shs / README.md
Shrijanagain's picture
Upload README.md with huggingface_hub
cfdf5b5 verified
|
Raw History Blame Contribute Delete
2.45 kB
metadata
pretty_name: Speed-5B-mX-parquet
task_categories:
  - text-generation
language:
  - en

Speed-5B-mX (compressed)

Zstd-compressed Parquet copy of Shrijanagain/Speed-5B-mX.

Totals

  • Rows: 3,036,552
  • Exact tokens (add_special_tokens=False): 2,789,251,861
  • Source JSONL bytes: 10.98 GiB
  • Parquet bytes: 3.31 GiB
  • Compression ratio: 3.315x
  • Tokenizer: Shrijanagain/SPEED2-V0-0B7-CPT-SFT/tokenizer.json

Schema

column type notes
text large_string document text
source string mixture source id
tokens int32 exact token count of text

Tokens per source

source rows tokens share
code_stack 599,671 828,986,469 29.72%
fineweb_edu 726,000 711,089,153 25.49%
fineweb 1,048,581 687,823,165 24.66%
finemath 167,228 259,464,482 9.30%
wikipedia 156,204 153,645,247 5.51%
cybersecurity 338,868 148,243,345 5.31%

Shards

source shard rows tokens parquet size
chunks_6b/mixed_6b_chunk_000000.jsonl 222,987 147,439,498 0.236 GiB
chunks_6b/mixed_6b_chunk_000001.jsonl 224,560 147,261,058 0.236 GiB
chunks_6b/mixed_6b_chunk_000002.jsonl 225,683 147,189,388 0.236 GiB
chunks_6b/mixed_6b_chunk_000003.jsonl 224,491 147,230,459 0.236 GiB
chunks_6b/mixed_6b_chunk_000004.jsonl 226,860 204,478,767 0.227 GiB
chunks_6b/mixed_6b_chunk_000005.jsonl 234,909 323,119,815 0.208 GiB
chunks_6b/mixed_6b_chunk_000006.jsonl 232,554 320,228,994 0.208 GiB
chunks_6b/mixed_6b_chunk_000007.jsonl 168,437 190,180,178 0.224 GiB
chunks_6b/mixed_6b_chunk_000008.jsonl 150,933 145,937,878 0.230 GiB
chunks_6b/mixed_6b_chunk_000009.jsonl 149,047 146,289,311 0.230 GiB
chunks_6b/mixed_6b_chunk_000010.jsonl 148,761 146,330,597 0.230 GiB
chunks_6b/mixed_6b_chunk_000011.jsonl 148,315 146,337,489 0.230 GiB
chunks_6b/mixed_6b_chunk_000012.jsonl 129,022 190,368,485 0.206 GiB
chunks_6b/mixed_6b_chunk_000013.jsonl 139,928 173,819,093 0.219 GiB
chunks_6b/mixed_6b_chunk_000014.jsonl 410,065 213,040,851 0.156 GiB

Audit

  • Record fields seen: {"source,text": 3036553}
  • Skipped rows: {"too_long": 1}

Use streaming=True or read row groups directly; each file is independently readable.