WeightSample
收藏资源简介:
该数据集包含模型的名称、训练和测试数据的大小、训练模型的超参数、性能指标以及内存分配和训练时间等信息。具体包括模型的训练规模、学习率、日志记录步数、优化器类型、训练轮数、输出目录、报告目标、保存策略、保存限制、随机种子、预热步数、权重衰减等参数,以及模型的准确率、F1宏观分数、F1加权分数、精确率、召回率等性能指标。同时,数据集分为训练集,提供了训练集的字节数和示例数。数据集的总下载大小和实际大小也一并提供。
This dataset contains information such as the model name, the sizes of training and test datasets, hyperparameters for model training, performance metrics, memory allocation and training duration. Specifically, it includes parameters like the model's training scale, learning rate, logging steps, optimizer type, training epochs, output directory, reporting target, saving strategy, saving limit, random seed, warm-up steps and weight decay, as well as performance metrics such as model accuracy, macro-F1 score, weighted-F1 score, precision and recall. Meanwhile, the dataset is divided into training subsets, with the byte size and sample count of the training set provided. The total download size and actual storage size of the dataset are also included.
WeightSample数据集概述
数据集基本信息
- 数据集名称:WeightSample
- 存储位置:https://huggingface.co/datasets/ngtranai09/WeightSample
- 数据量:3,490字节
- 下载大小:12,999字节
- 样本数量:16个训练样本
数据结构特征
主要字段
- Model_name:模型名称(字符串类型)
- Train_size:训练集大小(整数类型)
- Test_size:测试集大小(整数类型)
- Parameters:参数数量(整数类型)
- Trainable_parameters:可训练参数数量(整数类型)
- Memory Allocation:内存分配(字符串类型)
- Training Time:训练时间(字符串类型)
- lora:空值类型字段
- r:空值类型字段
训练参数结构(arg字段)
- auto_find_batch_size:自动寻找批次大小(布尔类型)
- gradient_accumulation_steps:梯度累积步数(整数类型)
- learning_rate:学习率(浮点类型)
- logging_steps:日志记录步数(整数类型)
- lr_scheduler_type:学习率调度器类型(字符串类型)
- num_train_epochs:训练轮数(整数类型)
- output_dir:输出目录(字符串类型)
- report_to:报告目标(字符串类型)
- save_strategy:保存策略(字符串类型)
- save_total_limit:保存总数限制(整数类型)
- seed:随机种子(整数类型)
- warmup_steps:预热步数(整数类型)
- weight_decay:权重衰减(浮点类型)
性能指标结构(Performance字段)
- accuracy:准确率(浮点类型)
- f1_macro:宏平均F1分数(浮点类型)
- f1_weighted:加权F1分数(浮点类型)
- precision:精确率(浮点类型)
- recall:召回率(浮点类型)
数据配置
- 配置名称:default
- 数据分割:train
- 文件路径:data/train-*




