Distributed training strategy submission

miner_diloco_sparse_ef.py

import math
import torch
import torch.nn.utils as nn_utils
import torch.distributed as dist

from copy import deepcopy
from dataclasses import dataclass
from torch.optim.lr_scheduler import LambdaLR
from typing import Dict, Any, Type, Optional, Union, List
from abc import ABC, abstractmethod


def all_reduce(tensor):
    return dist.all_reduce(tensor, op=dist.ReduceOp.SUM)


@dataclass
class OptimSpec:
    cls: Type[torch.optim.Optimizer]
    kwargs: Dict[str, Any]

    def build(self, model):
        return self.cls(model.parameters(), **self.kwargs)


def ensure_optim_spec(
    optim: Union[str, OptimSpec, None],
    default: Optional[OptimSpec] = None,
    **kwargs,
):
    if optim is None:
        return default or OptimSpec(torch.optim.AdamW, kwargs)
    if isinstance(optim, OptimSpec):
        return optim
    if isinstance(optim, str):
        name = optim.lower()
        mapping = {
            "adam": torch.optim.Adam,
            "adamw": torch.optim.AdamW,
            "sgd": torch.optim.SGD,
        }
        return OptimSpec(mapping[name], kwargs)
    raise TypeError


class Strategy(ABC):
    def __init__(self, **kwargs):
        self.max_steps = 100
        self.local_step = 0
        self.nbytes = 0
        for k, v in kwargs.items():
            setattr(self, k, v)

    def _init_node(self, model, rank, num_nodes):
        self.model = model
        self.rank = rank
        self.num_nodes = num_nodes
        if hasattr(self, "optim_spec"):
            self.optim = self.optim_spec.build(model)

    def zero_grad(self):
        self.optim.zero_grad()

    @abstractmethod
    def step(self):
        self.local_step += 1


class CommunicationModule(ABC):
    @abstractmethod
    def _init_node(self, model, rank, num_nodes):
        pass

    @abstractmethod
    def communicate(self, model, rank, num_nodes, local_step):
        pass


class SparseAsymDiLoCoEF(CommunicationModule):
    """
    ASYMMETRIC
    Worker -> Rank0 only
    NO broadcast back
    """

    def __init__(
        self,
        k_ratio=0.005,
        outer_lr=0.2,
    ):
        self.k_ratio = k_ratio
        self.outer_lr = outer_lr
        self.master_model = None
        self.outer_optim = None
        self.error_buffers = {}

    def _init_node(self, model, rank, num_nodes):
        for n, p in model.named_parameters():
            self.error_buffers[n] = torch.zeros_like(p.data)

        if rank == 0:
            self.master_model = deepcopy(model).cpu()
            self.outer_optim = torch.optim.SGD(
                self.master_model.parameters(),
                lr=self.outer_lr,
                momentum=0.9,
            )

    def _sparsify(self, tensor, name, strategy):
        ef = self.error_buffers[name]
        tensor = tensor + ef

        flat = tensor.view(-1)
        k = max(1, int(self.k_ratio * flat.numel()))

        _, idx = torch.topk(flat.abs(), k, sorted=False)
        sparse = torch.zeros_like(flat)
        sparse[idx] = flat[idx]

        self.error_buffers[name] = (flat - sparse).view_as(tensor)

        strategy.nbytes += sparse.numel() * sparse.element_size()
        return sparse.view_as(tensor)

    def communicate(self, model, rank, num_nodes, local_step):
        if num_nodes == 1:
            return

        if local_step < 20:
            H = 4
        elif local_step < 60:
            H = 12
        else:
            H = 40

        if local_step % H != 0:
            return

        with torch.no_grad():
            for name, p in model.named_parameters():
                sparse = self._sparsify(p.data, name, self.strategy)
                all_reduce(sparse)

                if rank == 0:
                    p.data.copy_(sparse / num_nodes)

        if rank == 0:
            self.outer_optim.zero_grad()
            for mp, p in zip(self.master_model.parameters(), model.parameters()):
                mp.grad = mp.data - p.data.cpu()
            self.outer_optim.step()

        # ❌ NO BROADCAST BACK (INTENTIONAL)


class CommunicateOptimizeStrategy(Strategy):
    def __init__(
        self,
        communication_modules: List[CommunicationModule],
        optim_spec=None,
        max_norm=1.0,
        **kwargs,
    ):
        super().__init__(**kwargs)
        self.optim_spec = ensure_optim_spec(
            optim_spec, OptimSpec(torch.optim.AdamW, {"lr": 2e-3})
        )
        self.communication_modules = communication_modules
        self.max_norm = max_norm

        for m in self.communication_modules:
            m.strategy = self

    def _init_node(self, model, rank, num_nodes):
        super()._init_node(model, rank, num_nodes)
        self.optim = self.optim_spec.build(model)
        for m in self.communication_modules:
            m._init_node(model, rank, num_nodes)

    def step(self):
        if self.max_norm:
            nn_utils.clip_grad_norm_(self.model.parameters(), self.max_norm)

        self.optim.step()

        for m in self.communication_modules:
            m.communicate(self.model, self.rank, self.num_nodes, self.local_step)

        self.local_step += 1


class DiLoCoSparseEFBenchmark(CommunicateOptimizeStrategy):
    def __init__(self, **kwargs):
        comm = SparseAsymDiLoCoEF(
            k_ratio=0.005,
            outer_lr=0.2,
        )
        super().__init__(
            communication_modules=[comm],
            **kwargs,
        )


STRATEGY = DiLoCoSparseEFBenchmark(
    optim_spec=OptimSpec(torch.optim.AdamW, {"lr": 2e-3}),
    max_norm=1.0,
)