Training Corpus
A large-scale collection of text or data used to train machine learning models. Its scale, quality, and diversity directly affect model capability.
Timeline (last 90 days)
在训练语料中，粗体和斜体多用于标题、强调或排版，而非语言学意义上的焦点标注，导致模型的生成分布被系统性偏移。
High-quality long-form text content (such as academic books) is crucial for improving the reasoning and knowledge capabilities of large language models.
In LLM projects, data simultaneously plays three roles: training corpus, retrieval knowledge base, and runtime context, and the three have inconsistent requirements for data.
If the training corpus and the retrieval knowledge base share the same set of raw documents, branching logic should be designed after the cleaning layer—the same source data produces instruction samples and retrieval chunks respectively, rather than each maintaining its own independent pipeline.
Indian languages account for a far smaller share of the training corpora of large language models than English, which leads to relatively weak safety guardrails for the model on languages such as Hindi.
Formally published books have been edited, proofread, and typeset, making them linguistically standardized and knowledge-dense, representing high-quality sources among training corpora.
All Facts (6)
在训练语料中，粗体和斜体多用于标题、强调或排版，而非语言学意义上的焦点标注，导致模型的生成分布被系统性偏移。
50%UnverifiedHigh-quality long-form text content (such as academic books) is crucial for improving the reasoning and knowledge capabilities of large language models.
50%UnverifiedIn LLM projects, data simultaneously plays three roles: training corpus, retrieval knowledge base, and runtime context, and the three have inconsistent requirements for data.
50%UnverifiedIf the training corpus and the retrieval knowledge base share the same set of raw documents, branching logic should be designed after the cleaning layer—the same source data produces instruction samples and retrieval chunks respectively, rather than each maintaining its own independent pipeline.
50%UnverifiedIndian languages account for a far smaller share of the training corpora of large language models than English, which leads to relatively weak safety guardrails for the model on languages such as Hindi.
50%UnverifiedFormally published books have been edited, proofread, and typeset, making them linguistically standardized and knowledge-dense, representing high-quality sources among training corpora.
50%