# DLCoal

**Repository Path**: larryivanhan/dlcoal

## Basic Information

- **Project Name**: DLCoal
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-11-03
- **Last Updated**: 2025-11-03

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Ne-Forge

Prototype repository for the Ne-Forge project. See `AGENTS.md` for
contributor guidance and the `configs/` directory for runnable examples.

## Quickstart

```bash
python -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
pip install -e .[dev]
```

### Training

Small MPS-friendly run:

```bash
python scripts/train_model.py --config configs/train_small_mps.yaml
```

Medium-stress synthetic run for reliability checks:

```bash
python scripts/train_model.py --config configs/train_medium_mps.yaml
```

Long CUDA run (48 GB vGPU, AMP + grad accumulation):

```bash
python scripts/train_model.py --config configs/train_large_cuda.yaml --device cuda
```

One-click launcher (auto env setup + sanity checks):

```bash
FORCE_DATASET_GENERATION=1 bash scripts/run_large_cuda.sh
```

`scripts/run_large_cuda.sh` will:

- Build/activate `.venv` if needed
- Generate the synthetic tensor dataset via `scripts/create_tensor_dataset.py` (leverages multi-core chunking)
- Launch CUDA training with AMP, gradient accumulation, checkpoints, and resumable state

Tune dataset generation via env vars, e.g.:

```bash
DATASET_NUM_WORKERS=16 DATASET_CHUNK_SIZE=256 DATASET_SUBCHUNK_SIZE=32 bash scripts/run_large_cuda.sh
```

### Visualisation

Plot learning curves:

```bash
python scripts/plot_training.py --run-dir runs/medium_mps --show
```

Compare predictions with ground truth (plots saved under `plots/`):

```bash
python scripts/compare_predictions.py \
  --config configs/train_medium_mps.yaml \
  --run-dir runs/medium_mps \
  --num-samples 6
```