Views
No views yet
| Version | Model | Params | Training | Tag |
|---|---|---|---|---|
| v017_local_baseline | LocalDiffusionPlanner | 7M | Offline BC, 200 demos/env, 30 epochs | — |
| v017_local_baseline | LocalDiffusionPlanner | 7M | Offline BC, 500 demos/env, 60 epochs | v0.17-local-baseline-gold (pending) |
ReMDM-MiniHack/
├── README.md # This file
├── v017_local_baseline/
│ ├── inference_weights.pth # EMA state dict (for evaluation)
│ ├── full_checkpoint.pth # Full training state (for resuming)
│ ├── config.json # Hyperparams + model args
│ └── eval_results.csv # Per-environment results
└── datasets/
└── oracle_demos_v017.pt # Oracle demonstration dataset1import torch
2from huggingface_hub import hf_hub_download
3
4# Download weights
5path = hf_hub_download("piotrwilam/ReMDM-MiniHack", "v017_local_baseline/inference_weights.pth")
6weights = torch.load(path, map_location="cpu", weights_only=False)
7
8# Load model
9from model import LocalDiffusionPlanner
10model = LocalDiffusionPlanner(action_dim=12)
11model.load_state_dict(weights)
12model.eval()| Environment | Win% | Avg Steps |
|---|---|---|
| Room-Random-5x5 | 94% | 18.3 |
| Room-Random-15x15 | 54% | 130.4 |
| Room-Dark-5x5 | 90% | 25.5 |
| Room-Ultimate-5x5 | 84% | 20.8 |
| Room-Ultimate-15x15 | 30% | 72.1 |
| Corridor-R2 | 42% | 132.1 |
| Corridor-R3 | 0% | 200.0 |
| MazeWalk-9x9 | 48% | 119.0 |
| MazeWalk-15x15 | 22% | 162.3 |