domyn/Domyn-Edge-Teacher-Agentic-Mix

Domyn Edge Teacher Agentic Mix

Phased release. Domyn Edge is a rolling release in three bundles. This dataset is part of the second bundle, together with the domain teachers and their post-training data. For an overview, see the launch blog post.

The training data of Domyn Edge Teacher Agentic

, in two configs:

Config	Stage	Train	Test
sft	Supervised fine-tuning, 128k context	5,562,889 (54.73B tokens)	291,167 (2.87B tokens)

sft: supervised fine-tuning mix

The agentic, tool-use, software-engineering, reasoning, and deep-research slices of the Domyn Edge SFT corpus. Code and math belong to the code and math teachers respectively, so both have weight 0 here. 5% of the mix is held out as the test split.

Task	Weight	Sources
Agentic (general tool-use)	1.0	Toucan-1.5M, ToolMind, INTELLECT-3-SFT (openreasoning_tool, toucan_tool), ToolACE, Nemotron-Agentic-v1, Nemotron-SFT-Agentic-v2, Nemotron-SFT-OpenCode-v1, Nemotron-Terminal-Corpus, deepseek-v4-pro-0813-agentic
Software engineering	1.0	daVinci-Dev, Scale-SWE-Distilled, SWE-agent-trajectories, Nemotron-SFT-SWE-v2, Nemotron-SFT-SWE-v3
Reasoning	1.0	Nemotron-SFT-ARC-AGI-v1 (tool-use splits only)
Deep research	1.0	Domyn internal deep-research SDG, MiroVerse-v0.1, OpenResearcher-Dataset
Function calling	1.0	Domyn internal NeMo-Gym tool-call trajectories

Released conversations and tokens per source, train and test splits together (licenses in short; full terms in ATTRIBUTION.md):

Source	License	Conversations	Tokens
Nemotron-Terminal-Corpus	CC BY 4.0	1,464,092	25.85B
Domyn internal (deep research SDG)	Domyn	1,261,267	3.16B
Nemotron-SFT-Agentic-v2	CC BY 4.0	970,967	5.39B
Nemotron-SFT-OpenCode-v1	CC BY 4.0	449,380	6.81B
INTELLECT-3-SFT	CC BY 4.0, MIT, Apache 2.0 (per subset)	394,060	1.74B
ToolMind	Apache 2.0	337,881	0.82B
Nemotron-Agentic-v1	CC BY 4.0	314,600	1.30B
Nemotron-SFT-SWE-v2	CC BY 4.0	256,194	3.69B
Toucan-1.5M	Apache 2.0	115,623	0.35B
daVinci-Dev	other (mixed permissive, CC BY 4.0; gated)	70,602	2.58B
Scale-SWE-Distilled	CC BY 4.0	65,950	3.24B
MiroVerse-v0.1	CC BY-NC 4.0, CC BY-SA 4.0, MIT, Apache 2.0 (per subset)	63,726	0.67B
Nemotron-SFT-ARC-AGI-v1	Apache 2.0	53,420	1.28B
ToolACE	Apache 2.0	10,417	0.01B
Domyn internal (NeMo-Gym tool trajectories)	Domyn	9,455	0.45B
Nemotron-SFT-SWE-v3	CC BY 4.0	5,932	0.17B
deepseek-v4-pro-0813-agentic	other (synthetic research corpus)	4,977	0.01B
SWE-agent-trajectories	CC BY 4.0	3,574	0.03B
OpenResearcher-Dataset	MIT	1,939	0.07B
Total		5,854,056	57.60B

How to use

from datasets import load_dataset

sft = load_dataset("domyn/Domyn-Edge-Teacher-Agentic-Mix", "sft", split="train", streaming=True)

Columns

sft columns

Column	Type	Description
messages	list	The conversation. Each message has role, content, reasoning_content (assistant reasoning, if any), tool_calls (each with id, type and function {name, arguments as a JSON string}), and name and tool_call_id for tool results.
tools	string	JSON-encoded list of the tool schemas available in the conversation, or null.
task	string	Task family: one of agentic, swe, reasoning, deep-research, function_calling.
source	string	Upstream dataset.
license	string	License of the upstream record.
metadata	string	JSON-encoded upstream metadata (for example topic or generation model), or null.
uuid	string	Record identifier.
token_length	int	Conversation length in tokens.

License

Our contributions to this dataset, including the curation and the mix composition, are released under the Apache License 2.0. Upstream datasets keep their original license terms. Read ATTRIBUTION.md before redistributing any part of this data.

Citation

@misc{domyn_edge_teacher_agentic_mix_2026,
  title     = {Domyn Edge Teacher Agentic Mix},
  author    = {Domyn},
  year      = {2026},
  note      = {Technical report, forthcoming in November 2026.},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/domyn/Domyn-Edge-Teacher-Agentic-Mix}
}