Hugging Face's logo

Datasets:
ByteSpanTokenisers
/
finewebedu-20B

Modalities:
Text
Formats:
parquet
Languages:
English
Size:
100M - 1B
Tags:
language modeling
Libraries:
Datasets
Dask
Dataset card Data Studio Files
xet
Community
1
codebyzeb commited on
Commit
2fa887c
·
1 Parent(s): 249da52

Update README

Browse files
Files changed (1) hide show
  1. README.md +28 -20
README.md CHANGED
@@ -23,38 +23,30 @@ configs:
23
  data_files:
24
  - split: train
25
  path: bytelevel-subset_2/train-*
26
- - config_name: frequency_64000
27
  data_files:
28
  - split: train
29
- path: frequency_64000/*.parquet
30
- - config_name: fw57M_Entropy_thresholdB_32000
31
  data_files:
32
  - split: train
33
- path: fw57M_Entropy_thresholdB_32000/*.parquet
34
- - config_name: fw57M_Surprisal_thresholdB_32000
35
  data_files:
36
  - split: train
37
- path: fw57M_Surprisal_thresholdB_32000/*.parquet
38
- - config_name: fw57M_Surprisal_bytespanP1-0T30_64000
39
  data_files:
40
  - split: train
41
- path: fw57M_Surprisal_bytespanP1-0T30_64000/*.parquet
42
- - config_name: fw57M_Surprisal_bytespanP1-0_64000
43
  data_files:
44
  - split: train
45
- path: fw57M_Surprisal_bytespanP1-0_64000/*.parquet
46
- - config_name: fw57M_Surprisal_bytespanP0-5_64000
47
  data_files:
48
  - split: train
49
- path: fw57M_Surprisal_bytespanP0-5_64000/*.parquet
50
- - config_name: fw57M_Surprisal_bytespanP0-5T30_64000
51
- data_files:
52
- - split: train
53
- path: fw57M_Surprisal_bytespanP0-5T30_64000/*.parquet
54
- - config_name: fw57M_Entropy_bytespanP0-0_64000
55
- data_files:
56
- - split: train
57
- path: fw57M_Entropy_bytespanP0-0_64000/*.parquet
58
  language:
59
  - en
60
  tags:
@@ -82,3 +74,19 @@ dataset_info:
82
  ---
83
 
84
  # FineWebEDU 20B
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  data_files:
24
  - split: train
25
  path: bytelevel-subset_2/train-*
26
+ - config_name: BPE_64000
27
  data_files:
28
  - split: train
29
+ path: BPE_64000/*.parquet
30
+ - config_name: ByteSpanSurprisalMonotonicSeeding_64000
31
  data_files:
32
  - split: train
33
+ path: ByteSpanSurprisalMonotonicSeeding_64000/*.parquet
34
+ - config_name: ByteSpanSurprisalMonotonicFrequency_64000
35
  data_files:
36
  - split: train
37
+ path: ByteSpanSurprisalMonotonicFrequency_64000/*.parquet
38
+ - config_name: ByteSpanSurprisalMonotonicSeeding_64000
39
  data_files:
40
  - split: train
41
+ path: ByteSpanSurprisalMonotonicSeeding_64000_64000/*.parquet
42
+ - config_name: ByteSpanSurprisalCombinedSeeding_64000_64000
43
  data_files:
44
  - split: train
45
+ path: ByteSpanSurprisalCombinedSeeding_64000_64000_64000/*.parquet
46
+ - config_name: BPEWP_64000
47
  data_files:
48
  - split: train
49
+ path: BPEWP_64000/*.parquet
 
 
 
 
 
 
 
 
50
  language:
51
  - en
52
  tags:
 
74
  ---
75
 
76
  # FineWebEDU 20B
77
+
78
+ A copy of [FineWebEDU-20B](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) used for out tokenizer experiments. The subsets are as follows:
79
+ - `bytelevel`: the full dataset tokenized using our bytelevel tokenizer
80
+ - `bytelevel-subset_1`: a 100k-row subset of the bytelevel subset, used to train bytelevel models.
81
+ - `bytelevel-subset_2`: a 100k-row subset of the bytelevel subset, used to extract llm predictions.
82
+ - `bytelevel-llm-data`: a copy of `bytelevel-subset_2` with lm predictions, used to train bytespan tokenizers
83
+ - `bytelevel-subset_3`: a 100k-row subset of the bytelevel subset, used to evaluate trained tokenizers
84
+
85
+ The remaining subsets are all versions of the dataset tokenized with our trained tokenizers:
86
+ - `BPE_64000`
87
+ - `BPEWP_64000`
88
+ - `ByteSpanSurprisalMonotonicFrequency_64000`
89
+ - `ByteSpanSurprisalMonotonicSeeding_64000`
90
+ - `ByteSpanSurprisalCombinedFrequency_64000`
91
+ - `ByteSpanSurprisalCombinedSeeding_64000`
92
+