Datasets:

Formats:
parquet
Libraries:
Datasets
Dask
License:
jwkirchenbauer commited on
Commit
90f78af
·
verified ·
1 Parent(s): c1aa036

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +11 -9
README.md CHANGED
@@ -4,6 +4,7 @@ configs:
4
  data_files:
5
  - split: train
6
  path: "*.parquet"
 
7
  ---
8
  Gemstones Training Dataset - Sequential version
9
 
@@ -17,6 +18,12 @@ reproduce the training batches across the gpus is/was the run the training code.
17
  This repo is the result of an attempt to simulate the way in which the training code loaded the data and
18
  stream it out to a portable file format for use in downstream analyses of the model suite.
19
 
 
 
 
 
 
 
20
  # Sharding format: sequential
21
 
22
  This version of the dataset approximates the order of the dataset _as if_ a model was being trained
@@ -35,7 +42,7 @@ have seen all of the same rows of the dataset during training. The sync granular
35
 
36
  This linearized recreation assumes a single worker is reading every row of the dataset and so at a microbatch size of 8 over packed sequences of 2048 tokens, 21247488 steps worth of "training" is required to reach ~350B tokens.
37
 
38
- At runtime, the the single worker receives the total dataset represented by the thousands of raw training format files (for reference, this format is defined by the `packed_cycle_dataset.py` file in this repo).
39
  The raw files were first shuffled globally, and then the single worker
40
  loaded 4 files at a time, and shuffled the "blocks" of 2048 tokens each in a temporary buffer so
41
  that the contents of the 4 packed files were not read in the exact order in which the tokens appeared in them.
@@ -44,12 +51,7 @@ that the contents of the 4 packed files were not read in the exact order in whic
44
  a time whose contents (blocks of tokens) are read in a shuffled order, does not exactly match any one of the
45
  Gemstones model sets. However, the key is that the synchronization argument above still holds and so analyses at a coarser granularity than ~8.6B tokens should be sound.
46
 
47
- The `train_mock_data_order_file.py` performs these operations and writes the resulting data order out to files named like
 
48
  `ordered_dataset_shard_{shard}-of-{total_shards}.parquet` where the total number of shards is arbitrary, but chosen to be 256 for
49
- portability.
50
-
51
- # Loading
52
-
53
- This data should be loadable using `load_dataset` in the standard manner to auto-download the data.
54
- Alternately, the dataset can be cloned using git to materialize the files locally, and then loaded
55
- using the default `parquet` builder as described here: https://huggingface.co/docs/datasets/en/loading#parquet
 
4
  data_files:
5
  - split: train
6
  path: "*.parquet"
7
+ license: odc-by
8
  ---
9
  Gemstones Training Dataset - Sequential version
10
 
 
18
  This repo is the result of an attempt to simulate the way in which the training code loaded the data and
19
  stream it out to a portable file format for use in downstream analyses of the model suite.
20
 
21
+ # Loading
22
+
23
+ This data should be loadable using `load_dataset` in the standard manner to auto-download the data.
24
+ Alternately, the dataset can be cloned using git to materialize the files locally, and then loaded
25
+ using the default `parquet` builder as described here: https://huggingface.co/docs/datasets/en/loading#parquet
26
+
27
  # Sharding format: sequential
28
 
29
  This version of the dataset approximates the order of the dataset _as if_ a model was being trained
 
42
 
43
  This linearized recreation assumes a single worker is reading every row of the dataset and so at a microbatch size of 8 over packed sequences of 2048 tokens, 21247488 steps worth of "training" is required to reach ~350B tokens.
44
 
45
+ In this setup, a single worker received the total dataset represented by the thousands of raw training format files (for reference, this format is defined by the `packed_cycle_dataset.py` file in this repo).
46
  The raw files were first shuffled globally, and then the single worker
47
  loaded 4 files at a time, and shuffled the "blocks" of 2048 tokens each in a temporary buffer so
48
  that the contents of the 4 packed files were not read in the exact order in which the tokens appeared in them.
 
51
  a time whose contents (blocks of tokens) are read in a shuffled order, does not exactly match any one of the
52
  Gemstones model sets. However, the key is that the synchronization argument above still holds and so analyses at a coarser granularity than ~8.6B tokens should be sound.
53
 
54
+ The `train_mock_data_order_file.py` performs these operations and writes the resulting data order out to files.
55
+ Each shard named like
56
  `ordered_dataset_shard_{shard}-of-{total_shards}.parquet` where the total number of shards is arbitrary, but chosen to be 256 for
57
+ portability, represents a contiguous subset of the approximated total ordering of the rows int the training dataset.