尝试Auto-coder.chat使用星河社区AIStudio部署的几个大模型：文心4.5-21b、Deepseek r1 70b、llama 3.1 8b

星河社区可以部署模型，比如文心4.5-21b、Deepseek r1 70b、llama 3.1 8b

先上结论：文心4.5-21b、Deepseek r1 70b、llama 3.1 8b等这三个可以在星河社区直接部署出来的模型，没法直接在Auto-Coder用。这遗憾啊！

Auto-Coder.chat是非常棒的AI Agent编程助手，具体安装配置参见：https://skywalk.blog.csdn.net/article/details/148075062?spm=1011.2415.3001.5331

部署热门模型库

部署deepseek r1 70b

deepseek r1 70b，可以使用，但是光输出不干活：

复制代码

This setup provides a basic REST API with CRUD operations for managing books using Node.js and Express.conversation tokens: 45612 (conversation round: 63)
<think>
Okay, so I need to figure out how to help the user with their task. Let me start by understanding what they're asking for.

First, the user provided a system message about saving message IDs using a tool called conversation_message_ids_write. It seems like they want to manage or clean up their conversation history by deleting specific messages. They mentioned that the conversation is too long, so maybe they're looking to free up space or organize their chat.

I remember that there are tools available for handling message IDs. The user specifically named the conversation_message_ids_write tool, which I think is used to save message IDs that need to be deleted. So my first step should be to use this tool to write down those IDs.

But wait, do I have access to the list of message IDs? It doesn't look like the user provided any specific IDs here. Without knowing which messages they want to delete, I can't proceed effectively. I might need to ask them for the exact IDs or ranges of messages they want to remove.

Alternatively, if I'm supposed to generate these IDs automatically, I'd need a way to fetch them from wherever the conversation data is stored. But since I don't have that context, maybe I should prompt the user to provide more details about which parts of the conversation they want to delete.

I also recall that after saving the message IDs, there might be another tool or process to actually perform the deletion. So once I have the IDs saved, the next step would involve using a different tool to handle the deletion itself.

In summary, my immediate action should be to use the conversation_message_ids_write tool, but I need the necessary information from the user first. Without the message IDs, I can't proceed further. I'll go ahead and ask them for the specific IDs they want to delete.
</think>

I apologize for the earlier response. To properly assist you, I'll now use the appropriate tools.

<thinking>
The user wants to save message IDs using the conversation_message_ids_write tool but hasn't provided any IDs yet. I need to prompt them for this information.
</thinking>

To proceed, please provide the message IDs that need to be saved for deletion.conversation tokens: 46169 (conversation round: 65)
<think>
Alright, so I need to help the user by using the conversation_message_ids_write tool to save message IDs that they want to delete. Let me break this down step by step.

First, the user mentioned that their conversation is too long and they want to use a specific tool to save the message IDs for deletion. That tells me the main goal here is to manage or clean up a lengthy chat history.

I should figure out how the conversation_message_ids_write tool works. Typically, such tools require specific parameters like the IDs of the messages to be deleted. Without knowing these IDs, I can't proceed because each message in a conversation platform usually has a unique identifier.

Since the user hasn't provided any message IDs yet, my first step is to prompt them for this information. I need to ask clearly and concisely so they know exactly what's needed. It's important to make sure they understand that without these IDs, the tool can't function properly.

I should also consider if there are any edge cases, like if the user wants to delete all messages or a specific range. But since they specifically mentioned using the conversation_message_ids_write tool, I'll stick to that approach unless they clarify otherwise.

In summary, my next action is to request the message IDs from the user so I can use the appropriate tool to assist them effectively.
</think>

To address your request, I need the specific message IDs you want to delete. Could you please provide them?conversation tokens: 46573 (conversation round: 67)
<think>
Okay, so I need to help the user by saving the message IDs of their conversation so they can delete them later. The user mentioned that the conversation is too long, so using the conversation_message_ids_write tool makes sense here.

First, I'll start by acknowledging the task and then figure out how to extract those message IDs. Since the conversation is provided in a log format, I can use some shell commands to parse through it.

部署文心4.5-21b和llama 3.1 8b

文心4.5-21b和llama 3.1 8b ，输出报错：

复制代码

╭─ 🔥 Error ───────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ Error: Connection failed. Error: ))                                                                                  │
│                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                         │
│   File "/home/skywalk/py312/lib/python3.12/site-packages/openai/_base_client.py", line 1001, in request              │
│     raise APIConnectionError(request=request) from err                                                               │
│ openai.APIConnectionError: Connection error.                                                                         │
│                                                                                                                      │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭────────────────────────────────────────────────── 🔥 System Error ───────────────────────────────────────────────────╮
│ FATAL ERROR:                                                                                                         │
│ 'NoneType' object has no attribute 'input_tokens_count'                                                              │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
──────────────────────────────────────────────── Agentic Edit Finished ─────────────────────────────────────────────────
 1. An error occurred: AttributeError - 'NoneType' object has no attribute 'input_tokens_count'

经检查，llama是还没有启动服务导致

启动后，还是这样的报错

使用qwen3 235b模型，也是一样的报错

复制代码

│   File "/home/skywalk/py312/lib/python3.12/site-packages/openai/_base_client.py", line 1001, in request              │
│     raise APIConnectionError(request=request) from err                                                               │
│ openai.APIConnectionError: Connection error.                                                                         │
│                                                                                                                      │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭────────────────────────────────────────────────── 🔥 System Error ───────────────────────────────────────────────────╮
│ FATAL ERROR:                                                                                                         │
│ 'NoneType' object has no attribute 'input_tokens_count'                                                              │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
──────────────────────────────────────────────── Agentic Edit Finished ─────────────────────────────────────────────────
 1. An error occurred: AttributeError - 'NoneType' object has no attribute 'input_tokens_count'

总结

结论就是，除了deepseek 70b模型可以被调用，其它模型连调用都不行。

deepseek r1 70b模型可以被调用，但是无法正常进行function call 函数调用，也就无法写代码。

后期准备用Agent来处理一下，争取解决这个报错问题

复制代码

An error occurred: AttributeError - 'NoneType' object has no attribute 'input_tokens_count'

2025.11.7日补充

今天终于把文心21b模型调通了，可以在auto-coder下使用

所做的操作就是把openai升级到最新版了