Datasets:

common-pile
/

comma_v0.1_training_dataset

Modalities:

Formats:

Size:

ArXiv:

Libraries:

Dataset card Data Studio Files Files and versions

craffel HF Staff commited on Jun 6

Commit

5afc546

·

verified ·

1 Parent(s): f5c81ad

Update README.md

Files changed (1) hide show

README.md +1 -0

README.md CHANGED Viewed

@@ -11,6 +11,7 @@ The Comma v0.1 models were trained in two stages, a "main" stage and a "cooldown
 During each stage, we heuristically set mixing rates to up or downweight different sources.
 In the two tables below, we provide per-source token count, repeat rate, and effective token count (after up/downweighting) for the main and cooldown stage of the Comma v0.1-1T training run.
 For the Comma v0.1-2T training run, all sources are repeated 2x as many times in both stages.
 | Main stage                    | Tokens (B) | Repeats | Effective tokens (B) |

 During each stage, we heuristically set mixing rates to up or downweight different sources.
 In the two tables below, we provide per-source token count, repeat rate, and effective token count (after up/downweighting) for the main and cooldown stage of the Comma v0.1-1T training run.
 For the Comma v0.1-2T training run, all sources are repeated 2x as many times in both stages.
+Token counts are as provided by the Comma v0.1 tokenizer; using a different tokenizer may change these counts significantly.
 | Main stage                    | Tokens (B) | Repeats | Effective tokens (B) |