Spaces:
Runtime error
Runtime error
File size: 7,617 Bytes
daeb223 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 |
import os
import gc
import time
import json
import math
import collections
from datetime import datetime
from typing import Optional, List, Dict, Tuple, Callable, Any, Union
import torch
import numpy as np
from transformers import (
is_datasets_available,
is_torch_tpu_available,
)
from transformers.trainer_utils import (
PredictionOutput,
EvalPrediction,
EvalLoopOutput,
denumpify_detensorize,
speed_metrics,
)
from transformers.utils import logging
from transformers.debug_utils import DebugOption
if is_datasets_available():
import datasets
if is_torch_tpu_available():
import torch_xla.core.xla_model as xm
import torch_xla.debug.metrics as met
from transformers import Trainer
logger = logging.get_logger(__name__)
class ToMixin:
def _optimizer_to(self, device: str = "cpu"):
# https://github.com/pytorch/pytorch/issues/8741
for param in self.optimizer.state.values():
# Not sure there are any global tensors in the state dict
if isinstance(param, torch.Tensor):
param.data = param.data.to(device)
if param._grad is not None:
param._grad.data = param._grad.data.to(device)
elif isinstance(param, dict):
for subparam in param.values():
if isinstance(subparam, torch.Tensor):
subparam.data = subparam.data.to(device)
if subparam._grad is not None:
subparam._grad.data = subparam._grad.data.to(
device)
def _scheduler_to(self, device: str = "cpu"):
# https://github.com/pytorch/pytorch/issues/8741
for param in self.lr_scheduler.__dict__.values():
if isinstance(param, torch.Tensor):
param.data = param.data.to(device)
if param._grad is not None:
param._grad.data = param._grad.data.to(device)
class BaseReader(Trainer, ToMixin):
name: str = None
def __init__(
self,
*args,
data_args = {},
eval_examples: datasets.Dataset = None,
**kwargs
):
super().__init__(*args, **kwargs)
self.data_args = data_args
self.eval_examples = eval_examples
def free_memory(self):
self.model.to("cpu")
self._optimizer_to("cpu")
self._scheduler_to("cpu")
torch.cuda.empty_cache()
gc.collect()
def postprocess(
self,
output: EvalLoopOutput,
) -> Union[Any, EvalPrediction]:
return output
def evaluate(
self,
eval_dataset: Optional[datasets.Dataset] = None,
eval_examples: Optional[datasets.Dataset] = None,
ignore_keys: Optional[List[str]] = None,
metric_key_prefix: str = "eval",
) -> Dict[str, float]:
# memory metrics - must set up as early as possible
self._memory_tracker.start()
eval_dataset = self.eval_dataset if eval_dataset is None else eval_dataset
eval_dataloader = self.get_eval_dataloader(eval_dataset)
start_time = time.time()
eval_examples = self.eval_examples if eval_examples is None else eval_examples
compute_metrics = self.compute_metrics
self.compute_metrics = None
eval_loop = self.prediction_loop if self.args.use_legacy_prediction_loop else self.evaluation_loop
try:
output = eval_loop(
eval_dataloader,
description="Evaluation",
prediction_loss_only=True if compute_metrics is None else None,
ignore_keys=ignore_keys,
metric_key_prefix=metric_key_prefix,
)
finally:
self.compute_metrics = compute_metrics
if isinstance(eval_dataset, datasets.Dataset):
eval_dataset.set_format(
type=eval_dataset.format["type"],
columns=list(eval_dataset.features.keys()),
)
eval_preds = self.postprocess(output, eval_examples, eval_dataset, mode="evaluate")
metrics = {}
if self.compute_metrics is not None:
metrics = self.compute_metrics(eval_preds)
# To be JSON-serializable, we need to remove numpy types or zero-d tensors
metrics = denumpify_detensorize(metrics)
# Prefix all keys with metric_key_prefix + '_'
for key in list(metrics.keys()):
if not key.startswith(f"{metric_key_prefix}_"):
metrics[f"{metric_key_prefix}_{key}"] = metrics.pop(key)
total_batch_size = self.args.eval_batch_size * self.args.world_size
metrics.update(
speed_metrics(
metric_key_prefix,
start_time,
num_samples=output.num_samples,
num_steps=math.ceil(output.num_samples / total_batch_size),
)
)
self.log(metrics)
# Log and save evaluation results
filename = "eval_results.txt"
eval_result_file = self.name + "_" + filename if self.name else filename
with open(os.path.join(self.args.output_dir, eval_result_file), "a") as writer:
logger.info("***** Eval results *****")
writer.write("***** Eval results *****\n")
writer.write(f"{datetime.now()}")
for key in sorted(metrics.keys()):
logger.info(" %s = %s", key, str(metrics[key]))
writer.write("%s = %s\n" % (key, str(metrics[key])))
writer.write("\n")
if DebugOption.TPU_METRICS_DEBUG in self.args.debug:
# tpu-comment: PyTorch/XLA에 대한 Logging debug metrics (compile, execute times, ops, etc.)
xm.master_print(met.metrics_report())
self.control = self.callback_handler.on_evaluate(
self.args, self.state, self.control, metrics
)
self._memory_tracker.stop_and_update_metrics(metrics)
return metrics
def predict(
self,
test_dataset: datasets.Dataset,
test_examples: datasets.Dataset,
ignore_keys: Optional[List[str]] = None,
metric_key_prefix: str = "test",
mode: bool = "predict",
) -> PredictionOutput:
# memory metrics - must set up as early as possible
self._memory_tracker.start()
test_dataloader = self.get_test_dataloader(test_dataset)
start_time = time.time()
compute_metrics = self.compute_metrics
self.compute_metrics = None
eval_loop = self.prediction_loop if self.args.use_legacy_prediction_loop else self.evaluation_loop
try:
output = eval_loop(
test_dataloader,
description="Prediction",
ignore_keys=ignore_keys,
metric_key_prefix=metric_key_prefix,
)
finally:
self.compute_metrics = compute_metrics
if isinstance(test_dataset, datasets.Dataset):
test_dataset.set_format(
type=test_dataset.format["type"],
columns=list(test_dataset.features.keys()),
)
predictions = self.postprocess(output, test_examples, test_dataset, mode=mode)
self._memory_tracker.stop_and_update_metrics(output.metrics)
return predictions |