Talos-Scenarios

# Talos Scenarios A collection of **602 unique agentic task scenarios/prompts** extracted from the Talos synthetic trace generation pipeline. These scenarios were used to generate the `DJLougen/Talos-kimi-k2.6-Hermes-synthetic` dataset via kimi-k2.6. ## Source Dataset These scenarios originate from: **DJLougen/Talos-kimi-k2.6-Hermes-synthetic** ## What's Inside | Field | Type | Description | |-------|------|-------------| | `scenario` | str | The user prompt / task description | | `category` | str | Domain tag: `coding`, `reasoning`, `creative`, `tool_use`, `science`, `history`, `business`, `philosophy`, `general` | | `complexity` | str | `low` / `medium` / `high` based on word count | | `requires_tools` | bool | Whether the scenario likely requires external tool/API access | | `word_count` | int | Length of the prompt in words | | `source` | str | Parent dataset reference | ## Statistics | Category | Count | |----------|-------| | general | 160 | | coding | 115 | | history | 115 | | science | 83 | | tool_use | 54 | | creative | 53 | | reasoning | 16 | | philosophy | 4 | | business | 2 | **Total:** 602 unique scenarios ## Usage ### Load with HuggingFace `datasets` ```python from datasets import load_dataset ds = load_dataset("DJLougen/Talos-Scenarios", split="train") print(ds[0]["scenario"]) ``` ### Filter by category ```python coding = ds.filter(lambda x: x["category"] == "coding") ``` ### Use for synthetic trace generation ```python for example in ds: prompt = example["scenario"] # Feed to your LLM to generate agentic traces ``` ## License MIT — synthetic data generated for training purposes. ## Contact Created by DJLougen as part of the Talos agentic trace curation pipeline.