gpt-5.6-luna-sft-900x

# GPT-5.6 Luna Diverse SFT 900 Diverse synthetic SFT traces distilled from **openai/gpt-5.6-luna** through an OpenAI-compatible endpoint. Each row is already formatted as a complete `system` / `user` / `assistant` conversation and can be passed directly to a tokenizer's chat template. The prompt mix is intentionally task-diverse rather than benchmark- or multiple-choice-shaped. The assistant response uses a consistent reasoning format: ```text <think> deliberate reasoning trace </think> final response ``` ## Dataset summary - **Rows:** 890 - **Task families:** 10 - **Language:** English - **Teacher model:** `openai/gpt-5.6-luna` - **Format:** chat-template-ready JSONL - **Generated:** 2026-07-16 - **Generator:** OpenStill - **Source file:** `sft.jsonl` OpenStill is Empero's internal distillation toolkit and is not publicly available yet. This release contains the 890 traces from a larger 3,500-prompt generation corpus. ## Using it for SFT The `messages` field is ready for `apply_chat_template`; no DPO conversion or prompt reconstruction is required. ```python from datasets import load_dataset from transformers import AutoTokenizer dataset = load_dataset("json", data_files="sft.jsonl", split="train") tokenizer = AutoTokenizer.from_pretrained("YOUR_BASE_MODEL") def render(example): return { "text": tokenizer.apply_chat_template( example["messages"], tokenize=False, add_generation_prompt=False, ) } dataset = dataset.map(render) ``` For trainers that accept conversational datasets directly, use `messages` without mapping it to a text field. ## Fields | field | description | |---|---| | `id` | Stable trace identifier from the source prompt corpus | | `topic` | Task family and generated domain, separated by ` / ` | | `model` | Teacher model used to generate the trace (`openai/gpt-5.6-luna`) | | `messages` | Ordered chat turns with `system`, `user`, and `assistant` roles | | `messages[0].content` | System instruction used for the training conversation | | `messages[1].content` | Diverse end-user request | | `messages[2].content` | Full `<think>...</think>` reasoning trace followed by the final response | ## Task distribution | task family | rows | what it contains | |---|---:|---| | `coding_and_debugging` | 135 | Implementation, debugging, refactoring, and code review | | `quantitative_problem_solving` | 133 | Applied mathematics, statistics, estimation, and numerical reasoning | | `planning_and_decision_making` | 106 | Practical plans with constraints, resources, and tradeoffs | | `scientific_and_technical_reasoning` | 106 | Science and engineering explanations, calculations, and predictions | | `analysis_and_argument` | 90 | Evidence-based analysis of claims, scenarios, and competing explanations | | `extraction_and_structured_output` | 90 | JSON, tables, classification, extraction, and data transformation | | `writing_and_transformation` | 72 | Rewriting, editing, summarization, and audience-aware drafting | | `troubleshooting_and_diagnosis` | 71 | Diagnosis from symptoms, logs, or operational context | | `logic_and_constraints` | 45 | Scheduling, ordering, logic, and constraint-satisfaction tasks | | `creative_generation_with_constraints` | 42 | Creative artifacts with testable structural or stylistic requirements | ## Statistics | measurement | minimum | median | mean | maximum | |---|---:|---:|---:|---:| | Reasoning length (words) | 74 | 164 | 167.1 | 274 | | User prompt length (characters) | 258 | 541 | 554.5 | 1,127 | | Final response length (characters) | 87 | 999 | 1,377.1 | 9,130 | All 890 rows have the same role sequence: ```text system → user → assistant ``` ## Validation & limitations - Every retained row passed structural validation: valid JSONL, the expected three-message role sequence, a non-empty reasoning trace, a non-empty final response, canonical `<think>` boundaries, and no hidden teacher/model metadata in the reasoning. - Structural validation is **not** deterministic correctness verification. Code was not executed, calculations were not checked against reference answers, and open-ended outputs were not judged by a separate model. - The data is synthetic and may contain factual mistakes, imperfect code, unsupported assumptions, or stylistic artifacts from the teacher model. Review or filter it for high-stakes use cases. - The dataset contains explicit reasoning traces. Confirm that the target model, tokenizer, and training recipe handle `<think>` tags as intended. - This is a partial snapshot: 890 traces from 3,500-prompt corpus. ## Support / Donate If this dataset or tooling helped you, consider supporting the project: - **BTC:** `bc1qx6zepu6sfkvshgdmc4ewu6pk6rpadvpgffpp7v` - **LTC:** `ltc1qv2mefzps2vtjcpwfx8xxdrpplrcvltswm68r7x` - **XMR:** `42Dbm5xg5Nq26fdyzfEU7KBnAJfhi7Cvz5J2ex5CzHXkfKuNEJzYCcmJ1GTbgjFZ5MBx72sdG1G9239Cd6rsZfv4QeDkYJY` ## Links - Project: [https://empero.org](https://empero.org) - Teacher route: `openai/gpt-5.6-luna` - Generated with OpenStill, Empero's internal distillation toolkit (not publicly available yet), using an OpenAI-compatible chat-completions API