ColossalAI/colossalai/builder/builder.py

80 lines
2.9 KiB
Python
Raw Normal View History

2021-10-28 16:21:23 +00:00
#!/usr/bin/env python
# -*- encoding: utf-8 -*-
import inspect
from colossalai.registry import *
def build_from_config(module, config: dict):
"""Returns an object of :class:`module` constructed from `config`.
2022-03-25 05:02:39 +00:00
Args:
module: A python or user-defined class
config: A python dict containing information used in the construction of the return object
Returns: An ``object`` of interest
Raises:
AssertionError: Raises an AssertionError if `module` is not a class
2021-10-28 16:21:23 +00:00
"""
assert inspect.isclass(module), 'module must be a class'
return module(**config)
def build_from_registry(config, registry: Registry):
2022-03-25 05:02:39 +00:00
r"""Returns an object constructed from `config`, the type of the object
2021-10-28 16:21:23 +00:00
is specified by `registry`.
2022-03-25 05:02:39 +00:00
Note:
2022-03-31 03:36:56 +00:00
the `config` is used to construct the return object such as `LAYERS`, `OPTIMIZERS`
and other support types in `registry`. The `config` should contain
all required parameters of corresponding object. The details of support
types in `registry` and the `mod_type` in `config` could be found in
`registry <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/registry/__init__.py>`_.
2022-03-25 05:02:39 +00:00
Args:
config (dict or :class:`colossalai.context.colossalai.context.Config`): information
used in the construction of the return object.
registry (:class:`Registry`): A registry specifying the type of the return object
2022-03-31 03:36:56 +00:00
Returns:
A Python object specified by `registry`.
2022-03-25 05:02:39 +00:00
Raises:
2022-03-31 03:36:56 +00:00
Exception: Raises an Exception if an error occurred when building from registry.
2021-10-28 16:21:23 +00:00
"""
config_ = config.copy() # keep the original config untouched
assert isinstance(registry, Registry), f'Expected type Registry but got {type(registry)}'
2021-10-28 16:21:23 +00:00
mod_type = config_.pop('type')
assert registry.has(mod_type), f'{mod_type} is not found in registry {registry.name}'
2021-10-28 16:21:23 +00:00
try:
obj = registry.get_module(mod_type)(**config_)
except Exception as e:
print(f'An error occurred when building {mod_type} from registry {registry.name}', flush=True)
2021-10-28 16:21:23 +00:00
raise e
return obj
2021-10-28 16:21:23 +00:00
def build_gradient_handler(config, model, optimizer):
"""Returns a gradient handler object of :class:`BaseGradientHandler` constructed from `config`,
`model` and `optimizer`.
2022-03-25 05:02:39 +00:00
Args:
config (dict or :class:`colossalai.context.Config`): A python dict or
a :class:`colossalai.context.Config` object containing information
used in the construction of the ``GRADIENT_HANDLER``.
model (:class:`nn.Module`): A model containing parameters for the gradient handler
optimizer (:class:`torch.optim.Optimizer`): An optimizer object containing parameters for the gradient handler
Returns:
An object of :class:`colossalai.engine.BaseGradientHandler`
2021-10-28 16:21:23 +00:00
"""
config_ = config.copy()
Develop/experiments (#59) * Add gradient accumulation, fix lr scheduler * fix FP16 optimizer and adapted torch amp with tensor parallel (#18) * fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes * fixed trainer * Revert "fixed trainer" This reverts commit 2e0b0b76990e8d4e337add483d878c0f61cf5097. * improved consistency between trainer, engine and schedule (#23) Co-authored-by: 1SAA <c2h214748@gmail.com> * Split conv2d, class token, positional embedding in 2d, Fix random number in ddp Fix convergence in cifar10, Imagenet1000 * Integrate 1d tensor parallel in Colossal-AI (#39) * fixed 1D and 2D convergence (#38) * optimized 2D operations * fixed 1D ViT convergence problem * Feature/ddp (#49) * remove redundancy func in setup (#19) (#20) * use env to control the language of doc (#24) (#25) * Support TP-compatible Torch AMP and Update trainer API (#27) * Add gradient accumulation, fix lr scheduler * fix FP16 optimizer and adapted torch amp with tensor parallel (#18) * fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes * fixed trainer * Revert "fixed trainer" This reverts commit 2e0b0b76990e8d4e337add483d878c0f61cf5097. * improved consistency between trainer, engine and schedule (#23) Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: ver217 <lhx0217@gmail.com> * add an example of ViT-B/16 and remove w_norm clipping in LAMB (#29) * add explanation for ViT example (#35) (#36) * support torch ddp * fix loss accumulation * add log for ddp * change seed * modify timing hook Co-authored-by: Frank Lee <somerlee.9@gmail.com> Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: binmakeswell <binmakeswell@gmail.com> * Feature/pipeline (#40) * remove redundancy func in setup (#19) (#20) * use env to control the language of doc (#24) (#25) * Support TP-compatible Torch AMP and Update trainer API (#27) * Add gradient accumulation, fix lr scheduler * fix FP16 optimizer and adapted torch amp with tensor parallel (#18) * fixed bugs in compatibility between torch amp and tensor parallel and performed some minor fixes * fixed trainer * Revert "fixed trainer" This reverts commit 2e0b0b76990e8d4e337add483d878c0f61cf5097. * improved consistency between trainer, engine and schedule (#23) Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: ver217 <lhx0217@gmail.com> * add an example of ViT-B/16 and remove w_norm clipping in LAMB (#29) * add explanation for ViT example (#35) (#36) * optimize communication of pipeline parallel * fix grad clip for pipeline Co-authored-by: Frank Lee <somerlee.9@gmail.com> Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: binmakeswell <binmakeswell@gmail.com> * optimized 3d layer to fix slow computation ; tested imagenet performance with 3d; reworked lr_scheduler config definition; fixed launch args; fixed some printing issues; simplified apis of 3d layers (#51) * Update 2.5d layer code to get a similar accuracy on imagenet-1k dataset * update api for better usability (#58) update api for better usability Co-authored-by: 1SAA <c2h214748@gmail.com> Co-authored-by: ver217 <lhx0217@gmail.com> Co-authored-by: puck_WCR <46049915+WANG-CR@users.noreply.github.com> Co-authored-by: binmakeswell <binmakeswell@gmail.com> Co-authored-by: アマデウス <kurisusnowdeng@users.noreply.github.com> Co-authored-by: BoxiangW <45734921+BoxiangW@users.noreply.github.com>
2021-12-09 07:08:29 +00:00
config_['model'] = model
config_['optimizer'] = optimizer
return build_from_registry(config_, GRADIENT_HANDLER)