← Search

Data · dataset · 2024

MiniMind-Dataset

Listed in Hugging Face Datasets

Description

📌 数据介绍 Ⅰ Tokenizer 分词器可以粗略理解成 LLM 使用的一本“词典”,负责把自然语言映射成 token id,再把 token id 解码回文本;项目中也提供了train_tokenizer.py作为词表训练示例。不建议重新训练 tokenizer,因为词表和切分规则一旦变化,模型权重、数据格式、推理接口与社区生态的兼容性都会下降,也会削弱模型的传播性。同时,tokenizer 还会影响 PPL 这类按 token 统计的指标,因此跨 tokenizer 比较时,BPB(Bits Per Byte)往往更有参考价值,可参考这篇。 对 MiniMind 这类小模型来说,词表大小还会直接影响 embedding 层和输出层的参数占比,因此保持词表精简通常是更合适的取舍。 Tokenizer介绍 第三方强大的开源模型例如 Yi、Qwen2、ChatGLM、Mistral、Llama 3 的 tokenizer 词表长度如下:… See the full description on the dataset page: huggingface.co/datasets/jingyaogong/minimind_dataset.

Links

Catalogue records · 1

Topics

Stated by source
text generation
Provenance · 1 source records, 9 field assertions
SourceKeyLast seenRaw
Hugging Face Datasetsjingyaogong/minimind_dataset15 d agoJSON v1
FieldAssertionExtractorEvidence
access_levelsource · Hugging Faceconnector:huggingface@1.0.0/gated
concepts[field].local:field:computer-science-aimapping · Hugging Faceconnector:huggingface@1.0.0
concepts[task].hf_task:text-generationsource · Hugging Faceconnector:huggingface@1.0.0/tags[task_categories:*]
created_datesource · Hugging Faceconnector:huggingface@1.0.0
descriptionsource · Hugging Faceconnector:huggingface@1.0.0/description
licensesource · Hugging Faceconnector:huggingface@1.0.0/tags[license:*]
publication_datesource · Hugging Faceconnector:huggingface@1.0.0
titlesource · Hugging Faceconnector:huggingface@1.0.0/id
updated_datesource · Hugging Faceconnector:huggingface@1.0.0