpantion

2026-10-21 01:53:15 +00:00

2954c6577e compare_pro_systems: --prafl-tag (pro3 = PRAFL→Listwise)

jsahn pushed to main at jsahn/QueryExpantion

2026-10-21 01:52:45 +00:00

176277dc70 RGFL client: recompute key order every pass (stuck 30h on the exhausted primary); standalone M3.1 RGFL driver

jsahn pushed to main at jsahn/sebrief

2026-10-20 05:06:28 +00:00

cdd070749b Backfill guard: a past week's digest only takes papers published by its end

jsahn pushed to main at jsahn/QueryExpantion

2026-10-20 03:54:21 +00:00

c1cd622238 Driver for official SweRank (SweRankEmbed-Small + SweRankLLM on the DGX) on Pro

jsahn pushed to main at jsahn/QueryExpantion

2026-10-20 03:48:32 +00:00

8fdded8787 SweRankLLM client: default endpoint is the DGX Spark

jsahn pushed to main at jsahn/QueryExpantion

2026-10-20 02:18:43 +00:00

c6047e9fdc SweRankLLM serve script for DGX Spark (NVIDIA vLLM container)

de55310e38 SweRankLLM (official reranker) via vLLM on the MacBook: serve scripts, saved-retrieval replay, swerankllm reranker

ae4c57e628 M3.1: Pro only for now (M31_JOBS override, guard stops Multi-SWE jobs)

280ed3f285 Pace M3.1 calls: per-process concurrency cap, longer timeouts

45ea78a1f3 Restore Pro worktrees; preflight-check repositories before the M3.1 run

Compare 31 commits »

jsahn pushed to main at jsahn/QueryExpantion

2026-10-14 13:05:07 +00:00

653b8a50a6 Go-75 ablation: final numbers on all 75 bugs

jsahn pushed to main at jsahn/QueryExpantion

2026-10-14 11:38:40 +00:00

7ca0601179 Table 8 ablation: diversified retrieval and verification each contribute; neither alone beats the SweRank re-implementation

55a7872d0c Add ablations (c) dense-only retrieval and (a') dense initial order for the listwise reranker

Compare 2 commits »

jsahn pushed to main at jsahn/QueryExpantion

2026-10-14 05:11:17 +00:00

239adac286 Add the Go-75 verification / union-dense ablation to Table 8 notes and the report

jsahn pushed to main at jsahn/QueryExpantion

2026-10-14 03:07:25 +00:00

aa14c54812 Label the re-implemented baselines as SweRank (re-impl.) in Table 8 and the report

ccd01e0013 Add the two Table 8 ablation switches: skip verification, union with dense top-100

Compare 2 commits »

jsahn pushed to main at jsahn/QueryExpantion

2026-10-14 02:24:57 +00:00

92037784ec Rename the dense retrieve-and-rerank baselines; they are not the published SweRank system

jsahn pushed to main at jsahn/QueryExpantion

2026-10-14 02:18:00 +00:00

701de09b2e Add the one-page SWE-bench Pro results report

jsahn pushed to main at jsahn/QueryExpantion

2026-10-13 14:44:49 +00:00

3f1868ce4d Add the SweRank runs (both retrievers) to Table 8 with their outputs

jsahn pushed to main at jsahn/Radxa_MIXER

2026-10-13 09:58:56 +00:00

767b1c8963 Mix inputs at 0 dB instead of -3 dB

jsahn pushed to main at jsahn/Radxa_MIXER

2026-10-12 13:01:57 +00:00

1cd00ae21b Show Bluetooth inputs when BlueZ objects can't be read as JSON