ning423/Hermes-OmniForge-Qwen36-27B-full-v0.3.0-unsloth

Hermes OmniForge Qwen3.6-27B数据集v0.3.0是一个合成的SFT数据集，支持Unsloth-ready导出。数据集包含多种任务类型，如文本生成、视觉问答、图像文本到文本等。数据集分为训练集、验证集和测试集，分别有150,000、5,000和5,000行数据。数据集的组成部分包括lambda_hermes_agent_reasoning_traces、gpt55_hermes_synthetic、repo_coding_terminal等，每个部分有不同的比例。数据集支持多种使用方式，包括与Hugging Face Datasets和Unsloth的集成。数据集的每一行都遵循特定的JSON模式，包含id、component、source_dataset、source_license、source_url、task_family、benchmark_targets、tools、messages、media、loss_mask和quality等字段。数据集还包含视觉媒体数据，但由于实际源图像不可用，使用了占位符PNG文件。数据集还提供了验证脚本和报告，以确保数据的完整性和正确性。最后，数据集的所有行都是合成的，并包含源元数据以供归因和审计。

The Hermes OmniForge Qwen3.6-27B Dataset v0.3.0 is a synthetic SFT dataset with Unsloth-ready exports. It includes various task categories such as text-generation, visual-question-answering, and image-text-to-text. The dataset is divided into train, validation, and test sets with 150,000, 5,000, and 5,000 rows respectively. The dataset components include lambda_hermes_agent_reasoning_traces, gpt55_hermes_synthetic, repo_coding_terminal, etc., each with different proportions. The dataset supports multiple usage methods, including integration with Hugging Face Datasets and Unsloth. Each row in the dataset follows a specific JSON schema, containing fields such as id, component, source_dataset, source_license, source_url, task_family, benchmark_targets, tools, messages, media, loss_mask, and quality. The dataset also includes vision media data, but due to the unavailability of real source images, placeholder PNG files are used. The dataset provides validation scripts and reports to ensure data integrity and correctness. Finally, all rows in the dataset are synthetic and include source metadata for attribution and auditability.