Dataset Schema & Parquet Specification
Detailed Apache Parquet field definitions, reasoning token tags, and multi-teacher provenance metadata.
Hugging Face Repository:Solstice-AI/Solace-1.0-GLM5.2-Fable5-GPT5.6Sol-DeepSeekV4Pro0813-Qwen3.8Max-KimiK3-Manus ↗
Parquet Columns11 Primary Fields
CompressionSnappy
Row Count482,910 Traces
Total Size4.82 GB (Compressed)
Overview
Solace 1.0 is distributed as Apache Parquet files partitioned by task domain. Each sample includes the raw problem statement, the multi-turn verified scratchpad, final grounded solution, and comprehensive provenance metadata attributing teacher models and validation metrics.
Parquet Schema Reference
{
"id": "solace_v1_math_0049182",
"domain": "mathematics",
"subdomain": "olympiad_number_theory",
"teacher_models": ["gpt-5.6-sol", "deepseek-v4-pro", "glm-5.2"],
"consensus_score": 0.942,
"verification_passed": true,
"compiler_feedback": null,
"prompt": "Find all positive integers n such that 2^n + 12^n + 2026^n is a perfect square.",
"thought_chain": "1. Analyze modulo arithmetic...\n2. Test base cases n=1, 2, 3...\n3. For n > 2, examine mod 7 and mod 13...",
"solution": "The only positive integer solution is n = 1.",
"token_count": 1420
}
Field Descriptions
| Field Name | Type | Description |
|---|---|---|
id |
String |
Unique cryptographic identifier based on prompt hash. |
domain |
String |
High-level domain: mathematics, code_synthesis, agentic_tool_use, strategic_reasoning. |
subdomain |
String |
Granular specialization (e.g. dynamic_programming, kernel_optimization, proof_verification). |
teacher_models |
List[String] |
List of teacher architectures that reached consensus on this solution path. |
consensus_score |
Float |
Semantic reasoning graph overlap score (ranges from $0.0$ to $1.0$). |
verification_passed |
Boolean |
Flag indicating successful pass through compiler sandboxes and proof engines. |
compiler_feedback |
Nullable[String] |
Output logs from unit tests, linter assertions, or Lean 4 proof logs if applicable. |
prompt |
String |
Unmodified task instruction or user problem statement. |
thought_chain |
String |
Full step-by-step thinking process without truncation or artificial summaries. |
solution |
String |
Final grounded answer and code artifacts. |
token_count |
Integer |
Total token count across prompt, thought chain, and solution. |