错误，怎么解决{"param":null,"type":"invalid_request_error","code":"invalid_request_error","message":"Error from provider (Console Go): Upstream request failed: [invalid_request_error] This model's maximum context length is 1048576 tokens. However, you requested 1048728 tokens (664728 in the messages, 384000 in the completion). Please reduce the length of the messages or completion."}

[wangzczgbaba-source]

{"param":null,"type":"invalid_request_error","code":"invalid_request_error","message":"Error from provider (Console Go): Upstream request failed: [invalid_request_error] This model's maximum context length is 1048576 tokens. However, you requested 1048728 tokens (664728 in the messages, 384000 in the completion). Please reduce the length of the messages or completion."}
运行错误，怎么解决？？？

[zoahdev]

That error is the model context window being exceeded, not a config bug: messages (664,728 tokens) + requested completion (384,000) = 1,048,728 > the 1,048,576 limit. The completion request (384K) is what pushed it over ? that's an unusually large max_tokens.

Quick checks:

1. If you set max_tokens (or a reasoning budget) very high, lower it ? the completion side is half the total.
2. The 664K messages side means the session is huge: start a new session, or trim the context. Our large-files profile check can flag oversized session logs before they hit the cap (Bug: session search crashes with `RangeError: Invalid string length` on large sessions (root cause + fix) #1859 class):
   npx dsh-plugin-doctor --profile ~/.dsh/profiles/web --json ? large-files
3. If this repeats mid-session, the retrieval-efficiency contract helps keep context bounded: https://zoahdev.github.io/dsh-tutorials/zh-15-retrieval-efficiency.html

Post the provider name and the exact request that failed if you want a deeper look.

[denial123789]

这个错误可以直接按预算来读：

664,728 messages + 384,000 completion = 1,048,728
model window                         = 1,048,576
overflow                            =       152

所以不是 API key 问题，而是这次请求的“历史输入 + 最大输出预留”超过了模型上下文。建议按这个顺序处理：

1. 在当前会话的模型设置里先把最大输出 token 从 384000 降到更实际的值，例如 128000，不要只减 152，应该留出后续消息和工具调用的余量。
2. 如果当前 profile 提供 /compact，先执行一次 /compact，成功后只重试原请求一次。
3. 如果没有可压缩历史或仍然超限，新建会话，只带目标、关键决策、文件路径、未解决错误和下一步，不要复制整个旧对话。

还需要区分两个同名配置：模型适配器的 maxTokens 是普通回复的输出上限；dsh-compaction-basic 里的 maxTokens 是摘要调用的上限。修改后可以用 dsh --profile web --dump-config 检查实际生效的 provider/model、contextWindow 和 maxTokens。

我根据官方 adapter、overflow recovery 和 /compact 源码整理了完整的英文排查页，包含配置示例与验证清单：

https://github.com/sandbaseai/deepseek-harness-handbook/blob/agent/context-overflow/docs/en/troubleshooting/context-window-exceeded.md

对应文档 PR： sandbaseai/deepseek-harness-handbook#24

[wangzczgbaba-source]

是处理/goal 目标的过程中，多轮执行后出现的错误。提示是 运行错误400 ，使用的是opencode go v4f 模型，我已经重开了对话，等会我在回头重新按照你们的方法去试试

[IronsDu]

我也在发出指令后，它的思考过程中遇到了此错误。 没有找到哪里可以compact。

[IronsDu]

新开启一个会话，让它阅读 https://github.com/sandbaseai/deepseek-harness-handbook/blob/agent/context-overflow/docs/en/troubleshooting/context-window-exceeded.md ，并让它自己解决那个错误。 就搞定了。