Dataset · Hugging Face

ArrowLabs/DCLM-Ultra

Name: DCLM-Ultra
Creator: ArrowLabs
License: apache-2.0
Keywords: task_categories:text-generation, task_categories:question-answering, language:en, license:apache-2.0, size_categories:1M<n<10M, format:parquet, modality:text, library:datasets, library:dask, library:polars, library:mlcroissant, region:us, DCLM, fineweb, edu

DCLM Ultra DCLM Ultra is an improved, deduplicated and filtered version of DCLM Pro (Filtered DCLM, 1.5% better compared to DCLM baseline). It achieves superior performance across benchmarks including hellaswag and piqa, also achieving better loss compared to pro. 3M model trained on only DCLM Pro vs only Ultra, same architecture, same training budget. Metric DCLM Pro DCLM Ultra Training loss 3.37 3.27 PIQA 53.65% 54.95% HellaSwag 25.45% 25.54%… See the full description on the dataset page: https://huggingface.co/datasets/ArrowLabs/DCLM-Ultra.

View on Hugging FaceReport an issue

Author / organization	ArrowLabs
License (as declared)	apache-2.0
Downloads	916 (as reported by Hugging Face)
Likes	2
Created	18 October 2026
Last updated at source	28 October 2026

DCLMfinewebedupretrainpretrainingwebreasonreasoningdensefinefine dataultraproDCLM ProDCLM baselinethoughtthoughtsthinkthinkingpost trainposttrain

Embed this entry

<iframe src="https://silytics.si/embed/item/4488338f-3eb3-4939-80d2-21dc53f908b6" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Popularity, last 22 days

916 now (+536 since 2026-10-20). Counts as reported by the source, snapshotted daily by Silytics.

Connected in the knowledge graph

ArrowLabs/AresXS1-PreviewModel · used to train
nkthebass/tinybrainbot-10m-v2-baseModel · used to train
onnx-community/tinybrainbot-10m-v2-base-ONNXModel · used to train

Explore this entry in the graph

Recent changes

28 Oct 2026 Description: DCLM Ultra DCLM Ultra is an improved, deduplicated and filtered version of DCLM Pro (Filtered DCLM, 1.5 percent better than the original). It achieves superior performance across benchmarks like hellaswag and piqa while also achieving bette → DCLM Ultra DCLM Ultra is an improved, deduplicated and filtered version of DCLM Pro (Filtered DCLM, 1.5% better compared to DCLM baseline). It achieves superior performance across benchmarks including hellaswag and piqa, also achieving bett

Changes detected by Silytics between refreshes of the source.

More from ArrowLabs

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 10 November 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.