Sign in to view source links and access this dataset
Description
462 examples of complete, unrestricted reasoning traces distilled from the Mythos V2 language model, totaling approximately 104.7 million characters. HelioAI Labs created this dataset for deep analytical research and process supervision, with a last recorded update in June 2026. The distillation process involved zero alignment truncation, preserving the full reasoning chain.
Use Cases
Training process supervision models based on the full-parameter reasoning traces.
Analyzing the structure and patterns of language model reasoning based on the complete traces.
Developing methods for AI interpretability based on the detailed reasoning steps.
Benchmarking model reasoning capabilities against the distilled examples.
Strengths
Contains 462 distinct reasoning examples.
Traces comprise approximately 104.7 million characters, indicating substantial textual detail.
Uses unrestricted full-parameter distillation, likely preserving nuanced internal model states.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Provenance
Source
HelioAI Labs, via Hugging Face.
Collection Method
Unrestricted full-parameter distillation from the Mythos V2 language model.
Freshness
Last updated 2026-06-15 22:35:42; freshness should be verified.
License is unknown; terms of use must be verified before application.