Hugging Face's logo

Datasets:
sxiong
/
DHSA_Long-Data-Collections

Modalities:
Text
Formats:
json
Size:
1M - 10M
ArXiv:
Libraries:
Datasets
Dask
License:
Dataset card Data Studio Files
xet
 Community
sxiong commited on
Commit
c62a837
·
verified ·
1 Parent(s): b5b40bc

Add files using upload-large-folder tool

Browse files
Files changed (44) hide show
  1. README.md +143 -0
  2. fine-tune/16k_32k/booksum.jsonl.zst +3 -0
  3. fine-tune/16k_32k/natural_questions_10_200_docs.jsonl.zst +3 -0
  4. fine-tune/32k_64k/booksum.jsonl.zst +3 -0
  5. fine-tune/32k_64k/natural_questions_10_200_docs.jsonl.zst +3 -0
  6. fine-tune/64k_128k/booksum.jsonl.zst +3 -0
  7. fine-tune/8k_16k/booksum.jsonl.zst +3 -0
  8. fine-tune/8k_16k/natural_questions_10_200_docs.jsonl.zst +3 -0
  9. fine-tune/gt_128k/booksum.jsonl.zst +3 -0
  10. fine-tune/lt_8k/booksum.jsonl.zst +3 -0
  11. fine-tune/lt_8k/natural_questions_10_200_docs.jsonl.zst +3 -0
  12. length_bucket_summary.json +103 -0
  13. pretrain/16k_32k/NI_decontaminated_materialized.jsonl.zst +3 -0
  14. pretrain/16k_32k/P3_decontaminated_materialized.jsonl.zst +3 -0
  15. pretrain/16k_32k/arxiv_doc_to_abs.jsonl.zst +3 -0
  16. pretrain/16k_32k/pile_sub.jsonl.zst +3 -0
  17. pretrain/16k_32k/rp_sub.jsonl.zst +3 -0
  18. pretrain/16k_32k/ul2_plus_oscar_en.jsonl.zst +3 -0
  19. pretrain/32k_64k/NI_decontaminated_materialized.jsonl.zst +3 -0
  20. pretrain/32k_64k/arxiv_doc_to_abs.jsonl.zst +3 -0
  21. pretrain/32k_64k/pile_sub.jsonl.zst +3 -0
  22. pretrain/32k_64k/rp_sub.jsonl.zst +3 -0
  23. pretrain/32k_64k/ul2_plus_oscar_en.jsonl.zst +3 -0
  24. pretrain/64k_128k/NI_decontaminated_materialized.jsonl.zst +3 -0
  25. pretrain/64k_128k/arxiv_doc_to_abs.jsonl.zst +3 -0
  26. pretrain/64k_128k/pile_sub.jsonl.zst +3 -0
  27. pretrain/64k_128k/rp_sub.jsonl.zst +3 -0
  28. pretrain/64k_128k/ul2_plus_oscar_en.jsonl.zst +3 -0
  29. pretrain/8k_16k/NI_decontaminated_materialized.jsonl.zst +3 -0
  30. pretrain/8k_16k/P3_decontaminated_materialized.jsonl.zst +3 -0
  31. pretrain/8k_16k/arxiv_doc_to_abs.jsonl.zst +3 -0
  32. pretrain/8k_16k/pile_sub.jsonl.zst +3 -0
  33. pretrain/8k_16k/rp_sub.jsonl.zst +3 -0
  34. pretrain/8k_16k/ul2_plus_oscar_en.jsonl.zst +3 -0
  35. pretrain/gt_128k/arxiv_doc_to_abs.jsonl.zst +3 -0
  36. pretrain/gt_128k/pile_sub.jsonl.zst +3 -0
  37. pretrain/gt_128k/rp_sub.jsonl.zst +3 -0
  38. pretrain/gt_128k/ul2_plus_oscar_en.jsonl.zst +3 -0
  39. pretrain/lt_8k/NI_decontaminated_materialized.jsonl.zst +3 -0
  40. pretrain/lt_8k/P3_decontaminated_materialized.jsonl.zst +3 -0
  41. pretrain/lt_8k/arxiv_doc_to_abs.jsonl.zst +3 -0
  42. pretrain/lt_8k/pile_sub.jsonl.zst +3 -0
  43. pretrain/lt_8k/rp_sub.jsonl.zst +3 -0
  44. pretrain/lt_8k/ul2_plus_oscar_en.jsonl.zst +3 -0
README.md ADDED
@@ -0,0 +1,143 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ ---
4
+
5
+ # DHSA_Long-Data-Collections
6
+
7
+ A length-bucketed release of [togethercomputer/Long-Data-Collections](https://huggingface.co/datasets/togethercomputer/Long-Data-Collections), used in *[Long-Context Modeling with Dynamic Hierarchical Sparse Attention for Memory-Constrained LLM Inference](https://arxiv.org/pdf/2510.24606)* (ICML 2026 Spotlight).
8
+
9
+ Each example is assigned to exactly one length bucket based on token count with `meta-llama/Llama-3.1-8B-Instruct` (`add_special_tokens=True`):
10
+
11
+ | Bucket | Token range |
12
+ |--------|-------------|
13
+ | `lt_8k` | [0, 8K) |
14
+ | `8k_16k` | [8K, 16K) |
15
+ | `16k_32k` | [16K, 32K) |
16
+ | `32k_64k` | [32K, 64K) |
17
+ | `64k_128k` | [64K, 128K) |
18
+ | `gt_128k` | ≥ 128K |
19
+
20
+ - **Pretrain** examples are bucketed by the `text` field.
21
+ - **Fine-tune** examples are bucketed by the `prompt` field.
22
+
23
+ ## Directory layout
24
+
25
+ ```
26
+ Long-Data-Collections/
27
+ ├── README.md
28
+ ├── length_bucket_summary.json
29
+ ├── pretrain/
30
+ │ ├── lt_8k/
31
+ │ │ ├── arxiv_doc_to_abs.jsonl.zst
32
+ │ │ ├── NI_decontaminated_materialized.jsonl.zst
33
+ │ │ ├── P3_decontaminated_materialized.jsonl.zst
34
+ │ │ ├── pile_sub.jsonl.zst
35
+ │ │ ├── rp_sub.jsonl.zst
36
+ │ │ └── ul2_plus_oscar_en.jsonl.zst
37
+ │ ├── 8k_16k/
38
+ │ ├── 16k_32k/
39
+ │ ├── 32k_64k/
40
+ │ ├── 64k_128k/
41
+ │ └── gt_128k/
42
+ └── fine-tune/
43
+ ├── lt_8k/
44
+ │ ├── booksum.jsonl.zst
45
+ │ └── natural_questions_10_200_docs.jsonl.zst
46
+ ├── 8k_16k/
47
+ ├── 16k_32k/
48
+ ├── 32k_64k/
49
+ ├── 64k_128k/
50
+ └── gt_128k/
51
+ ```
52
+
53
+ All files are zstd-compressed JSONL (`.jsonl.zst`). Each line is a JSON object preserving the original fields (`text`, and optionally `meta` / `metadata` for pretrain; `text`, `prompt`, `completion` for fine-tune).
54
+
55
+ ## Example counts by bucket
56
+
57
+ ### Pretrain (bucketed by `text` length)
58
+
59
+ | Dataset | lt_8k | 8k_16k | 16k_32k | 32k_64k | 64k_128k | gt_128k | **Total** |
60
+ |---------|------:|-------:|--------:|--------:|---------:|--------:|----------:|
61
+ | arxiv_doc_to_abs | 377,854 | 559,745 | 436,259 | 150,632 | 28,557 | 5,258 | 1,558,305 |
62
+ | rp_sub | 913,818 | 11,265 | 3,893 | 1,019 | 368 | 151 | 930,514 |
63
+ | ul2_plus_oscar_en | 3,148,215 | 342,739 | 60,306 | 9,213 | 1,804 | 236 | 3,562,513 |
64
+ | pile_sub | 1,878,091 | 43,443 | 12,818 | 3,022 | 1,799 | 1,285 | 1,940,458 |
65
+ | NI_decontaminated_materialized | 460,066 | 117,188 | 567 | 210 | 12 | 0 | 578,043 |
66
+ | P3_decontaminated_materialized | 757,542 | 56,465 | 8 | 0 | 0 | 0 | 814,015 |
67
+ | **All pretrain** | **7,535,586** | **1,130,845** | **513,851** | **164,096** | **32,540** | **6,930** | **9,383,848** |
68
+
69
+ ### Fine-tune (bucketed by `prompt` length)
70
+
71
+ | Dataset | lt_8k | 8k_16k | 16k_32k | 32k_64k | 64k_128k | gt_128k | **Total** |
72
+ |---------|------:|-------:|--------:|--------:|---------:|--------:|----------:|
73
+ | booksum | 7,909 | 1,207 | 413 | 64 | 5 | 2 | 9,600 |
74
+ | natural_questions_10_200_docs | 21,370 | 25,879 | 41,639 | 69 | 0 | 0 | 88,957 |
75
+ | **All fine-tune** | **29,279** | **27,086** | **42,052** | **133** | **5** | **2** | **98,557** |
76
+
77
+ Full counts are also available in `length_bucket_summary.json`.
78
+
79
+ ## Dataset description
80
+
81
+ This collection compiles long-context datasets for training and evaluating models that require extensive comprehension over large text inputs. It is derived from the public Together AI release and reorganized into length buckets for long-context experiments.
82
+
83
+ ### Pretrain data (`pretrain/`)
84
+
85
+ | File | Description |
86
+ |------|-------------|
87
+ | `rp_sub` | RedPajama book subset — diverse literary text |
88
+ | `arxiv_doc_to_abs` | RedPajama ArXiv papers with abstract appended after the paper body |
89
+ | `ul2_plus_oscar_en` | UL2-style fill-in-the-blank completions (LAION Open-Instruction-Generalist) |
90
+ | `pile_sub` | Subsample of The Pile |
91
+ | `NI_decontaminated_materialized` | Natural Instructions, decontaminated against HELM core scenarios |
92
+ | `P3_decontaminated_materialized` | Public Pool of Prompts (P3), decontaminated against HELM core scenarios |
93
+
94
+ ### Fine-tune data (`fine-tune/`)
95
+
96
+ | File | Description |
97
+ |------|-------------|
98
+ | `natural_questions_10_200_docs` | Multi-passage QA from Natural Questions (10–200 Wiki passages per question) |
99
+ | `booksum` | Long-context book summarization |
100
+
101
+ Fine-tune records contain `prompt` (context + instruction), `completion` (target answer/summary), and `text` (`prompt` + `completion`).
102
+
103
+ ## Usage
104
+
105
+ Load a specific bucket and file with the Hugging Face `datasets` library:
106
+
107
+ ```python
108
+ from datasets import load_dataset
109
+
110
+ ds = load_dataset(
111
+ "json",
112
+ data_files="/path/to/Long-Data-Collections/pretrain/16k_32k/arxiv_doc_to_abs.jsonl.zst",
113
+ split="train",
114
+ )
115
+ ```
116
+
117
+ Or stream locally with zstd:
118
+
119
+ ```python
120
+ import json, zstandard as zstd
121
+
122
+ path = "pretrain/8k_16k/rp_sub.jsonl.zst"
123
+ with open(path, "rb") as f:
124
+ with zstd.ZstdDecompressor().stream_reader(f) as reader:
125
+ for line in reader:
126
+ row = json.loads(line)
127
+ ...
128
+ ```
129
+
130
+ ## Licensing
131
+
132
+ Please refer to the original sources of each sub-dataset for their respective licenses. This dataset is a length-bucketed release of [togethercomputer/Long-Data-Collections](https://huggingface.co/datasets/togethercomputer/Long-Data-Collections); upstream licensing terms apply.
133
+
134
+ ## Citation
135
+
136
+ ```bibtex
137
+ @article{xiong2025long,
138
+ title={Long-Context Modeling with Dynamic Hierarchical Sparse Attention for On-Device LLMs},
139
+ author={Xiong, Siheng and Zou, Joe and Fekri, Faramarz and Cho, Yae Jee},
140
+ journal={arXiv preprint arXiv:2510.24606},
141
+ year={2025}
142
+ }
143
+ ```
fine-tune/16k_32k/booksum.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b152445bb6da44feca6d1cf69b5799fe4d914f1b0fabddef426944f1e81ecf93
3
+ size 12044335
fine-tune/16k_32k/natural_questions_10_200_docs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c81e00bc189b2eecbfeb835b655fbae34797ac222b04eeb4b2aa93f44ddc5ab
3
+ size 1446375399
fine-tune/32k_64k/booksum.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9bfaacbe2a7f6054e1df74488ba115b95c4489586cdff2c5e0b6757fca8c4a9
3
+ size 3899162
fine-tune/32k_64k/natural_questions_10_200_docs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53d21084dee2cba56fd94af3a252473c83f6c34c2fcf1d42c708c0fe38177845
3
+ size 3079078
fine-tune/64k_128k/booksum.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c0ca8f6db8b4e2be3e21a45d65ce286960efffeb7c30b0418d9bb38c232170bb
3
+ size 605476
fine-tune/8k_16k/booksum.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a31fbb53daa68d0ac61202badad47c2e086d511115db2b0591c4c6f97075c189
3
+ size 15620413
fine-tune/8k_16k/natural_questions_10_200_docs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a5757e58892e2f1af6da9c1be5ccc148269239ce9922a2772bb92cdd69c90d0
3
+ size 485828033
fine-tune/gt_128k/booksum.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b311967ecfececaa1694bd75afbb7e8404e4695113ed5f61ba89630ed84908f
3
+ size 483590
fine-tune/lt_8k/booksum.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:95cee98620e2134756e278851a029e03484053dbab52d179ca9ec36ffa460765
3
+ size 29291053
fine-tune/lt_8k/natural_questions_10_200_docs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5a33a8c7d8f7b4581b9ca02cb2b6610b2bfb134dd66392570fe34c9ace4f25e2
3
+ size 162772652
length_bucket_summary.json ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tokenizer": "meta-llama/Llama-3.1-8B-Instruct",
3
+ "buckets": {
4
+ "lt_8k": {
5
+ "min": 0,
6
+ "max": 8192
7
+ },
8
+ "8k_16k": {
9
+ "min": 8192,
10
+ "max": 16384
11
+ },
12
+ "16k_32k": {
13
+ "min": 16384,
14
+ "max": 32768
15
+ },
16
+ "32k_64k": {
17
+ "min": 32768,
18
+ "max": 65536
19
+ },
20
+ "64k_128k": {
21
+ "min": 65536,
22
+ "max": 131072
23
+ },
24
+ "gt_128k": {
25
+ "min": 131072,
26
+ "max": null
27
+ }
28
+ },
29
+ "pretrain": {
30
+ "length_field": "text",
31
+ "counts": {
32
+ "arxiv_doc_to_abs": {
33
+ "lt_8k": 377854,
34
+ "8k_16k": 559745,
35
+ "16k_32k": 436259,
36
+ "32k_64k": 150632,
37
+ "64k_128k": 28557,
38
+ "gt_128k": 5258
39
+ },
40
+ "rp_sub": {
41
+ "lt_8k": 913818,
42
+ "8k_16k": 11265,
43
+ "16k_32k": 3893,
44
+ "32k_64k": 1019,
45
+ "64k_128k": 368,
46
+ "gt_128k": 151
47
+ },
48
+ "ul2_plus_oscar_en": {
49
+ "lt_8k": 3148215,
50
+ "8k_16k": 342739,
51
+ "16k_32k": 60306,
52
+ "32k_64k": 9213,
53
+ "64k_128k": 1804,
54
+ "gt_128k": 236
55
+ },
56
+ "pile_sub": {
57
+ "lt_8k": 1878091,
58
+ "8k_16k": 43443,
59
+ "16k_32k": 12818,
60
+ "32k_64k": 3022,
61
+ "64k_128k": 1799,
62
+ "gt_128k": 1285
63
+ },
64
+ "NI_decontaminated_materialized": {
65
+ "lt_8k": 460066,
66
+ "8k_16k": 117188,
67
+ "16k_32k": 567,
68
+ "32k_64k": 210,
69
+ "64k_128k": 12,
70
+ "gt_128k": 0
71
+ },
72
+ "P3_decontaminated_materialized": {
73
+ "lt_8k": 757542,
74
+ "8k_16k": 56465,
75
+ "16k_32k": 8,
76
+ "32k_64k": 0,
77
+ "64k_128k": 0,
78
+ "gt_128k": 0
79
+ }
80
+ }
81
+ },
82
+ "fine-tune": {
83
+ "length_field": "prompt",
84
+ "counts": {
85
+ "booksum": {
86
+ "lt_8k": 7909,
87
+ "8k_16k": 1207,
88
+ "16k_32k": 413,
89
+ "32k_64k": 64,
90
+ "64k_128k": 5,
91
+ "gt_128k": 2
92
+ },
93
+ "natural_questions_10_200_docs": {
94
+ "lt_8k": 21370,
95
+ "8k_16k": 25879,
96
+ "16k_32k": 41639,
97
+ "32k_64k": 69,
98
+ "64k_128k": 0,
99
+ "gt_128k": 0
100
+ }
101
+ }
102
+ }
103
+ }
pretrain/16k_32k/NI_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa311f67467b97ce3860b0929240bec849373b71b02ad80068758616ec98b359
3
+ size 14346410
pretrain/16k_32k/P3_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78eef66ab92773b526884f3ebe991bd1c0349e16542fed0e8403d1518dc4b9a8
3
+ size 126602
pretrain/16k_32k/arxiv_doc_to_abs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4da759139e3445bc5b706dbb9488330e01a0abcffcc469deb046523a07a25e1d
3
+ size 9237421956
pretrain/16k_32k/pile_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f42a76d2bef60dd2f1602302fe3764500828068c727cf6c3f21891b5fa48ca24
3
+ size 298260337
pretrain/16k_32k/rp_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7dd44c81d54e7bac5bf4e675e5a82ec2926f2442d9eed3ed3d8ffc43b256fac5
3
+ size 120850953
pretrain/16k_32k/ul2_plus_oscar_en.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c3667ccc7d50a9fa56cac1215f99e5875da4d6a13aa03d6cec14e179c26991af
3
+ size 2033775807
pretrain/32k_64k/NI_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c29f4bb97091947d4be3b6db84b5e72f5e054080bfe5e7001ed5c942092ef494
3
+ size 8679743
pretrain/32k_64k/arxiv_doc_to_abs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc0c74e509b9dcbd9b0b14bd94a617edb33737468b83e439b936c80b6fca5aad
3
+ size 5152350883
pretrain/32k_64k/pile_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:32770f9bf8031df4f64e90ea09780676d085250a5a553b3c2ae48648bf8f8fc0
3
+ size 139689345
pretrain/32k_64k/rp_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:866fbdac6796b21740c6db4095aebe3243f3a300c8609e2c232eebec402dad5e
3
+ size 58331120
pretrain/32k_64k/ul2_plus_oscar_en.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:18993d727f381e242143ae2631098a01d31dbf4442046aa1baffbced45729eaf
3
+ size 601118058
pretrain/64k_128k/NI_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecaa31b01e2f7bf70bbf44230f8011480f273f53e3d8385d89626427a02b77f0
3
+ size 349205
pretrain/64k_128k/arxiv_doc_to_abs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c34a4d6236f13065de4e3c7008cabfc3dce4d85f8425c229985c7dd2477c3824
3
+ size 1731854726
pretrain/64k_128k/pile_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10d0225375675ae70fbc7725eabd5df57c5e84422e2d5ca66faf011463959391
3
+ size 242078467
pretrain/64k_128k/rp_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5c2ee8846e9f4caf85e91141294e2883d41ffc6daa224d6f6c390091c4764f1a
3
+ size 43814314
pretrain/64k_128k/ul2_plus_oscar_en.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b89015cd8021c8bdc0087c3f343ce02814e6016d01d5c8837b768c0dbbb778ca
3
+ size 226482363
pretrain/8k_16k/NI_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:487cad4bc8fa5b978ae6d34aafa283103490031602c957aad5d520397db2e368
3
+ size 1004080076
pretrain/8k_16k/P3_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d1be5933168b2d2493584efa70a771e71f40856a79747439178a859fe74d0659
3
+ size 795283951
pretrain/8k_16k/arxiv_doc_to_abs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8328f7006930bb92730995cf4ea308670cce9b5b7c7c583f2bbb53a43ea8d61
3
+ size 7204802163
pretrain/8k_16k/pile_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d463632bba24dccfd1e93a8319201a8f29910b0813fe1309201a6b4fde82b93
3
+ size 612190563
pretrain/8k_16k/rp_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1aa0010b631dbf6c019274d9b04fdb74c9cb52c45140c148e61bbc7696aefa8
3
+ size 193692147
pretrain/8k_16k/ul2_plus_oscar_en.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ebad4b85569489b01278013c62fb421077861b7516f982e2a74f9798466c6df3
3
+ size 6135262983
pretrain/gt_128k/arxiv_doc_to_abs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a6450dfa3da8c65eae1c58b53c0c427f7719e800eadf0ad30339515831c1187
3
+ size 676435349
pretrain/gt_128k/pile_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12b9973913e3a7f3f953560d2ee3c9f3bcaad1598a95aac0bc67270d5e013c9f
3
+ size 466034209
pretrain/gt_128k/rp_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e35fdc80beb9596703d6fdeb4d26500e60e23acad7e8a59e2ab0b95f4f40e6cd
3
+ size 38751678
pretrain/gt_128k/ul2_plus_oscar_en.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82b6bda88c3afed2cdb6cad754c02aff295aeb702dabedf803b4512634d12848
3
+ size 55911570
pretrain/lt_8k/NI_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3dfbfc31fd3e2a73275731ca941e2f34434601d3159fc37ab7fda563df1265c5
3
+ size 3993987525
pretrain/lt_8k/P3_decontaminated_materialized.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d9206217c9380e177eac61b6157ee3d139dd273f1e7d3480f50c43fe1e57ce2c
3
+ size 9373286073
pretrain/lt_8k/arxiv_doc_to_abs.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4633db0d9ae7d8b1590b6fea21880a7cd4d5f6c5554456c01807aa993d46a5c4
3
+ size 2405405082
pretrain/lt_8k/pile_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:485f5529bc7f69db480ba61352ab586484479266158e9278220476449a4bdae1
3
+ size 2442803283
pretrain/lt_8k/rp_sub.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2caef3a9981b345617785cd8f10a973b6627afd8f5509cedbd14a84cbe0044be
3
+ size 1500201842
pretrain/lt_8k/ul2_plus_oscar_en.jsonl.zst ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:592011c63dedd4c76ebf20aa10527050184f4a1bf04a89a9d3fb7c6f4fddd2b5
3
+ size 21857262057