mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-05 05:02:50 +08:00
53 lines
1.7 KiB
Python
53 lines
1.7 KiB
Python
import os
|
|
import math
|
|
import json
|
|
import matplotlib.pyplot as plt
|
|
from typing import List, Optional
|
|
from transformers.trainer import TRAINER_STATE_NAME
|
|
|
|
from llmtuner.extras.logging import get_logger
|
|
|
|
|
|
logger = get_logger(__name__)
|
|
|
|
|
|
def smooth(scalars: List[float]) -> List[float]:
|
|
r"""
|
|
EMA implementation according to TensorBoard.
|
|
"""
|
|
last = scalars[0]
|
|
smoothed = list()
|
|
weight = 1.8 * (1 / (1 + math.exp(-0.05 * len(scalars))) - 0.5) # a sigmoid function
|
|
for next_val in scalars:
|
|
smoothed_val = last * weight + (1 - weight) * next_val
|
|
smoothed.append(smoothed_val)
|
|
last = smoothed_val
|
|
return smoothed
|
|
|
|
|
|
def plot_loss(save_dictionary: os.PathLike, keys: Optional[List[str]] = ["loss"]) -> None:
|
|
|
|
with open(os.path.join(save_dictionary, TRAINER_STATE_NAME), "r", encoding="utf-8") as f:
|
|
data = json.load(f)
|
|
|
|
for key in keys:
|
|
steps, metrics = [], []
|
|
for i in range(len(data["log_history"])):
|
|
if key in data["log_history"][i]:
|
|
steps.append(data["log_history"][i]["step"])
|
|
metrics.append(data["log_history"][i][key])
|
|
|
|
if len(metrics) == 0:
|
|
logger.warning(f"No metric {key} to plot.")
|
|
continue
|
|
|
|
plt.figure()
|
|
plt.plot(steps, metrics, alpha=0.4, label="original")
|
|
plt.plot(steps, smooth(metrics), label="smoothed")
|
|
plt.title("training {} of {}".format(key, save_dictionary))
|
|
plt.xlabel("step")
|
|
plt.ylabel(key)
|
|
plt.legend()
|
|
plt.savefig(os.path.join(save_dictionary, "training_{}.png".format(key)), format="png", dpi=100)
|
|
print("Figure saved:", os.path.join(save_dictionary, "training_{}.png".format(key)))
|