Distributed training strategy submission

miner_diloco_sparseloco.py

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

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

from exogym.aux.utils import LogModule

ParamsT: TypeAlias = Union[Iterable[torch.Tensor], Iterable[dict[str, Any]]]


if torch.cuda.is_available():
    torch.backends.cuda.matmul.allow_tf32 = True
    torch.backends.cudnn.allow_tf32 = True
    torch.backends.cudnn.benchmark = True
    try:
        torch.set_float32_matmul_precision("high")
    except Exception:
        pass


def mps_compatible(func):
    def all_gather_wrapper(tensor_list, tensor, *args, **kwargs):
        is_tensor_mps = hasattr(tensor, "device") and tensor.device.type == "mps"
        is_list_mps = any(hasattr(t, "device") and t.device.type == "mps" for t in tensor_list)
        if is_tensor_mps or is_list_mps:
            cpu_tensor = tensor.data.to("cpu") if is_tensor_mps else tensor
            cpu_tensor_list = [
                t.data.to("cpu") if hasattr(t, "device") and t.device.type == "mps" else t
                for t in tensor_list
            ]
            result = func(cpu_tensor_list, cpu_tensor, *args, **kwargs)
            if is_tensor_mps:
                tensor.data.copy_(cpu_tensor.to("mps"))
            for i, t in enumerate(tensor_list):
                if hasattr(t, "device") and t.device.type == "mps":
                    t.data.copy_(cpu_tensor_list[i].to("mps"))
            return result
        return func(tensor_list, tensor, *args, **kwargs)

    def standard_wrapper(tensor, *args, **kwargs):
        if hasattr(tensor, "device") and tensor.device.type == "mps":
            cpu_tensor = tensor.data.to("cpu")
            result = func(cpu_tensor, *args, **kwargs)
            tensor.data.copy_(cpu_tensor.to("mps"))
            return result
        return func(tensor, *args, **kwargs)

    return all_gather_wrapper if func.__name__ == "all_gather" else standard_wrapper


@mps_compatible
def broadcast(tensor, src=0):
    return dist.broadcast(tensor, src=src)


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


@mps_compatible
def all_gather(tensor_list, tensor, group=None, async_op=False):
    return dist.all_gather(tensor_list, tensor, group=group, async_op=async_op)


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

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


class Strategy(ABC, LogModule):
    def __init__(self, lr_scheduler=None, lr_scheduler_kwargs=None, **kwargs):
        self.lr_scheduler = lr_scheduler
        self.lr_scheduler_kwargs = lr_scheduler_kwargs or {}
        self.kwargs = kwargs
        self.scheduler = None
        self.lr_callbacks = []
        self.max_steps = 1

    def _init_node(self, model, rank, num_nodes):
        if torch.cuda.is_available():
            local_rank = int(os.environ.get("LOCAL_RANK", "0"))
            torch.cuda.set_device(local_rank)

        self.model = model
        self.rank = rank
        self.num_nodes = num_nodes
        self.local_step = 0

    @abstractmethod
    def step(self):
        self.nbytes = 0

        if self.scheduler is not None:
            self.scheduler.step()
            if self.rank == 0:
                for cb in self.lr_callbacks:
                    cb(self.scheduler.get_last_lr()[0])

        self.local_step += 1

    def zero_grad(self):
        self.optim.zero_grad(set_to_none=True)

    def _setup_scheduler(self):
        def lr_lambda(step: int):
            warmup = int(self.lr_scheduler_kwargs.get("warmup_steps", 80))
            max_steps = int(self.lr_scheduler_kwargs.get("max_steps", self.max_steps))
            max_steps = max(1, max_steps)
            warmup = max(1, min(warmup, max_steps))

            if step < warmup:
                return step / warmup

            progress = (step - warmup) / max(1, max_steps - warmup)
            progress = min(max(progress, 0.0), 1.0)
            return 0.5 * (1.0 + math.cos(math.pi * progress))

        if self.lr_scheduler == "lambda_cosine":
            self.scheduler = LambdaLR(self.optim, lr_lambda)

    def __config__(self):
        return {"strategy": self.__class__.__name__}


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

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


class CommunicateOptimizeStrategy(Strategy):
    def __init__(self, communication_modules, optim_spec=None, max_norm=None, **kwargs):
        super().__init__(**kwargs)
        self.communication_modules = communication_modules
        self.optim_spec = optim_spec
        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)
        self._setup_scheduler()
        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)

        super().step()


class DiLoCoCommunicator(CommunicationModule):
    def __init__(self, H=25, outer_optim_spec=None):
        self.H = H
        self.outer_optim_spec = outer_optim_spec

    def _init_node(self, model, rank, num_nodes):
        self.pg = dist.new_group(backend="gloo", timeout=datetime.timedelta(60))
        self.master_model = deepcopy(model).to("cpu")
        for p in self.master_model.parameters():
            p.requires_grad = True

        self.outer_optim = self.outer_optim_spec.cls(
            self.master_model.parameters(),
            process_group=self.pg,
            **self.outer_optim_spec.kwargs,
        )

    def communicate(self, model, rank, num_nodes, local_step):
        if num_nodes > 1 and local_step > 0 and local_step % self.H == 0:
            self.outer_optim.zero_grad(set_to_none=True)

            local_sd = model.state_dict()
            for n, p in self.master_model.named_parameters():
                p.grad = p.data - local_sd[n].data.to("cpu")

            self.outer_optim.step()

            master_sd = self.master_model.state_dict()
            for n, p in model.named_parameters():
                p.data.copy_(master_sd[n].to(p.device))


class DiLoCoStrategy(CommunicateOptimizeStrategy):
    def __init__(self, optim_spec, outer_optim_spec, H=25, **kwargs):
        self.comm = DiLoCoCommunicator(H=H, outer_optim_spec=outer_optim_spec)
        super().__init__(
            communication_modules=[self.comm],
            optim_spec=optim_spec,
            **kwargs,
        )


class SparseLoCo(torch.optim.SGD):
    def __init__(
        self,
        params,
        lr,
        momentum=0.9,
        weight_decay=0.05,
        top_k=64,
        chunk_size=64,
        use_dct=True,
        use_quantization=True,
        quantization_bins=4,
        quantization_range=6,
        process_group=None,
        **kwargs,
    ):
        super().__init__(params, lr=lr, momentum=momentum, weight_decay=0.0, **kwargs)
        self.decoupled_weight_decay = weight_decay
        self.process_group = process_group

    @torch.no_grad()
    def step(self, closure=None):
        return super().step(closure=closure)


STRATEGY = DiLoCoStrategy(
    optim_spec=OptimSpec(
        torch.optim.AdamW,
        {
            "lr": 0.0012,
            "betas": (0.9, 0.95),
            "weight_decay": 0.08,
            "eps": 1e-8,
        },
    ),
    outer_optim_spec=OptimSpec(
        SparseLoCo,
        {
            "lr": 0.6,
            "momentum": 0.9,
            "weight_decay": 0.05,
            "top_k": 64,
            "chunk_size": 64,
            "use_dct": True,
            "use_quantization": True,
            "quantization_bins": 4,
            "quantization_range": 6,
        },
    ),
    lr_scheduler="lambda_cosine",
    lr_scheduler_kwargs={
        "warmup_steps": 80,
        "max_steps": 100,
    },
    max_norm=1.0,
    H=25,
)