This website requires JavaScript.
733c13c31c
Mark issues.md Issue 5 as fixed
lars
2026-08-12 15:31:17 +02:00
818c380fd0
Extract predict/rollout's duplicated inference bootstrap into giant.checkpoint_io (issues.md Issue 5)
lars
2026-08-12 15:31:14 +02:00
6a21c3b908
Mark issues.md Issues 3 & 4 as fixed
lars
2026-08-12 15:05:33 +02:00
2bfb1ab056
Extract giant train/new-run's CLI override mapping into a table-driven function (issues.md Issues 3 & 4)
lars
2026-08-12 15:04:49 +02:00
01acbfed61
Add unknown-key validation to config.toml merge (issues.md Issue 2)
lars
2026-08-12 14:47:29 +02:00
9bf5874308
Make config dataclasses the single source of truth for DEFAULT_CONFIG
lars
2026-08-12 14:35:14 +02:00
55332db67a
Bump ruff line-length to 120 and reformat
lars
2026-08-12 13:33:09 +02:00
9ce7b32324
Fix test_render_all_run_gallery_invokes_subprocess clobbering LaTeX's own subprocess.run
lars
2026-08-10 11:56:36 +02:00
82772e4e09
Add render.py coverage: figure params, router diagnostics plots, gallery/condor glue
lars
2026-08-10 11:47:31 +02:00
b1cf9d345d
Downgrade coverage-report upload to actions/upload-artifact@v3
lars
2026-08-10 11:38:56 +02:00
24445b7427
Add coverage for router-center seeding, geometry batch reader, material topN cache, and setup-cache corruption paths
lars
2026-08-10 11:38:04 +02:00
451bdc210e
Apply ruff format
lars
2026-08-10 11:26:28 +02:00
adb7a8663e
Add pytest-cov to dev deps and run coverage in CI
lars
2026-08-10 11:25:50 +02:00
878e9ddca3
Delete docs/v0.3.0-design.md and strip all references to it
lars
2026-08-10 11:19:02 +02:00
f46628141d
Bump version to 0.3.0
lars
2026-08-10 10:50:38 +02:00
630d8d3992
Rewrite README for v0.3.0 architecture, quick start, and data columns
lars
2026-08-10 10:49:50 +02:00
fff61ebd61
Deduplicate giant/training/trainers.py shared per-stage logic
lars
2026-08-10 10:32:26 +02:00
d3271bc798
Silence the fork-safety warning from num_workers>0 pipeline tests
lars
2026-08-10 09:48:58 +02:00
8019a80563
Refactor train.py into giant/training/ around a metrics collector
lars
2026-08-07 17:03:20 +02:00
da7cde3ef9
v0.3.0 post-implementation audit: resolve all 9 tracked discrepancies
lars
2026-08-07 16:12:58 +02:00
200c6d243b
v0.3.0 step 7: AttentionHistory (KV-cached) + scheduled/never teacher forcing
lars
2026-08-07 13:15:56 +02:00
93b19911f8
v0.3.0 step 6: sample.py/rollout.py AR generation + class->PDG decode
lars
2026-08-07 10:37:57 +02:00
c9d255b1c5
v0.3.0 step 5: Stage2Autoregressive (history=markov) + §11.4 grad instrumentation
lars
2026-08-07 09:36:49 +02:00
4fc15ecdfc
v0.3.0 step 4: type map + particle_type.target = "onehot"/"embedding"
lars
2026-08-06 15:43:48 +02:00
9112e845e0
v0.3.0 step 3: per-stage train.py trainers + pipeline.py/cli.py rewrite
lars
2026-08-06 11:31:49 +02:00
9ce55e5013
v0.3.0 step 2: network.py refactor to composable stage models
lars
2026-08-06 10:55:29 +02:00
eb6dd27406
v0.3.0 step 1: new nested config schema, v0.2 migration shim
lars
2026-08-06 10:15:20 +02:00
a489991a3b
Document the differentiability position and its validation obligation
lars
2026-08-06 09:57:44 +02:00
376bdb9d08
Refine v0.3.0 design: defaults, deferred scope, open questions
lars
2026-08-06 09:54:48 +02:00
f390884f67
Add v0.3.0 design doc: Stage-2 autoregressive redesign
lars
2026-08-04 15:25:32 +02:00
2fe887b49a
Bump version to 0.2.0
lars
2026-08-04 14:01:33 +02:00
803aae364e
format: Format tests/test_condor.py according to ruff styling
lars
2026-08-03 14:57:45 +02:00
057d637080
Fix test_write_submit_requires_synced_venv for active-venv resolution
lars
2026-08-03 14:52:41 +02:00
c3e5956718
Resolve giant condor wrapper from the active venv, not a hardcoded path
lars
2026-08-03 14:46:54 +02:00
dae6451203
Apply ruff format
lars
2026-08-03 14:04:17 +02:00
ca3a2a3462
Fix CLI/tooling robustness gaps and dedupe the Conditioning enum
lars
2026-08-03 13:49:09 +02:00
ad1b8e7835
Fix stale-partial reuse and n_chunks mismatch in analysis condor pipeline
lars
2026-08-03 13:48:43 +02:00
ad0341a9d4
Fix training-loop checkpoint/resume and WGAN bugs
lars
2026-08-03 13:48:22 +02:00
5b63dfd588
Fix conditioning="physical" so it can actually generalize past training vocab
lars
2026-08-03 13:47:59 +02:00
74343d3e48
Add data-integrity guards against silent NaN/Inf propagation and races
lars
2026-08-03 13:47:29 +02:00
a4c0443e01
Add bigger WGAN config (hidden_dim=512, n_blocks=6)
lars
2026-07-30 17:08:46 +02:00
22fdca7697
Merge pull request 'Add opt-in straight-through Gumbel-softmax combine weights to MoE router' (#26 ) from feat/router-gumbel-softmax into master
lars
2026-07-30 17:05:08 +02:00
8065df896e
Scope wandb run config to only-active hyperparameters
lars
2026-07-30 16:46:19 +02:00
5eec4c250a
Add gumbel/learn_centers/learn_width/learn_temperature to out-dir naming
lars
2026-07-30 16:42:56 +02:00
af2ee7c7ce
Add gumbel router configs sweeping learnable-knob combinations
lars
2026-07-30 16:40:02 +02:00
b51eafcfa5
Add opt-in straight-through Gumbel-softmax combine weights to MoE router
lars
2026-07-30 16:29:20 +02:00
43cb6dd9ae
Merge pull request 'Add learnable per-expert width and shared temperature to EnergyRouter' (#25 ) from feat/router-learnable-width into master
lars
2026-07-30 16:18:35 +02:00
aa55c407ab
Merge branch 'master' into feat/router-learnable-width
lars
2026-07-30 15:58:55 +02:00
da5f54ea1c
Add learnable per-expert width and shared temperature to EnergyRouter
lars
2026-07-30 15:33:52 +02:00
78297456e3
Merge pull request 'Store a quantile grid instead of a raw reservoir sample in the setup cache' (#24 ) from perf/quantile-energy-reservoir into master
lars
2026-07-30 13:59:17 +02:00
d656cf3109
Store a quantile grid instead of a raw reservoir sample in the setup cache
lars
2026-07-30 13:32:46 +02:00
de5db25e3f
Add giant new-run to scaffold a config.toml + run dir ahead of training
lars
2026-07-30 12:55:17 +02:00
1fd2625889
Merge pull request 'Offset event_id per file to avoid cross-file collisions' (#23 ) from fix/event-id-file-offset into master
lars
2026-07-30 11:44:31 +02:00
b944bba8fb
Offset event_id per file to avoid cross-file collisions
lars
2026-07-30 11:36:48 +02:00
e288c3fe21
Merge pull request 'Cache giant train's setup stage in a sidecar file' (#21 ) from perf/setup-stage into master
lars
2026-07-30 11:02:11 +02:00
5aaf6cde4d
Merge branch 'master' into perf/setup-stage
lars
2026-07-30 10:57:48 +02:00
471a81b5e7
Add dwarf warm-cache to precompute the setup-stage sidecar
lars
2026-07-30 10:53:31 +02:00
26aa9d3fde
Pass --seed through to the train/val event split
lars
2026-07-30 10:29:19 +02:00
e7478c36fb
Cache giant train's setup stage in a sidecar file
lars
2026-07-30 10:28:37 +02:00
9112625a08
Merge pull request 'Speed up giant train's setup stage' (#20 ) from perf/setup-stage into master
lars
2026-07-29 14:02:27 +02:00
84efbf5c2c
Merge branch 'master' into perf/setup-stage
lars
2026-07-29 13:56:59 +02:00
759b67a9e1
Speed up _WelfordAccumulator's per-chunk update
lars
2026-07-29 13:55:07 +02:00
47a6c9db1f
Add regression coverage for vocab/process index-map builders
lars
2026-07-29 13:45:33 +02:00
e331148afa
Merge pull request 'Make default checkpoint out_dir name reflect only non-default hyperparams' (#19 ) from feature/smart-outdir-naming into master
lars
2026-07-29 13:38:16 +02:00
09e4c765c7
Speed up giant train's setup stage
lars
2026-07-29 13:34:32 +02:00
1115451c8e
Make default checkpoint out_dir name reflect only non-default hyperparams
lars
2026-07-29 11:21:02 +02:00
f2f89023d5
Merge pull request 'Feature/wandb integration' (#18 ) from feature/wandb-integration into master
lars
2026-07-29 10:52:52 +02:00
4c19072724
Skip empty-slice mean/std in sec phys validation print
lars
2026-07-29 10:49:12 +02:00
539b6f61e1
Add test coverage for resolve_expert_dims
lars
2026-07-29 10:42:44 +02:00
f427d3384f
Timestamp default checkpoint dir to avoid W&B run-id collisions
lars
2026-07-29 10:42:40 +02:00
bb699d41b2
Persist global_step across --resume so W&B step stays monotonic
lars
2026-07-29 10:42:35 +02:00
a986f96ba3
Log router health, WGAN grad-norm split, n_sec accuracy, GPU/throughput to W&B
lars
2026-07-29 10:21:03 +02:00
969c5c6e9a
Fix router experts silently ignoring --hidden-dim/--n-blocks
lars
2026-07-27 16:34:25 +02:00
29459ab1f7
Log batch-level metrics to W&B, not just per-epoch summaries
lars
2026-07-27 16:20:05 +02:00
a05837f918
Apply ruff format
lars
2026-07-27 16:11:05 +02:00
0778a61360
Add opt-in Weights & Biases logging for the training loop
lars
2026-07-27 16:07:30 +02:00
db0f12be58
Add configs for router energy (embedding/physical) and WGAN baseline runs
lars
2026-07-27 15:53:04 +02:00
91aed19d4b
Merge remote-tracking branch 'origin/master' into condor-gpu-train-rollout
lars
2026-07-27 15:01:54 +02:00
6208810089
Merge pull request 'Refactor the analysis plot creation with focus on rollout' (#16 ) from analysis-rollout-plots into master
lars
2026-07-27 14:58:51 +02:00
5594866eac
Merge remote-tracking branch 'origin/master' into analysis-rollout-plots
lars
2026-07-27 14:55:00 +02:00
0f7febfd02
ci: set UV_LINK_MODE=copy to silence the cross-filesystem hardlink warning
lars
2026-07-27 14:40:01 +02:00
19be455346
ci: re-pin UV_CACHE_DIR after setup-uv, which exports its own value regardless of enable-cache
lars
2026-07-27 14:31:22 +02:00
1787d26d1c
ci: stop setup-uv from overriding UV_CACHE_DIR
lars
2026-07-27 14:22:49 +02:00
0afa75ee30
ci: replace unsupported artifact sharing with a bind-mounted uv cache
lars
2026-07-27 14:18:54 +02:00
08c76a9614
ci: share one uv sync across jobs, gate tests on lint+type-check, sync tag/version on release tags
lars
2026-07-27 14:00:59 +02:00
b8a4dc7d63
analyze: thread full model/training/rollout/dataset params to plots
lars
2026-07-27 13:35:02 +02:00
4feb9007f3
docs: note the EnergyRouter centers_init fix in the roadmap
lars
2026-07-27 13:14:49 +02:00
ee29b9a303
router: seed EnergyRouter centers from data quantiles instead of a fixed linspace
lars
2026-07-27 13:11:16 +02:00
641bbb0a68
docs: record first MoE router rollout benchmark result in the roadmap
lars
2026-07-27 12:02:59 +02:00
61410ddee3
analyze: default run directory to <repo>/analysis_runs, gitignored
lars
2026-07-27 11:29:35 +02:00
a88b21ef70
style: ruff format
lars
2026-07-27 11:19:06 +02:00
eb751d968d
transforms: pad legacy cond normalizers for pre-physical-conditioning checkpoints
lars
2026-07-27 11:06:48 +02:00
4d6101dcd7
analyze: recalibrate condor walltime model from real cluster timings
lars
2026-07-27 11:06:40 +02:00
c7194701f6
analyze: raise default condor job memory request to 8192 MB
lars
2026-07-27 10:31:46 +02:00
dbd5c7e083
analyze: default condor docker image to alma9-gridjob
lars
2026-07-27 10:14:15 +02:00
fa59443339
analyze: run condor compute jobs via .venv/bin/giant, not uv run
lars
2026-07-27 10:03:02 +02:00
e380400fe9
analyze: estimate per-job HTCondor walltime from chunk row count
lars
2026-07-27 09:47:44 +02:00
85d3914a4d
analyze: expose bin/pdg options on analyze submit
lars
2026-07-27 09:30:54 +02:00
86fc46b5a8
analyze: chunk per-plot aggregation across HTCondor jobs
lars
2026-07-27 09:24:19 +02:00
70d0f04326
analyze: normalize pdg dtype in open_side to fix rollout/reference concat
lars
2026-07-24 14:32:06 +02:00
lars