2026-07-12 - 2026-08-12
Overview
1 Release published by 1 user
Published
v0.2.0
14 Pull requests merged by 1 user
Merged
#26 Add opt-in straight-through Gumbel-softmax combine weights to MoE router
Merged
#25 Add learnable per-expert width and shared temperature to EnergyRouter
Merged
#24 Store a quantile grid instead of a raw reservoir sample in the setup cache
Merged
#23 Offset event_id per file to avoid cross-file collisions
Merged
#21 Cache giant train's setup stage in a sidecar file
Merged
#20 Speed up giant train's setup stage
Merged
#19 Make default checkpoint out_dir name reflect only non-default hyperparams
Merged
#18 Feature/wandb integration
Merged
#16 Refactor the analysis plot creation with focus on rollout
Merged
#14 Draft: Resolve "Wasserstein-GAN throwaway experiment (fast single-pass eval)"
Merged
#13 Resolve "Condition on material + particle physical properties, not learned embeddings"
Merged
#12 Rewrite analysis module as a lean, fully-streaming pipeline
Merged
#11 Add mixture-of-experts routing prototype for Stage 1 and Stage 2
Merged
#9 Implement Phase 2: secondary particle prediction
2 Pull requests proposed by 1 user
Proposed
#17 WIP: Condor gpu train rollout
Proposed
#27 V0.3.0 stage2 autoregressive
5 Issues closed from 1 user
Closed
#15 Refactor Analysis Pipeline
Closed
#3 Wasserstein-GAN throwaway experiment (fast single-pass eval)
Closed
#7 Condition on material + particle physical properties, not learned embeddings
Closed
#2 Phase 2 secondary-prediction baseline: train + validate
Closed
#4 Prototype a mixture-of-experts / routing-tree architecture
2 Issues created by 1 user
Opened
#15 Refactor Analysis Pipeline
Opened
#22 event_id collisions across multi-file/multi-shard reference reads in giant analyze