rebasin 0.0.21

An implementation of methods described in Git Re-basin-paper by Ainsworth et al.

rebasin

An implementation of methods described in "Git Re-basin"-paper by Ainsworth et al.

Installation

pip install rebasin

Usage

Currently, only weight-matching is implemented as a method for rebasing, and only a simplified form of linear interpolation is implemented.

import torch
from torch import nn
from rebasin import PermutationCoordinateDescent
from rebasin import interpolation

model_a, model_b, train_dl, val_dl, loss_fn = ...

def eval_fn(model: nn.Module, device: str | torch.device | None = None) -> float:
    loss = 0.0
    for inputs, logits in val_dl:
        if device is not None:
            inputs = inputs.to(device)
            logits = logits.to(device)
        outputs = model(inputs)
        loss = loss_fn(outputs, logits)
    return loss / len(val_dl)

input_data = next(iter(train_dl))[0]

# Rebasin
pcd = PermutationCoordinateDescent(model_a, model_b, input_data)
pcd.calculate_permutations()
pcd.apply_permutations()

# Interpolate
lerp = interpolation.LerpSimple(
    models=[model_a, model_b], 
    eval_fn=eval_fn,  # Can be any metric as long as the function takes a model and a device
    eval_mode="min",  # "min" or "max"
    train_dataloader=train_dl,  # Used to recalculate BatchNorm statistics; optional
)
lerp.interpolate(steps=10)

# Access model with lowest validation loss:
lerp.best_model