Laz4rz's picture
Update README.md
329a380 verified
metadata
language:
  - en
pretty_name: ScienceWikiSmallChunk256
tags:
  - RAG
  - Retrieval Augmented Generation
  - Small Chunks
  - Wikipedia
  - Science
  - Scientific
  - Scientific Wikipedia
  - Science Wikipedia
  - 256 tokens
license: cc-by-sa-3.0
task_categories:
  - text-generation
  - text-classification
  - question-answering

ScienceWikiSmallChunk

Processed version of millawell/wikipedia_field_of_science, prepared to be used in small context length RAG systems. Chunk length is tokenizer dependent, but each chunk should be around 256 tokens. Longer wikipedia pages have been split into smaller entries, with title added as a prefix.

There is also 512 tokens dataset available: Laz4rz/wikipedia_science_chunked_small_rag_512

If you wish to prepare some other chunk length:

  • use millawell/wikipedia_field_of_science
  • adapt chunker function:
def chunker_clean(results, example, length=512, approx_token=3, prefix=""):
    if len(results) == 0:
        regex_pattern = r'[\n\s]*\n[\n\s]*'
        example = re.sub(regex_pattern, " ", example).strip().replace(prefix, "")
    chunk_length = length * approx_token
    if len(example) > chunk_length:
        first = example[:chunk_length]
        chunk = ".".join(first.split(".")[:-1])
        if len(chunk) == 0:
            chunk = first
        rest = example[len(chunk)+1:]
        results.append(prefix+chunk.strip())
        if len(rest) > chunk_length:
            chunker_clean(results, rest.strip(), length=length, approx_token=approx_token, prefix=prefix)
        else:
            results.append(prefix+rest.strip())
    else:
        results.append(prefix+example.strip())
    return results