-
Notifications
You must be signed in to change notification settings - Fork 51
Pull requests: goodfire-ai/param-decomp
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
feat(lab): pd-cold R2 archive — auto-push finished runs' decomposition item + cross-cluster pull
#964
opened Jul 8, 2026 by
claude-spd1
Collaborator
Loading…
feat(tools): gc_dead_runs — GC early-crashed runs' checkpoints
#961
opened Jul 7, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(train): stale-CI replay (S34) — repeat steps reuse the window-first CI envelope
#953
opened Jul 6, 2026 by
danbraunai-goodfire
Collaborator
Loading…
fix(consumers): metadata consumers parse only the config subset they read
#952
opened Jul 6, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(checkpoint): split checkpoints into decomposition | training items (S22/S33)
#950
opened Jul 6, 2026 by
ocg-goodfire
Collaborator
Loading…
fix(harvest): per-worker local_batch_size; GPU assert + mem-fraction for 8B restores
#949
opened Jul 5, 2026 by
ocg-goodfire
Collaborator
Loading…
perf(train): per-entry recon fwd/bwd sequencing (+ save_gathered_weights, both default-off)
#946
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
configs: full32L bf16-source batch ladder — b192 + b224 probes
#945
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(optim): bf16 Adam m/v storage for the V/U + CI-fn optimizers (moments_dtype)
#944
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(adversary): stateless sign-ascent SRC_STEP for persistent-PGD sources (SPEC §6)
#943
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
fix(slow-eval): remove slow_on_first_step — the slow tier fires on slow_every multiples only
#942
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
perf(compile): 13x faster V/U init compile + compile-time findings
#941
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
fix(invariance_check): feed int tokens, not a float residual batch (SPEC D4)
#940
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
bf16 PPGD source: quality A/B + full32L memory cash-in configs (lp-bf16-source-ab)
#938
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(adversary): optional global-norm grad clip on PPGD source ascent + src-grad-norm metric
#937
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
tests: un-xfail torch↔JAX goldens under the embed-internal token contract
#936
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
deps: unpin pydantic, bump wandb 0.21.0 -> 0.22.3
#935
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(eval): per-site faithfulness observability (rel/abs Frobenius)
#934
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
test(recon): confirm PGDReconLayerwiseLoss fresh-PGD per-site independence (#918 item 6)
#933
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
fix(checkpoint): make orbax-restored TrainState donatable — solve the resume OOM (S22)
#932
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(optim): config-gated Muon for the main optimizers (amends SPEC S20)
#931
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
•
Draft
docs(pretrain): fix stale clean_suffix_logits reference missed by #923
#930
opened Jul 3, 2026 by
ocg-goodfire
Collaborator
Loading…
feat(ci-health): CI-fn training-health telemetry (eval/ci_health/*)
#929
opened Jul 2, 2026 by
ocg-goodfire
Collaborator
Loading…
fix(toys): pd-tms/pd-resid-mlp eval crash — missing remat kwarg
#928
opened Jul 2, 2026 by
ocg-goodfire
Collaborator
Loading…
perf(sharding): fsdp-major ZeRO-1 linearization — −14% step time at dp32
#927
opened Jul 2, 2026 by
ocg-goodfire
Collaborator
Loading…
Previous Next
ProTip!
What’s not been updated in a month: updated:<2026-06-08.