MiniMax M3 70K Codemath: Verified Math-Code Distillation Traces

Verified distillation traces generated with faststill v0.0.1. The dataset contains (prompt, reasoning, output) triplets from an OpenAI-compatible chat-completions endpoint, with each row verified by a machine check before inclusion. The dataset was created by empero-ai and last updated on 2026-06-16.

Use Cases

• Training student models for math problem-solving based on verified reasoning traces.
• Fine-tuning language models for code generation tasks based on verified output examples.
• Evaluating the quality of instruction-following models based on the verified triplet structure.

Strengths

• Rows are verified via deterministic machine checks, filtering out incorrect labels.
• The dataset was generated using a specific pipeline (faststill v0.0.1) designed for creating verified triplets.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

huggingface

Collection Method

Generated via the faststill v0.0.1 pipeline from an OpenAI-compatible chat-completions endpoint.

Freshness

Last updated 2026-06-16 00:19:51; freshness should be verified.