Upload 20 files
Browse files- README.md +53 -3
- dataset.json +13 -0
- packed_manifest.json +16 -0
- tokens-000000.bin +3 -0
- tokens-000001.bin +3 -0
- tokens-000002.bin +3 -0
- tokens-000003.bin +3 -0
- tokens-000004.bin +3 -0
- tokens-000005.bin +3 -0
- tokens-000006.bin +3 -0
- tokens-000007.bin +3 -0
- tokens-000008.bin +3 -0
- tokens-000009.bin +3 -0
- tokens-000010.bin +3 -0
- tokens-000011.bin +3 -0
- tokens-000012.bin +3 -0
- tokens-000013.bin +3 -0
- tokens-000014.bin +3 -0
- tokens-000015.bin +3 -0
- tokens-000016.bin +3 -0
README.md
CHANGED
|
@@ -1,3 +1,53 @@
|
|
| 1 |
-
-
|
| 2 |
-
|
| 3 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Auralis-Coder Training Dataset
|
| 2 |
+
|
| 3 |
+
Packed token dataset for training the **Auralis-Coder** 1.5B parameter language model.
|
| 4 |
+
|
| 5 |
+
## Dataset Overview
|
| 6 |
+
|
| 7 |
+
| Property | Value |
|
| 8 |
+
|---|---|
|
| 9 |
+
| Total tokens | 176,544,387 |
|
| 10 |
+
| Tokenizer | Qwen2.5 (Qwen/Qwen2.5-1.5B) |
|
| 11 |
+
| Format | uint32 token IDs (packed) |
|
| 12 |
+
| Shards | 16 |
|
| 13 |
+
| Shard size | ~10M tokens each |
|
| 14 |
+
| Documents | 235,219 |
|
| 15 |
+
| Duplicates removed | 19 |
|
| 16 |
+
|
| 17 |
+
## Source Data
|
| 18 |
+
|
| 19 |
+
The dataset combines high-quality sources targeting programming and artificial intelligence:
|
| 20 |
+
|
| 21 |
+
- **FineWeb-Edu** — Educational web content
|
| 22 |
+
- **StarCoder** — Programming/code data
|
| 23 |
+
- **FineWeb** — General high-quality web content
|
| 24 |
+
|
| 25 |
+
## Tokenization & Packing
|
| 26 |
+
|
| 27 |
+
- Tokenized with Qwen2.5 tokenizer (vocab: 151,936)
|
| 28 |
+
- Packed into fixed-size shards (10M tokens each)
|
| 29 |
+
- Sequence length: 4,096 tokens during training
|
| 30 |
+
- Maximum context: 131,072 tokens
|
| 31 |
+
|
| 32 |
+
## Usage
|
| 33 |
+
|
| 34 |
+
```python
|
| 35 |
+
import numpy as np
|
| 36 |
+
|
| 37 |
+
# Load a shard
|
| 38 |
+
tokens = np.fromfile("tokens-000000.bin", dtype=np.uint32)
|
| 39 |
+
|
| 40 |
+
# Reshape into sequences (e.g., 4096 tokens per sequence)
|
| 41 |
+
seq_len = 4096
|
| 42 |
+
sequences = tokens.reshape(-1, seq_len)
|
| 43 |
+
```
|
| 44 |
+
|
| 45 |
+
## Files
|
| 46 |
+
|
| 47 |
+
- `tokens-000000.bin` through `tokens-000015.bin` — Packed token shards (uint32)
|
| 48 |
+
- `packed_manifest.json` — Dataset statistics and metadata
|
| 49 |
+
- `dataset.json` — Additional metadata
|
| 50 |
+
|
| 51 |
+
## License
|
| 52 |
+
|
| 53 |
+
Refer to the source dataset licenses (FineWeb, StarCoder, FineWeb-Edu) and the Auralis project distribution.
|
dataset.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"dataset_name": "Auralis_packed_dataset",
|
| 3 |
+
"token_format": "uint32",
|
| 4 |
+
"tokenizer": "Qwen/Qwen2.5-1.5B",
|
| 5 |
+
"existing_tokens_before_run": 176544387,
|
| 6 |
+
"new_tokens_this_run": 4141,
|
| 7 |
+
"total_tokens_after_run": 176548528,
|
| 8 |
+
"documents_tokenized": 1,
|
| 9 |
+
"duplicates_skipped": 0,
|
| 10 |
+
"records_skipped": 0,
|
| 11 |
+
"existing_shards_preserved": true,
|
| 12 |
+
"new_shards_start": 16
|
| 13 |
+
}
|
packed_manifest.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"format": "uint32 token IDs",
|
| 3 |
+
"tokenizer": "Qwen/Qwen2.5-1.5B",
|
| 4 |
+
"original_tokens": 147256664,
|
| 5 |
+
"new_tokens": 29287723,
|
| 6 |
+
"total_tokens": 176544387,
|
| 7 |
+
"requested_new_tokens": 70000000,
|
| 8 |
+
"documents_used": 235201,
|
| 9 |
+
"documents_skipped": 0,
|
| 10 |
+
"duplicates_removed": 19,
|
| 11 |
+
"approx_new_source_characters": 123902601,
|
| 12 |
+
"new_shard_size_tokens": 10000000,
|
| 13 |
+
"original_shards_preserved": true,
|
| 14 |
+
"storage": "packed",
|
| 15 |
+
"note": "Existing packed shards were preserved. New data was appended as subsequent token shards."
|
| 16 |
+
}
|
tokens-000000.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:89710b24f3ac74cf83b121f18095e1300a59a83d51bfad724d6cf7745e2fe0d9
|
| 3 |
+
size 120432176
|
tokens-000001.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b402d00eb537c6f8be7f4f50c3010a9610b3b519c4c7d3c60cb8a69ff5ccea2d
|
| 3 |
+
size 40000000
|
tokens-000002.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2566ee93e95643052f9f9644a514f8e9e1f6995e6ba36048707a785a6faeb2bb
|
| 3 |
+
size 40000000
|
tokens-000003.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:40115e20ecd5a1c0b41f99f2a45320ea2ad7dee786f4fafff41c68e441efd945
|
| 3 |
+
size 37148620
|
tokens-000004.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b402d00eb537c6f8be7f4f50c3010a9610b3b519c4c7d3c60cb8a69ff5ccea2d
|
| 3 |
+
size 40000000
|
tokens-000005.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2566ee93e95643052f9f9644a514f8e9e1f6995e6ba36048707a785a6faeb2bb
|
| 3 |
+
size 40000000
|
tokens-000006.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:40115e20ecd5a1c0b41f99f2a45320ea2ad7dee786f4fafff41c68e441efd945
|
| 3 |
+
size 37148620
|
tokens-000007.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b402d00eb537c6f8be7f4f50c3010a9610b3b519c4c7d3c60cb8a69ff5ccea2d
|
| 3 |
+
size 40000000
|
tokens-000008.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2566ee93e95643052f9f9644a514f8e9e1f6995e6ba36048707a785a6faeb2bb
|
| 3 |
+
size 40000000
|
tokens-000009.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:40115e20ecd5a1c0b41f99f2a45320ea2ad7dee786f4fafff41c68e441efd945
|
| 3 |
+
size 37148620
|
tokens-000010.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b402d00eb537c6f8be7f4f50c3010a9610b3b519c4c7d3c60cb8a69ff5ccea2d
|
| 3 |
+
size 40000000
|
tokens-000011.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2566ee93e95643052f9f9644a514f8e9e1f6995e6ba36048707a785a6faeb2bb
|
| 3 |
+
size 40000000
|
tokens-000012.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:40115e20ecd5a1c0b41f99f2a45320ea2ad7dee786f4fafff41c68e441efd945
|
| 3 |
+
size 37148620
|
tokens-000013.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b402d00eb537c6f8be7f4f50c3010a9610b3b519c4c7d3c60cb8a69ff5ccea2d
|
| 3 |
+
size 40000000
|
tokens-000014.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2566ee93e95643052f9f9644a514f8e9e1f6995e6ba36048707a785a6faeb2bb
|
| 3 |
+
size 40000000
|
tokens-000015.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e5d0e80c93420529abb1305660531877205acc7356a0d8bec25fe35b167c4feb
|
| 3 |
+
size 37150892
|
tokens-000016.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e2614aecf5aec10d27f0c9d3a3687232c508c8f98c57fa5e1d43dfa97d8cef5
|
| 3 |
+
size 16564
|