Commit Graph

Select branches

Hide Pull Requests

main

Mono Color

• f2cbce5a03 related_work: log 12b_adaptive outcome as confirmation of the anchoring account NilsandClaude Fable 5 2026-07-14 12:57:08 +02:00
• e152a91c05 parse_known_args: module-level parser must tolerate importers' argv NilsandClaude Fable 5 2026-07-14 12:50:45 +02:00
• cd8e94dda5 gpuq: idle-time class dispatcher (single-writer, race-free) + docs NilsandClaude Fable 5 2026-07-14 12:47:23 +02:00
• 9329f22a63 gpuq: class-based scheduling at submit time (pending count + live util) NilsandClaude Fable 5 2026-07-14 12:41:34 +02:00
• c718c636a5 GPUQ: as-deployed cheat sheet + job conventions NilsandClaude Fable 5 2026-07-14 12:40:39 +02:00
• d2d372c909 feedforward flag for HumanEval eval NilsandClaude Fable 5 2026-07-14 12:34:57 +02:00
• 12ec01becc adaptive flag for GSM eval NilsandClaude Fable 5 2026-07-14 12:33:03 +02:00
• 95be35d20f adaptive flag for unified trainer NilsandClaude Fable 5 2026-07-14 12:32:36 +02:00
• 32728e7b07 bootstrap via public https clone (repo now public) NilsandClaude Fable 5 2026-07-14 12:25:07 +02:00
• 471cf336d0 gpuq: document the scheduler (design, contract, ops, failure modes, alternatives) NilsandClaude Fable 5 2026-07-14 11:54:38 +02:00
• 0252641701 gpuq: per-node supervisor (worker resurrection + bucket heartbeat) NilsandClaude Fable 5 2026-07-14 11:50:10 +02:00
• 5efe5b790a gpuq: bucket-backed per-GPU job queue (worker loop + submit/status helper) NilsandClaude Fable 5 2026-07-14 11:48:03 +02:00
• 1460cf241f GSM plan-distillation control (width/depth law falsification test) NilsandClaude Fable 5 2026-07-14 11:22:46 +02:00
• 8fe8e34b47 rung-2 pilot: loop-only entrance-faded band LoRA, warm-start, inline eval; BW hard-only fallback NilsandClaude Fable 5 2026-07-14 11:00:42 +02:00
• 507c170744 train_merge_bw: fall back to hard-only curriculum when easy pool is empty (BW direct pass ~1%) NilsandClaude Fable 5 2026-07-14 10:59:32 +02:00
• 1ad019d76d add --lr flag (wide-adapter LR rescale diagnostics at 12B) NilsandClaude Fable 5 2026-07-14 10:46:46 +02:00
• 53a8c9b609 adaptive-alpha merge (Lys-inspired, trained) + truncated-BPTT deep-k training NilsandClaude Fable 5 2026-07-14 10:29:57 +02:00
• cd0ed80ea6 related_work: track the two arXiv PDFs (retrofitted recurrence, inner loop inference) NilsandClaude Fable 5 2026-07-14 10:28:34 +02:00
• c08c8d503d related_work: archive the 2026 workspace paper (HTML + 18 figures + bib) for offline reference NilsandClaude Fable 5 2026-07-14 10:26:54 +02:00
• ba3bacc252 unification paragraph: anchor coefficient as the stability dial across Lys/ours/McLeish NilsandClaude Fable 5 2026-07-14 10:25:04 +02:00
• 8e7bd2019e integrate McLeish et al. + Lys et al.: reposition related work, rung-2 Muon note NilsandClaude Fable 5 2026-07-14 10:24:17 +02:00
• 7ac365271a related_work: unify anchoring across McLeish/Lys/ours; link 12B alpha miscalibration to their stability results NilsandClaude Fable 5 2026-07-14 10:24:12 +02:00
• 89df135631 related_work: relevance notes for 2511.07384 (retrofitted recurrence) and 2602.14759 (inner loop inference) NilsandClaude Fable 5 2026-07-14 10:22:31 +02:00
• f48011c470 decouple --pause from feedforward: enable loop+pause combo arm NilsandClaude Fable 5 2026-07-14 10:09:18 +02:00
• 5c4b76ba14 bootstrap via bucket git-bundle (no ssh keys on rented nodes) NilsandClaude Fable 5 2026-07-14 10:06:41 +02:00
• 01ac15d3e1 node kit: alpha flag through train/eval, carry env+width fixes, bootstrap script NilsandClaude Fable 5 2026-07-14 10:05:30 +02:00
• 674b814b65 pre-register L9 discriminator arm NilsandClaude Fable 5 2026-07-14 09:57:04 +02:00
• 263a07eab9 three-paper split: plans, claim/evidence manifests, work plans NilsandClaude Fable 5 2026-07-14 09:52:25 +02:00
• 54f2b3c6cc MultiPL-E Rust harness: compile-run verifier, STaR labeling, transfer eval NilsandClaude Fable 5 2026-07-14 09:45:46 +02:00
• 6c39c38eb9 track protocol pre-registrations and interim report in git NilsandClaude Fable 5 2026-07-14 02:54:52 +02:00
• 42304aacf8 fix: load task data files only for selected tasks NilsandClaude Fable 5 2026-07-14 01:28:12 +02:00
• 69ce16ab53 vast.ai runbook + per-tokenizer eos handling in loop_common NilsandClaude Fable 5 2026-07-14 01:14:57 +02:00
• ef9c08966c J-lens workspace reproduction + loop retrofit: lens, band looping, adapters, controls, multi-task evals NilsandClaude Fable 5 2026-07-14 00:54:12 +02:00