XunziALLM 3.0
发布筹备中Release in preparation
荀子系列新一代模型,完成预训练、指令微调、GRPO 和 Agentic RL 四阶段训练。模型版本与权重信息如下。
A new generation of the Xunzi model family, trained through pre-training, SFT, GRPO and Agentic RL.
- 版本 / Version
- Agentic RL · iter_0000999
- 基座 / Base checkpoint
- Qwen3.5-9B → merged_grpo_2400
- 训练链 / Training
- Pre-training → SFT → GRPO → Agentic RL
- 参数量 / Parameters
- 约 / Approximately 9.41B
- 模型架构 / Architecture
- Qwen3_5ForConditionalGeneration
- 权重 / Weights
- bfloat16 · 4 shards · ≈18.8 GB
下载地址与许可证待公布。公开发布后,将在这里提供权重获取、使用说明和引用信息。Download links and license will be announced. Model access, usage instructions and citation information will be added here.
了解荀子 3.0 项目About the Xunzi 3.0 project
计划开放Planned release唐诗生成模型Tang Poetry Generation
面向唐诗创作的专项模型。版本、使用说明及下载地址将在发布后补充。
Model version, usage details and downloads will be added upon release.
计划开放Planned release宋词生成模型Song Ci Generation
面向宋词创作的专项模型。版本、使用说明及下载地址将在发布后补充。
Model version, usage details and downloads will be added upon release.
计划开放Planned release骈文生成模型Parallel Prose Generation
面向骈文创作的专项模型。版本、使用说明及下载地址将在发布后补充。
Model version, usage details and downloads will be added upon release.
This section will expand with more Xunzi models. Existing model access links are listed below.