File size: 4,937 Bytes
7155cf2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 |
# Copyright 2024 Bytedance Ltd. and/or its affiliates
#
# 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.
"""
A unified tracking interface that supports logging data to different backend
"""
import os
from abc import ABC, abstractmethod
from typing import Any, Dict, List, Optional, Tuple, Union
from torch.utils.tensorboard import SummaryWriter
from ..py_functional import convert_dict_to_str, flatten_dict, is_package_available, unflatten_dict, filter_config_for_hparams
from .gen_logger import AggregateGenerationsLogger
if is_package_available("mlflow"):
import mlflow # type: ignore
if is_package_available("wandb"):
import wandb # type: ignore
if is_package_available("swanlab"):
import swanlab # type: ignore
class Logger(ABC):
@abstractmethod
def __init__(self, config: Dict[str, Any]) -> None: ...
@abstractmethod
def log(self, data: Dict[str, Any], step: int) -> None: ...
def finish(self) -> None:
pass
class ConsoleLogger(Logger):
def __init__(self, config: Dict[str, Any]) -> None:
print("Config\n" + convert_dict_to_str(config))
def log(self, data: Dict[str, Any], step: int) -> None:
print(f"Step {step}\n" + convert_dict_to_str(unflatten_dict(data)))
class MlflowLogger(Logger):
def __init__(self, config: Dict[str, Any]) -> None:
mlflow.start_run(run_name=config["trainer"]["experiment_name"])
mlflow.log_params(flatten_dict(config))
def log(self, data: Dict[str, Any], step: int) -> None:
mlflow.log_metrics(metrics=data, step=step)
class TensorBoardLogger(Logger):
def __init__(self, config: Dict[str, Any]) -> None:
tensorboard_dir = os.getenv("TENSORBOARD_DIR", "tensorboard_log")
os.makedirs(tensorboard_dir, exist_ok=True)
print(f"Saving tensorboard log to {tensorboard_dir}.")
self.writer = SummaryWriter(tensorboard_dir)
filtered_config = filter_config_for_hparams(config)
self.writer.add_hparams(flatten_dict(filtered_config), {})
def log(self, data: Dict[str, Any], step: int) -> None:
for key, value in data.items():
self.writer.add_scalar(key, value, step)
def finish(self):
self.writer.close()
class WandbLogger(Logger):
def __init__(self, config: Dict[str, Any]) -> None:
wandb.init(
project=config["trainer"]["project_name"],
name=config["trainer"]["experiment_name"],
config=config,
)
def log(self, data: Dict[str, Any], step: int) -> None:
wandb.log(data=data, step=step)
def finish(self) -> None:
wandb.finish()
class SwanlabLogger(Logger):
def __init__(self, config: Dict[str, Any]) -> None:
swanlab_key = os.getenv("SWANLAB_API_KEY")
swanlab_dir = os.getenv("SWANLAB_DIR", "swanlab_log")
swanlab_mode = os.getenv("SWANLAB_MODE", "cloud")
if swanlab_key:
swanlab.login(swanlab_key)
swanlab.init(
project=config["trainer"]["project_name"],
experiment_name=config["trainer"]["experiment_name"],
config={"UPPERFRAMEWORK": "EasyR1", "FRAMEWORK": "veRL", **config},
logdir=swanlab_dir,
mode=swanlab_mode,
)
def log(self, data: Dict[str, Any], step: int) -> None:
swanlab.log(data=data, step=step)
def finish(self) -> None:
swanlab.finish()
LOGGERS = {
"wandb": WandbLogger,
"mlflow": MlflowLogger,
"tensorboard": TensorBoardLogger,
"console": ConsoleLogger,
"swanlab": SwanlabLogger,
}
class Tracker:
def __init__(self, loggers: Union[str, List[str]] = "console", config: Optional[Dict[str, Any]] = None):
if isinstance(loggers, str):
loggers = [loggers]
self.loggers: List[Logger] = []
for logger in loggers:
if logger not in LOGGERS:
raise ValueError(f"{logger} is not supported.")
self.loggers.append(LOGGERS[logger](config))
self.gen_logger = AggregateGenerationsLogger(loggers)
def log(self, data: Dict[str, Any], step: int) -> None:
for logger in self.loggers:
logger.log(data=data, step=step)
def log_generation(self, samples: List[Tuple[str, str, str, float]], step: int) -> None:
self.gen_logger.log(samples, step)
def __del__(self):
for logger in self.loggers:
logger.finish()
|