Source code for cerebras.modelzoo.trainer.extensions.wandb.wandb

# Copyright 2022 Cerebras Systems.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Contains the WandbLogger class for logging metrics to Weights and Biases."""

from typing import List, Literal, Optional
from warnings import warn

import torch

from cerebras.modelzoo.trainer.loggers import Logger


[docs]class WandbLogger(Logger): """ Logger class for logging metrics to Weights and Biases. """ def __init__( self, project: Optional[str] = None, group: Optional[str] = None, run_id: Optional[str] = None, run_name: Optional[str] = None, job_type: Optional[str] = None, tags: Optional[List[str]] = None, resume: Literal["never", "allow", "auto", "must"] = "auto", entity: Optional[str] = None, ): """ Args: project: The name of the project to which the run belongs. group: The name of the group to which the run belongs. run_id: The unique identifier for the run. run_name: The name of the run. job_type: The type of job. tags: List of tags to be associated with the run. resume: Resume mode for the run. It can be one of the following: - "never": Do not resume the run. - "allow": Allow the run to resume if a previous run exists. - "auto": Automatically resume the run if a previous run exists. - "must": Resume the run if a previous run exists. entity: An entity is a username or team name where you're sending runs. This entity must exist before you can send runs there, so make sure to create your account or team in the UI before starting to log runs. """ self.project = project self.group = group self.run_id = run_id self.run_name = run_name self.job_type = job_type self.tags = tags self.resume = resume self.entity = entity def pre_setup(self, trainer): # pylint: disable=no-self-use try: # pylint: disable=unused-import import wandb # noqa except ImportError: raise RuntimeError( "wandb is an optional dependency of modelzoo. " "In order to use it, 'pip install wandb==0.16.2' into this venv" ) def finalize(self): try: import wandb if wandb.run is not None: wandb.run.finish() finally: pass
[docs] def check_presence_of_wandb_dir( # pylint: disable=no-self-use self, rundir ): """Check if the wandb directory is present in the run directory. Args: rundir: The directory where the run is being stored. """ # Ensure that the wandb directory is not already present or empty wandb_dir = rundir / "wandb" if wandb_dir.exists(): # Ensure there are no run-* folders in the wandb directory. if any( dir.is_dir() and dir.name.startswith('run-') for dir in wandb_dir.iterdir() ): raise FileExistsError( f"A previous run seems to already exist in {wandb_dir}. " "Please specify a different 'model_dir'." )
def setup(self, trainer): import wandb from wandb.sdk.lib import RunDisabled from wandb.wandb_run import Run rundir = trainer.model_dir previous_run_id = None run_files = list((rundir / "wandb").glob("run-*")) if run_files: previous_run_id = str(run_files[0]).split('-')[-1] if self.resume == "never": if ( self.run_id is not None and previous_run_id is not None and self.run_id == previous_run_id ): raise ValueError( f"The specified run_id ({self.run_id}) matches with a " f"previous_run_id ({previous_run_id}) " "but 'never' mode requires them to be different." ) self.check_presence_of_wandb_dir(rundir) elif self.resume in ["allow", "auto"]: if self.run_id is not None and previous_run_id is not None: if self.run_id == previous_run_id: # Log into this previous run as it's the same run pass else: # Raise an error if a wandb run already exists inside the specified run dir. self.check_presence_of_wandb_dir(rundir) elif previous_run_id is not None: # No new run ID provided, so default to the previous run ID self.run_id = previous_run_id elif self.resume == "must" and previous_run_id: if self.run_id is not None and self.run_id != previous_run_id: raise ValueError( f"The specified run_id ({self.run_id}) does not match " f"previous_run_id ({previous_run_id}) " "but resume mode 'must' requires them to be the same." ) self.run_id = previous_run_id if wandb.run is None: # pylint: disable=all self.run = wandb.init( dir=rundir, job_type=self.job_type, # config=params, project=self.project, group=self.group, tags=self.tags, name=self.run_name, id=self.run_id, resume=self.resume, entity=self.entity, ) # define default x-axis if isinstance(self.run, (Run, RunDisabled)) and getattr( self.run, "define_metric", None ): self.run.define_metric("global_step") self.run.define_metric( "*", step_metric="global_step", step_sync=True ) else: self.run = wandb.run def log_metrics(self, metrics, step): m = {"global_step": step} summary = {} for name, value in metrics.items(): if isinstance(value, torch.Tensor): if value.numel() == 1: m[name] = value.item() else: warn( "Attempting to log a non-scalar tensor for {name}. " "WandB Logger does not support logging non-scalar tensors." ) elif isinstance(value, (int, float)): m[name] = value elif isinstance(value, str): summary[name] = value else: try: import pandas as pd import wandb if isinstance(value, pd.DataFrame): m[name] = wandb.Table(dataframe=value) continue except ImportError: pass warn( f"Attempting to log a {type(value)} for {name}. " f"WandB Logger does not support logging {type(value)}" ) self.run.log(m) self.run.summary.update(summary)