g5 d dl Z d dlZd dlZd dlZddlmZmZ ddlmZm Z m
Z
m Z m Z m
Z
mZmZ ddlmZ d Z ddZddZdS ) N )AcceleratorState PartialState)
PrecisionTypePrepareForLaunchare_libraries_initializedcheck_cuda_p2p_ib_support get_gpu_infois_mps_availableis_torch_versionpatch_environment)0ELASTIC_LOG_LINE_PREFIX_TEMPLATE_PYTORCH_VERSIONc " t } dS )z+Verify a `PartialState` can be initialized.N)r )_s P/var/www/html/ai-engine/env/lib/python3.11/site-packages/accelerate/launchers.py test_launchr # s AAA no29500 127.0.0.1static none皙?c
d}d}t d t j D rd}nAdt j v r3dt
t j d v } t | }nL# t $ r? t d|j dt j d w xY w|s|rt j
d d
xd d
lmc m} t# t$ j d k rt d |d
}t) | d }t+ d| d | |||d d
S |r\t/ d dk rDt0 j rt+ d nt+ d | | d
S |t d ||k rt d |dk rBd dlm }m } d dl m } d d l m!} t# t$ j d k rt d tE d }t# | d k r d }|D ] }|d | d!z
} tG | tI ||||z |||" }tK s
d#|d$< d#|d%< tM d*i |5 t j
d&d' d(k rmt) tN d) } ||d*|d nF# |$ r>}d+}d,|j( d v rtG | d- |tG | d.| |d
}~ww xY wt) | d) }t+ d/| d0 |
i }
|d1k r||
d2< | s| d3| } tI |||| | ||
| |
d4
} tS d5tT r|| d6< | |d*i | | 7 | n?# |$ r7}d,|j( d v rtG d8 |tG d9| |d
}~ww xY w d
d
d
d
S # 1 swxY w Y d
S tW r d#t j d:< t+ d; n=t0 j rt+ d nt+ d< | | d
S )=a
Launches a training function, using several processes or multiple nodes if it's possible in the current environment
(TPU with multiple cores for instance).
To use this function absolutely zero calls to a CUDA device must be made in the notebook session before calling. If
any have been made, you will need to restart the notebook and make sure no cells use any CUDA capability.
Setting `ACCELERATE_DEBUG_MODE="1"` in your environment will run a test before truly launching to ensure that none
of those calls have been made.
Args:
function (`Callable`):
The training function to execute. If it accepts arguments, the first argument should be the index of the
process run.
args (`Tuple`):
Tuple of arguments to pass to the function (it will receive `*args`).
num_processes (`int`, *optional*):
The number of processes to use for training. Will default to 8 in Colab/Kaggle if a TPU is available, to
the number of GPUs available otherwise.
mixed_precision (`str`, *optional*, defaults to `"no"`):
If `fp16` or `bf16`, will use mixed precision training on multi-GPU.
use_port (`str`, *optional*, defaults to `"29500"`):
The port to use to communicate between processes when launching a multi-GPU training.
master_addr (`str`, *optional*, defaults to `"127.0.0.1"`):
The address to use for communication between processes.
node_rank (`int`, *optional*, defaults to 0):
The rank of the current node.
num_nodes (`int`, *optional*, defaults to 1):
The number of nodes to use for training.
rdzv_backend (`str`, *optional*, defaults to `"static"`):
The rendezvous method to use, such as 'static' (the default) or 'c10d'
rdzv_endpoint (`str`, *optional*, defaults to `""`):
The endpoint of the rdzv sync. storage.
rdzv_conf (`Dict`, *optional*, defaults to `None`):
Additional rendezvous configuration.
rdzv_id (`str`, *optional*, defaults to `"none"`):
The unique run id of the job.
max_restarts (`int`, *optional*, defaults to 0):
The maximum amount of restarts that elastic agent will conduct on workers before failure.
monitor_interval (`float`, *optional*, defaults to 0.1):
The interval in seconds that is used by the elastic_agent as a period of monitoring workers.
log_line_prefix_template (`str`, *optional*, defaults to `None`):
The prefix template for elastic launch logging. Available from PyTorch 2.2.0.
Example:
```python
# Assume this is defined in a Jupyter Notebook on an instance with two GPUs
from accelerate import notebook_launcher
def train(*args):
# Your training function here
...
notebook_launcher(train, args=(arg1, arg2), num_processes=2, mixed_precision="fp16")
```
Fc 3 @ K | ]}| d V dS )KAGGLEN)
startswith).0keys r z$notebook_launcher..{ s.
A
A3>>( # #
A
A
A
A
A
Ar TIPythonz google.colabz Unknown mixed_precision mode: z. Choose between .TPU_NAMENr zTo train on TPU in Colab or Kaggle Kernel, the `Accelerator` should only be initialized inside your training function. Restart your notebook and make sure no cells initializes an `Accelerator`. XLA)distributed_typezLaunching a training on z TPU cores.forkargsnprocs start_methodr z Launching training on one GPU.z Launching training on one CPU.zcYou have to specify the number of GPUs you would like to use, add `num_processes=...` to your call.z4The node_rank must be less than the number of nodes.) LaunchConfigelastic_launchstart_processes)ProcessRaisedExceptionzTo launch a multi-GPU training from your notebook, the `Accelerator` should only be initialized inside your training function. Restart your notebook and make sure no cells initializes an `Accelerator`. bitsandbyteszCould not start distributed process. Libraries known to initialize CUDA upon import have been imported already. Please keep these imports inside your training function to try and help with this:z
* ``)nproc node_rank
world_size master_addr master_portmixed_precision1nccl_p2p_disablenccl_ib_disableACCELERATE_DEBUG_MODEfalsetrue MULTI_GPUr zQAn issue was found when verifying a stable environment for the notebook launcher.z.Cannot re-initialize CUDA in forked subprocesszThis likely stems from an outside import causing issues once the `notebook_launcher()` is called. Please review your imports and test them when running the `notebook_launcher()` to identify which one is problematic and causing CUDA to be initialized.z! The following error was raised: zLaunching training on z GPUs.r rank:)
min_nodes max_nodesnproc_per_noderun_id
rdzv_endpoint rdzv_backend rdzv_configs max_restartsmonitor_intervalr- z>=log_line_prefix_template)config
entrypointaU CUDA has been initialized before the `notebook_launcher` could create a forked subprocess. This likely stems from an outside import causing issues once the `notebook_launcher()` is called. Please review your imports and test them when running the `notebook_launcher()` to identify which one is problematic and causing CUDA to be initialized.z0An issue was found when launching the training: PYTORCH_ENABLE_MPS_FALLBACKzLaunching training on MPS.zLaunching training on CPU.),anyosenvironkeyssysmodulesstr get_ipythonr lower
ValueErrorr; listget)torch_xla.distributed.xla_multiprocessing distributedxla_multiprocessinglenr
_shared_stater printspawnr
torchcuda is_available torch.distributed.launcher.apir/ r0 torch.multiprocessingr2 torch.multiprocessing.spawnr3 r RuntimeErrordictr r
r r+ r r r ) functionr+
num_processesr; use_portr9 r7 num_nodesrJ rI rdzv_confrdzv_idrL rM rN in_colab in_kagglexmplauncherr/ r0 r2 r3 problematic_importserrlib_name patched_envelaunch_config_kwargss r notebook_launcherr} ( s b HI
A
Arz/@/@
A
A
AAA O ck ! !!SY)?)K)K)M)M%N%NN
'(=(=(?(?@@
sT-A-G-G-I-I s s\i\n\p\p s s s
C I C BJNN:t$D$D$P????????? - . . 2 2 !
M#HuEEE
C CCCDDD (m& QQQQQ s lnnQ'!++ : " " $ $ 4 2 3 3 3 3 2 3 3 3$ u
! ! STT T 1 S S S S S S S S = = = = = = J J J J J J#122Q66 % #>"97CCIIKKvUU / k Z Z ZH d'r-^deeeee1
d
d
d qKqvVWyXX".#& !_ !_ !_# #
$% % #/#/[/[XY/[/["\"\bc c
d ,H{SSSD}DDDEEE j ($& #x//,5 &) , H/:,G,GX,G,GM+/"+"+'4 &&3%1%.%1)9%+ , , ,( (.^__ dKc,-GHdNN,,*N*N9M*N*N[cdddfjkkk- j j jG16RS9TT * [
! ! ++a^_+a+a b bhii j lS3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
j 3
jn !!
4
This function is provided for internal testing and debugging, but it's not intended for real trainings. It will
only use the CPU.
Args:
function (`Callable`):
The training function to execute.
args (`Tuple`):
Tuple of arguments to pass to the function (it will receive `*args`).
num_processes (`int`, *optional*, defaults to 2):
The number of processes to use for training.
r r1 r r r yes)r8 r9 r: accelerate_mixed_precisionaccelerate_debug_rdv_fileaccelerate_use_cpuT)debugr) r* N)ri r2 tempfileNamedTemporaryFiler
namer )rm r+ rn r2 tmp_filerv s r debug_launcherr
sX & 655555 $ & & \( $# '+&.m $
\ \ (===H OH4
TZ [ [ [ [ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \s4 A:!A" A:"A& &A:)A& *A::A>A>)r Nr r r r r r r Nr r r N)r r. )rS rV r re stater r utilsr r r r r
r r r
utils.constantsr r r} r r r r r s4
1 1 1 1 1 1 1 1 N M M M M M
! b b b b J!\ !\ !\ !\ !\ !\r