Discrete Flow Matching Model

Discrete Flow Matching model trained on discretized Moons dataset

Model Details

Model Type: discrete_flow_matching_moons
Training Epochs: 20000
Batch Size: 2000
Learning Rate: 0.001
CFG Parameter (eta): 0.1
Vocabulary Size: 100
Data Scale: 10.0
Data Standard Deviation: 1.0

Architecture

Embedding Dimension: 128
Data Dimensionality: 2

Usage

from huggingface_hub import hf_hub_download
import torch
from model_config import FlowMatchingConfig

# Download model and config
model_path = hf_hub_download("derekwong/discrete-flow-matching-moons", "pytorch_model.bin")
config_path = hf_hub_download("derekwong/discrete-flow-matching-moons", "config.json")

# Load config
import json
with open(config_path) as f:
    config_dict = json.load(f)
config = FlowMatchingConfig.from_dict(config_dict)

# Load model (you'll need the DiscreteFlow class)
model = DiscreteFlow(dim=config.dim, h=config.embedding_dimension, v=config.vocab_size)
model.load_state_dict(torch.load(model_path))
model.eval()

Training

This model was trained using Discrete Flow Matching on the discretized Moons dataset.