gameloader 50b1f76263 feat(models/denoiser): Add Mamba2 model implementation ...

Implement a minimal, single-file implementation of the Mamba-2 model in PyTorch.
Based on the paper "Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality".

2026-01-20 14:44:03 +08:00

..

callbacks

feat(callbacks): add Weights & Biases image logging to SaveImagesHook\n\n- Introduce optional wandb_max_save_num parameter in __init__\n- Add helper methods _get_wandb_logger and _log_wandb_samples\n- Log sampled images as wandb.Image artifacts during validation epoch end (global zero only)\n- Handle proper dtype conversion and clipping for wandb compatibility\n- Respect both general max_save_num and wandb-specific limit\n\nThis enables visualization of validation samples directly in the W&B dashboard without manual intervention.

2026-01-19 16:03:54 +08:00

data

feat(dataset): add HuggingFace image dataset

2026-01-18 17:10:01 +08:00

diffusion

disperse loss

2025-06-14 12:14:01 +08:00

models

feat(models/denoiser): Add Mamba2 model implementation

2026-01-20 14:44:03 +08:00

plugins

submit code

2025-04-09 11:01:16 +08:00

utils

feat(data): add support for Hugging Face datasets

2026-01-16 14:25:32 +08:00

__init__.py

submit code

2025-04-09 11:01:16 +08:00

lightning_data.py

feat(data): add support for Hugging Face datasets

2026-01-16 14:25:32 +08:00

lightning_model.py

torch.compile

2025-07-03 20:23:59 +08:00