github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-07 20:10:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[misc] update pre-commit and run all files (#4752)

Browse Source 
* [misc] update pre-commit

* [misc] run pre-commit

* [misc] remove useless configuration files

* [misc] ignore cuda for clang-format
This commit is contained in:
Hongxin Liu
2023-09-19 14:20:26 +08:00
committed by GitHub
parent 3c6b831c26
commit 079bf3cb26
1268 changed files with 50037 additions and 38444 deletions
Download Patch File Download Diff File Expand all files Collapse all files

205
tests/test_shardformer/test_model/test_shard_t5.py
View File

@@ -1,6 +1,5 @@
import pytest
import torch
from torch.nn.parallel import DistributedDataParallel as DDP
import colossalai
from colossalai.logging import disable_existing_loggers
@@ -21,19 +20,13 @@ from tests.test_shardformer.test_model._utils import (
def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn, test_config):
org_model, org_optimizer, sharded_model, sharded_optimizer, criterion, booster = build_model_from_hybrid_plugin(
model_fn, loss_fn, test_config
)
org_model, org_optimizer, sharded_model, sharded_optimizer, criterion, booster = \
build_model_from_hybrid_plugin(model_fn, loss_fn, test_config)
org_loss, org_output, sharded_loss, sharded_output = \
run_forward_backward_with_hybrid_plugin(
org_model,
sharded_model,
sharded_optimizer,
data_gen_fn,
output_transform_fn,
criterion,
booster)
org_loss, org_output, sharded_loss, sharded_output = run_forward_backward_with_hybrid_plugin(
org_model, sharded_model, sharded_optimizer, data_gen_fn, output_transform_fn, criterion, booster
)
stage_manager = booster.plugin.stage_manager
tp_group = booster.plugin.tp_group
@@ -42,22 +35,18 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
t5 = unwrap_model(org_model)
sharded_t5 = unwrap_model(sharded_model)
row_layer_for_check = ['shared', 'encoder.block[0].layer[0].SelfAttention.q']
row_layer_for_check = ["shared", "encoder.block[0].layer[0].SelfAttention.q"]
# Save gradient tensors for comparison between the original model and the sharded model before optimizer step.
grads_to_check = {}
if test_config['precision'] == 'fp32':
if test_config["precision"] == "fp32":
atol, rtol = 1e-5, 1e-3
else:
atol, rtol = 5e-3, 5e-3
if (stage_manager is None or stage_manager.is_first_stage()) and booster.plugin.zero_stage == 0:
row_layer_grads = get_grad_tensors_for_check(t5,
sharded_t5,
row_layer_for_check,
tp_group,
atol=atol,
rtol=rtol,
dim=0)
row_layer_grads = get_grad_tensors_for_check(
t5, sharded_t5, row_layer_for_check, tp_group, atol=atol, rtol=rtol, dim=0
)
grads_to_check.update(row_layer_grads)
# optimizer executes step
@@ -66,18 +55,18 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
# check last hidden state & loss
if stage_manager is None or stage_manager.is_last_stage():
if test_config['precision'] == 'fp32':
if test_config["precision"] == "fp32":
atol, rtol = 1e-5, 1e-3
else:
atol, rtol = 5e-3, 5e-3
if org_model.__class__.__name__ != 'T5ForConditionalGeneration':
if org_model.__class__.__name__ != "T5ForConditionalGeneration":
check_output_hidden_state(org_output, sharded_output, stage_manager, atol=atol, rtol=rtol)
check_loss(org_loss, sharded_loss, atol=atol, rtol=rtol)
# check weights
if test_config['precision'] == 'fp32':
if test_config["precision"] == "fp32":
atol, rtol = 5e-4, 1e-3
else:
atol, rtol = 5e-3, 5e-3
@@ -90,67 +79,70 @@ def check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn,
torch.cuda.empty_cache()
@parameterize('test_config', [{
'tp_size': 2,
'pp_size': 2,
'num_microbatches': 2,
'enable_all_optimization': True,
'use_lazy_init': True,
'precision': 'fp16',
'initial_scale': 1,
}, {
'tp_size': 1,
'pp_size': 2,
'num_microbatches': 4,
'use_lazy_init': False,
'precision': 'fp16',
'initial_scale': 1,
}, {
'tp_size': 4,
'pp_size': 1,
'enable_all_optimization': True,
'use_lazy_init': False,
'precision': 'fp32',
}, {
'tp_size': 1,
'pp_size': 4,
'num_microbatches': 4,
'enable_all_optimization': False,
'use_lazy_init': False,
'precision': 'fp32'
}, {
'tp_size': 2,
'pp_size': 1,
'enable_all_optimization': True,
'use_lazy_init': False,
'precision': 'fp32'
}, {
'tp_size': 2,
'pp_size': 1,
'enable_all_optimization': True,
'use_lazy_init': True,
'zero_stage': 2,
'precision': 'fp16',
'initial_scale': 1
}, {
'tp_size': 1,
'pp_size': 2,
'num_microbatches': 2,
'enable_all_optimization': True,
'use_lazy_init': True,
'zero_stage': 1,
'precision': 'fp16',
'initial_scale': 1
}])
@parameterize(
"test_config",
[
{
"tp_size": 2,
"pp_size": 2,
"num_microbatches": 2,
"enable_all_optimization": True,
"use_lazy_init": True,
"precision": "fp16",
"initial_scale": 1,
},
{
"tp_size": 1,
"pp_size": 2,
"num_microbatches": 4,
"use_lazy_init": False,
"precision": "fp16",
"initial_scale": 1,
},
{
"tp_size": 4,
"pp_size": 1,
"enable_all_optimization": True,
"use_lazy_init": False,
"precision": "fp32",
},
{
"tp_size": 1,
"pp_size": 4,
"num_microbatches": 4,
"enable_all_optimization": False,
"use_lazy_init": False,
"precision": "fp32",
},
{"tp_size": 2, "pp_size": 1, "enable_all_optimization": True, "use_lazy_init": False, "precision": "fp32"},
{
"tp_size": 2,
"pp_size": 1,
"enable_all_optimization": True,
"use_lazy_init": True,
"zero_stage": 2,
"precision": "fp16",
"initial_scale": 1,
},
{
"tp_size": 1,
"pp_size": 2,
"num_microbatches": 2,
"enable_all_optimization": True,
"use_lazy_init": True,
"zero_stage": 1,
"precision": "fp16",
"initial_scale": 1,
},
],
)
@clear_cache_before_run()
def run_t5_test(test_config):
sub_model_zoo = model_zoo.get_sub_registry('transformers_t5')
sub_model_zoo = model_zoo.get_sub_registry("transformers_t5")
for name, (model_fn, data_gen_fn, output_transform_fn, loss_fn, _) in sub_model_zoo.items():
# skip 4-stage pp test for t5_encoder
if test_config['pp_size'] > 2 and name == 'transformers_t5_encoder_model':
if test_config["pp_size"] > 2 and name == "transformers_t5_encoder_model":
continue
check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn, test_config)
@@ -160,29 +152,32 @@ def run_t5_test(test_config):
torch.cuda.empty_cache()
@parameterize('test_config', [
{
'tp_size': 2,
'pp_size': 2,
'num_microbatches': 4,
'enable_all_optimization': False,
'use_lazy_init': False,
'precision': 'fp32',
'initial_scale': 1,
},
{
'tp_size': 2,
'pp_size': 2,
'num_microbatches': 4,
'enable_all_optimization': False,
'use_lazy_init': False,
'precision': 'fp16',
'zero_stage': 1,
'initial_scale': 1,
},
])
@parameterize(
"test_config",
[
{
"tp_size": 2,
"pp_size": 2,
"num_microbatches": 4,
"enable_all_optimization": False,
"use_lazy_init": False,
"precision": "fp32",
"initial_scale": 1,
},
{
"tp_size": 2,
"pp_size": 2,
"num_microbatches": 4,
"enable_all_optimization": False,
"use_lazy_init": False,
"precision": "fp16",
"zero_stage": 1,
"initial_scale": 1,
},
],
)
def run_t5_3d_test(test_config):
sub_model_zoo = model_zoo.get_sub_registry('transformers_t5')
sub_model_zoo = model_zoo.get_sub_registry("transformers_t5")
for name, (model_fn, data_gen_fn, output_transform_fn, loss_fn, _) in sub_model_zoo.items():
check_forward_backward(model_fn, data_gen_fn, output_transform_fn, loss_fn, test_config)
@@ -193,13 +188,13 @@ def run_t5_3d_test(test_config):
def check_t5(rank, world_size, port):
disable_existing_loggers()
colossalai.launch(config={}, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
colossalai.launch(config={}, rank=rank, world_size=world_size, host="localhost", port=port, backend="nccl")
run_t5_test()
def check_t5_3d(rank, world_size, port):
disable_existing_loggers()
colossalai.launch(config={}, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
colossalai.launch(config={}, rank=rank, world_size=world_size, host="localhost", port=port, backend="nccl")
run_t5_3d_test()