Geson-anko

流れはこんな感じです。

graph TD
    A["Create Shared Object Pool"] --> B["Instantiate Threads and Share."]
    B --> C["Start Interaction Thread"]
    B --> D["Start Training Thread"]
    B --> E["Start Main Thread"]

Geson-anko

import copy
import threading
import time

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader, TensorDataset

Geson-anko

import torch
import torch.nn as nn
import torch.backends.mps


class ModuleWithDevice(nn.Module):
    def __init__(self, *args, default_device=torch.device("cpu"), **kwargs) -> None:
        super().__init__(*args, **kwargs)
        # If this module has no parameters, returns this value.
        self._default_device = default_device

Geson-anko

import copy
from typing import Any, Self


class Reconstructable:
    _init_args: tuple[Any, ...]
    _init_kwds: dict[str, Any]

    @classmethod
    def reconstructable_init(cls, *args: Any, **kwds: Any) -> Self:

Geson-anko

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch import Tensor, Size
from torch.distributions import Distribution, Normal
import warnings

class NormalMixture(Distribution):
    """Computes Mixture Density Distribution of Normal distribution."""
    SQRT_2_PI = (2 * torch.pi) ** 0.5

Geson-anko

"""
AMI Time Module

This module provides a custom implementation of time-related functions with time acceleration and pause/resume features.
It wraps the standard Python time module and allows for consistent time control across the AMI system.

Key features:
- Time acceleration: Adjust the speed of time passage in the system.
- Pause/Resume: Ability to pause and resume the flow of time in the system.
- Thread-safe: All operations are protected by locks for use in multi-threaded environments.

Geson-anko

Benchmarking serialization methods...
Format: numpy().tobytes() / data_ptr() / torch.save
------------------------------------------------------------

Tensor shape: (100,)
Number of iterations: 10000

Average times per operation (ms):
Serialization:   ['0.001', '0.000', '0.020']
Deserialization: ['0.001', '0.001', '0.024']

Geson-anko

"""PyTorch trainer implementation for pamiq-core.

This module provides a base class for implementing PyTorch model training within the
pamiq-core framework. It handles optimizer and learning rate scheduler configuration,
state management, and integrates with the pamiq-core training system.
"""

from abc import abstractmethod
from pathlib import Path
from typing import Any, override