dfm8
Datasets
All datasets matching “dfm8”dfm8-openhermes-en
DFM8 OpenHermes English
An English instruction-conversation dataset derived from OpenHermes 2.5 by
model-based auditing, selective answer repair, and post-repair auditing.
This dataset contains both retained original conversations and regenerated
conversations; it is not exclusively synthetic repairs.
Data and provenance
Source: teknium/OpenHermes-2.5. Local source IDs use the form
openhermes_2_5:train:<index>. The precise upstream revision is not established
by… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm8-openhermes-en.dfm8-synthetic-native-tool-calling
Native Tool Calling
Synthetic DFM8 training data generated with Gemma 4 31B and filtered by deterministic checks plus a Gemma 4 31B judge.
Schema
Rows are JSONL chat records:
{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
Tool-calling rows may also include a top-level tools list and assistant tool_calls.
Counts
accepted rows: 836675
generated rows seen: 4800000
audit rows seen: 4580233… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm8-synthetic-native-tool-calling.dfm8-synthetic-multiturn-danish-english-chat
Multi-turn Danish and English Chat
Synthetic DFM8 training data generated with Gemma 4 31B and filtered by deterministic checks plus a Gemma 4 31B judge.
Schema
Rows are JSONL chat records:
{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
Tool-calling rows may also include a top-level tools list and assistant tool_calls.
Counts
accepted rows: 414196
generated rows seen: 4800000
audit rows seen: 4580233… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm8-synthetic-multiturn-danish-english-chat.dfm8-synthetic-strict-math-answer-contract
Strict Math Answer Contract
Synthetic DFM8 training data generated with Gemma 4 31B and filtered by deterministic checks plus a Gemma 4 31B judge.
Schema
Rows are JSONL chat records:
{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
Tool-calling rows may also include a top-level tools list and assistant tool_calls.
Counts
accepted rows: 552952
generated rows seen: 4800000
audit rows seen: 4580233… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm8-synthetic-strict-math-answer-contract.dfm8-synthetic-constrained-format-following
Constrained Format Following
Synthetic DFM8 training data generated with Gemma 4 31B and filtered by deterministic checks plus a Gemma 4 31B judge.
Schema
Rows are JSONL chat records:
{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
Tool-calling rows may also include a top-level tools list and assistant tool_calls.
Counts
accepted rows: 838905
generated rows seen: 4800000
audit rows seen: 4580233… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm8-synthetic-constrained-format-following.dfm8-synthetic-code-debugging
Code Generation and Debugging
Synthetic DFM8 training data generated with Gemma 4 31B and filtered by deterministic checks plus a Gemma 4 31B judge.
Schema
Rows are JSONL chat records:
{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
Tool-calling rows may also include a top-level tools list and assistant tool_calls.
Counts
accepted rows: 340711
generated rows seen: 4800000
audit rows seen: 4580233… See the full description on the dataset page: https://huggingface.co/datasets/schneiderkamplab/dfm8-synthetic-code-debugging.