issues Search Results · language:Dune language:Python language:JavaScript language:Python language:Java language:JavaScript
Filter by
55.1M results
S24 s PPGD-warmup-routes-all-even-for-the-Subset-variant quirk is faithfully replicated from a torch path that carries
its OWN warning (persistent_pgd_state.py:321: hardcoded AllLayersRouter() in warmup, ...
kind:decision
needs-decision
parity
subsystem:recon-losses
wave-1
ocg-goodfire
The JAX production stochastic recon term (ChunkwiseSubsetReconLoss) has no runnable torch Metric on this branch — only
the config schema exists (param_decomp_config/losses.py:93); it is absent from param_decomp/metrics/dispatch.py ...
kind:decision
needs-decision
parity
subsystem:recon-losses
wave-1
For ConstantSources strategies (Unmasked, CIMasked, CIMaskedSubset, CIMaskedLayerwise), torch passes
weight_deltas_and_masks=None — the delta path is absent, not zero-masked. JAX instead runs delta_mask=0, ...
kind:gap
parity
subsystem:recon-losses
wave-1
JAX supports only constant + linear-warmup for the persistent source LR (warmup_then_constant_lr);
_assert_supported_persistent refuses any post-warmup decay (verified: assert schedule.fn_type == constant ...
kind:gap
parity
subsystem:recon-losses
wave-1
What happened
In PR #29, the InputFileMap type was changed from dict[str, str] to dict[str, list[str]] to support multiple
rpms.in.yaml files referencing the same Dockerfile. The PR added a test for find_rpm_input_files_in_repo ...
enhancement
feature
good first issue
triaged
fullsend-ai-retro[bot]
StochasticHiddenActsReconLoss (per-element MSE on cached module output activations) is a real torch training-loss
capability deliberately not ported. build_recon_terms raises on any unsupported training ...
kind:deferred-feature
parity
subsystem:recon-losses
wave-1
The static live-set frozen-site realization of S2 (non-live site runs frozen x@W, no (B,T,C) acts) is only reached on
the JAX side via subset_chunk_plan (the production ChunkwiseSubsetReconLoss term). ...
kind:test
parity
subsystem:recon-losses
wave-1
Bug
In sensor.py, UNotifierVolumeSensor.native_value and extra_state_attributes call get_current_slot_info() without passing
now_weekday / weekend_days:
# sensor.py, native_value (~line 33-34)
now_time ...
jokerigno
build_recon_terms consumes the shared AnyLossMetricConfig tuple, requires each training loss to carry a coeff, and keys
terms by instance_key (cfg.name or cfg.type) with a distinct-name assert (torch instantiate_metrics ...
kind:done
parity
subsystem:recon-losses
Description: When users proceed to the checkout section in the Enatega Multivendor Customer App and attempt to switch
payment methods, two distinct issues are observed:
Credit/Debit Card Section Displaying ...
waleedahmedhere
Learn how you can use GitHub Issues to plan and track your work.
Save views for sprints, backlogs, teams, or releases. Rank, sort, and filter issues to suit the occasion. The possibilities are endless.Learn more about GitHub IssuesProTip! Restrict your search to the title by using the in:title qualifier.
Learn how you can use GitHub Issues to plan and track your work.
Save views for sprints, backlogs, teams, or releases. Rank, sort, and filter issues to suit the occasion. The possibilities are endless.Learn more about GitHub IssuesProTip! Restrict your search to the title by using the in:title qualifier.