zchen/b2txt25
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Enhance dataset caching logic for training and validation sets with improved messaging

Browse Source
This commit is contained in:
Zchen
2025-10-19 10:31:31 +08:00
parent 558be0ad98
commit cfd9653da9
1 changed files with 15 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
model_training_nnn_tpu/dataset_tf.py
View File

@@ -778,20 +778,21 @@ def create_input_fn(dataset_tf: BrainToTextDatasetTF,
dataset = dataset_tf.create_individual_dataset() dataset = dataset_tf.create_individual_dataset()
# ========================= I/O OPTIMIZATION SOLUTION ========================= # ========================= I/O OPTIMIZATION SOLUTION =========================
# 在数据加载之后、随机操作（如数据增强）之前进行缓存 # 对训练集和验证集都进行缓存，因为：
if training: # 1. 训练集：每个epoch都要完整遍历
# 对于训练，缓存到磁盘文件或内存 # 2. 验证集：每200轮验证一次 + 早停检查，会被频繁使用
if cache_path: if cache_path:
dataset = dataset.cache(cache_path) dataset = dataset.cache(cache_path)
print(f"🗃️ Dataset caching enabled: {cache_path}") split_name = "training" if training else "validation"
print("⚠️ First epoch will be slow while building cache, subsequent epochs will be much faster") print(f"🗃️ {split_name.capitalize()} dataset caching enabled: {cache_path}")
print(f"⚠️ First access will be slow while building {split_name} cache, subsequent access will be much faster")
else: else:
# 如果内存足够大，可以缓存到内存，速度更快 # 如果没有指定缓存路径，默认使用内存缓存
# 但对于大型数据集，推荐使用文件缓存 # 对于大型数据集，建议在调用时显式指定磁盘缓存路径
dataset = dataset.cache() dataset = dataset.cache()
print("🗃️ Dataset caching enabled: in-memory cache") split_name = "training" if training else "validation"
print("⚠️ First epoch will be slow while building cache, subsequent epochs will be much faster") print(f"🗃️ {split_name.capitalize()} dataset caching enabled: in-memory cache")
# (对于验证集通常不需要缓存，因为它只运行一次) print(f"⚠️ First access will be slow while building {split_name} cache, subsequent access will be much faster")
# ================================================================ # ================================================================
def apply_transforms(example): def apply_transforms(example):