Repository navigation
feat(training): dual-peak timestep sampling, cyclic LR schedules, que… - #9
Merged
Merged
Conversation
…ue ETA Timestep sampling - New `dual_peak` mode backed by the standalone runtime/dual_peak_sampling package: a mixture of two log-SNR components with independent position, width and weight. Wired into the baseline and InfoNoise samplers and exposed in the schema as fields gated on timestep_sampling==dual_peak. LR schedulers - `cosine_cycles`: training is split into N equal cosine cycles, with a shared max/min LR or per-cycle lists. - `constant_then_cosine`: hold the LR, then cosine-decay over the final fraction of training (decay_start_ratio, decay_min_lr). - Warmup/restart fields are now shown only for the schedulers that use them. Web - FloatListField for per-cycle LR lists; ru/en/zh strings. - Projects and Queue show a training-finish forecast derived from queue order and live monitor progress (lib/queueEstimates). Projects can be sorted by queue position, finish time, updated, created or title. Tests cover the sampler, both schedulers, the schema bridge and the forecast. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
…ue ETA
Timestep sampling
dual_peakmode backed by the standalone runtime/dual_peak_sampling package: a mixture of two log-SNR components with independent position, width and weight. Wired into the baseline and InfoNoise samplers and exposed in the schema as fields gated on timestep_sampling==dual_peak.LR schedulers
cosine_cycles: training is split into N equal cosine cycles, with a shared max/min LR or per-cycle lists.constant_then_cosine: hold the LR, then cosine-decay over the final fraction of training (decay_start_ratio, decay_min_lr).Web
Tests cover the sampler, both schedulers, the schema bridge and the forecast.