Lars Bogner

lars pushed to master at lars/giant

2026-06-22 07:27:02 +02:00

46068f356f Add hyperparameter scan

lars pushed to master at lars/giant

2026-06-19 13:24:42 +02:00

aef0a588ce Add --batch-size auto to estimate batch size from free GPU memory

lars pushed to master at lars/giant

2026-06-19 13:17:15 +02:00

74d0883868 Remove scripts/train.py in favor of the giant train CLI

lars pushed to master at lars/giant

2026-06-19 10:43:31 +02:00

cf653f7664 Add linear warmup before cosine LR decay

lars pushed to master at lars/giant

2026-06-18 17:52:14 +02:00

7f62141445 Make sampler step count configurable for validation

lars pushed to master at lars/giant

2026-06-18 17:43:24 +02:00

e505e2d141 Fix miniCaloSim link in README

lars pushed to master at lars/giant

2026-06-18 17:43:04 +02:00

0cdb5db947 Update README to match current architecture and tooling

lars pushed to master at lars/giant

2026-06-18 17:40:27 +02:00

8cebc4809d Apply ruff format and document lint/type tooling in CLAUDE.md

lars pushed to master at lars/giant

2026-06-18 17:38:40 +02:00

53fd2e4405 Add ruff and ty as dev dependencies, fix lint/type findings

lars pushed to master at lars/giant

2026-06-18 17:30:35 +02:00

c1a8a900d3 Split torch into cpu/cuda extras and pin dependency version bounds

lars pushed to master at lars/giant

2026-06-18 17:06:58 +02:00

80b198c1a5 Add lazy polars I/O and duplicate KL/constraint checks for giant.analysis

lars pushed to master at lars/giant

2026-06-18 16:36:19 +02:00

a867fc4aae Add giant.analysis module for notebook-based model quality diagnostics

lars pushed to master at lars/giant

2026-06-18 15:59:20 +02:00

da84801625 Add configurable dropout to ResBlocks

lars pushed to master at lars/giant

2026-06-18 14:00:01 +02:00

5d161a52b2 Add graceful shutdown on SIGINT/SIGTERM

lars pushed to master at lars/giant

2026-06-18 13:45:12 +02:00

893d91e749 Add KL divergence to marginal validation and hook it into the training loop

lars pushed to master at lars/giant

2026-06-18 13:32:37 +02:00

43634ef77a Dedup training pipeline, add seeding/resume and per-epoch metrics logging

lars pushed to master at lars/giant

2026-06-18 10:55:58 +02:00

f1a82b5853 Add --coord local mode to predict for raw-space prediction debugging

lars pushed to master at lars/giant

2026-06-18 10:36:55 +02:00

72bd65ff9f Add post_pos as a model target via travel_dir decomposition

lars pushed to master at lars/giant

2026-06-18 10:06:53 +02:00

6e2fe12a6d Fix installed torch version to be compatible with cuda drivers

lars pushed to master at lars/giant

2026-06-18 10:06:15 +02:00

c3b7b2744c Batch StreamingStepsDataset internally instead of per-row collate