imageat 平台将 Qwen 3.8 27B Uncensored 模型封装为托管式 Chat API，开发者无需下载模型或规划 GPU，即可通过 OpenAI 兼容接口进行调用 1。该模型支持 262,144 个 token 的上下文预算，并提供可选的 thinking 模式，适用于长文档分析、代码审查和研究笔记整理等场景 1。
在模型调用配置方面，平台提供了两组采样参数建议：第一组为 temperature 1.0、top_p 0.95、top_k 20 和 repetition_penalty 1.0；第二组为 temperature 0.7、top_p 0.80、top_k 20 和 repetition_penalty 1.0 1。此外，平台提示了安全合规要求，并指出采样参数会影响输出质量 1。
The imageat platform has packaged the Qwen 3.8 27B Uncensored model as a hosted Chat API, enabling developers to access it via an OpenAI-compatible chat completion interface without downloading the model or provisioning GPUs 1. This deployment is suitable for applications such as long document analysis, code review, and organizing research notes 1.
The model supports a 262,144-token context budget and features an optional thinking mode enabled via the enable_thinking parameter 1. The platform also highlights the impact of safety compliance and sampling parameters on output quality, providing specific configurations for testing 1. These include a setup with a temperature of 1.0, top_p of 0.95, top_k of 20, and repetition_penalty of 1.0, alongside an alternative configuration with a temperature of 0.7, top_p of 0.80, top_k of 20, and repetition_penalty of 1.0 1.
评论
还没有评论，欢迎留下第一条。