Xuchen Pan

AgentScope 主要面向的是 multi-Agent 场景，其中各个Agent之间的交互的消息需要满足一定的格式信息，而流式传输会打破这一规则，因此目前的 AgentScope 没有提供 agent 级别的流式输出。如果仅仅是希望每个 Agent 对用户显示的内容为流式显示，可以通过修改 agent 内部的模型调用以及 speak 方法，让模型返回流式生成器并使用 speak 依次打印。 我们在未来的版本中可能会考虑加入一类特殊的支持流式输出的 Agent 对外提供流式 API 服务，但 AgentScope 内其他的 Agent 为了保证输入输出格式仍会保持现有的一次性输出。

Because the gRPC server relies on the `RpcEnv`,`rpc_env.py` and the modules it depends on can't be moved to `examples`, all other modules have been moved to `examples/environments`

To be discussed 1. the name of distributed functions. In the current version, `async_func/sync_func` may be confused with `async` in Python `asyncio`. `unblocking_func / blocking_func` may be a better name....

As a multi-agent platform, AgentScope can use third-party agents by implementing a sub-class of `AgentBase` and using the third-party agent in the `reply` method. Additionally, AgentScope already has an RPC-based...

You can find some examples in `test_agent_server_management_funcs` of `tests/rpc_agent_test.py`. The wrong version of AgentScope may cause the reported error. Please reinstall AgentScope from the latest source code `pip install -e...

The main reason seems to be the API concurrency limitation. For remote APIs, providers generally limit the access frequency, which results in the inability to further increase the concurrency. In...

Sorry for the late reply. The `--is-human` was accidentally removed in a past refactor. We will fix it later.

ray默认情况应该会启动cpu数量个worker，日志中的现象可能是ray没有正确识别环境中cpu数量导致的

当前 Ray Object Store 确实支持磁盘映射，但是仅适用于存储于 Object Store 中的数据，如果是 Ray Actor 或 Task 本身超过了内存限制则无法处理，详细情况建议参考 Ray 的官方文档

能否提供更详细的模型部署信息 使用 openai_chat 类型的模型配置要求你的模型服务已经正常启动且服务接口兼容 OpenAI