Datasets:
ArXiv:
License:
Commit
•
c6cd9fc
1
Parent(s):
1fbd7c4
Define features explicitly in the dataset card
Browse filesDefine features explicitly in the dataset card, so they are not inferred.
Note that the features are inferred from the first samples. Therefore, if subsequent samples have features different from the inferred ones, an error is raised.
Fix #8.
README.md
CHANGED
@@ -1,5 +1,51 @@
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
4 |
## DCLM-baseline
|
5 |
DCLM-baseline is a 4T token / 3B document pretraining dataset that achieves strong performance on language model benchmarks.
|
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
3 |
+
dataset_info:
|
4 |
+
features:
|
5 |
+
- name: bff_contained_ngram_count_before_dedupe
|
6 |
+
dtype: int64
|
7 |
+
- name: language_id_whole_page_fasttext
|
8 |
+
struct:
|
9 |
+
- name: en
|
10 |
+
dtype: float64
|
11 |
+
- name: metadata
|
12 |
+
struct:
|
13 |
+
- name: Content-Length
|
14 |
+
dtype: string
|
15 |
+
- name: Content-Type
|
16 |
+
dtype: string
|
17 |
+
- name: WARC-Block-Digest
|
18 |
+
dtype: string
|
19 |
+
- name: WARC-Concurrent-To
|
20 |
+
dtype: string
|
21 |
+
- name: WARC-Date
|
22 |
+
dtype: timestamp[s]
|
23 |
+
- name: WARC-IP-Address
|
24 |
+
dtype: string
|
25 |
+
- name: WARC-Identified-Payload-Type
|
26 |
+
dtype: string
|
27 |
+
- name: WARC-Payload-Digest
|
28 |
+
dtype: string
|
29 |
+
- name: WARC-Record-ID
|
30 |
+
dtype: string
|
31 |
+
- name: WARC-Target-URI
|
32 |
+
dtype: string
|
33 |
+
- name: WARC-Type
|
34 |
+
dtype: string
|
35 |
+
- name: WARC-Warcinfo-ID
|
36 |
+
dtype: string
|
37 |
+
- name: WARC-Truncated
|
38 |
+
dtype: string
|
39 |
+
- name: previous_word_count
|
40 |
+
dtype: int64
|
41 |
+
- name: text
|
42 |
+
dtype: string
|
43 |
+
- name: url
|
44 |
+
dtype: string
|
45 |
+
- name: warcinfo
|
46 |
+
dtype: string
|
47 |
+
- name: fasttext_openhermes_reddit_eli5_vs_rw_v2_bigram_200k_train_prob
|
48 |
+
dtype: float64
|
49 |
---
|
50 |
## DCLM-baseline
|
51 |
DCLM-baseline is a 4T token / 3B document pretraining dataset that achieves strong performance on language model benchmarks.
|