docs: add model fallback docs and update config example
Add cross-provider model_fallbacks section to README.md and README.ja.md. Update config.example.json with provider, model_fallbacks, and task_reminder_interval fields. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
parent
ae2541e9b8
commit
c67c4a9eb3
3 changed files with 64 additions and 1 deletions
30
README.ja.md
30
README.ja.md
|
|
@ -621,6 +621,36 @@ HEARTBEAT_OK 応答 ユーザーが直接結果を受け取る
|
||||||
- `PICOCLAW_HEARTBEAT_ENABLED=false` で無効化
|
- `PICOCLAW_HEARTBEAT_ENABLED=false` で無効化
|
||||||
- `PICOCLAW_HEARTBEAT_INTERVAL=60` で間隔変更
|
- `PICOCLAW_HEARTBEAT_INTERVAL=60` で間隔変更
|
||||||
|
|
||||||
|
### モデルフォールバック
|
||||||
|
|
||||||
|
`model_fallbacks` はプライマリモデルがリトライ可能なエラー(レート制限、サーバーエラー等)を返した時に起動するフォールバックチェーンです。**異なるプロバイダー**をターゲットにでき、プロバイダーは初回使用時に遅延生成・キャッシュされます。
|
||||||
|
|
||||||
|
```json
|
||||||
|
{
|
||||||
|
"agents": {
|
||||||
|
"defaults": {
|
||||||
|
"provider": "vllm",
|
||||||
|
"model": "MiniMax-M1-80k",
|
||||||
|
"model_fallbacks": ["openai/gpt-4o", "anthropic/claude-sonnet-4-5-20250929"]
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"providers": {
|
||||||
|
"vllm": {
|
||||||
|
"api_key": "minimax-key",
|
||||||
|
"api_base": "https://api.minimax.io/v1"
|
||||||
|
},
|
||||||
|
"openai": {
|
||||||
|
"auth_method": "oauth"
|
||||||
|
},
|
||||||
|
"anthropic": {
|
||||||
|
"api_key": "sk-ant-xxx"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
フォーマットは `provider/model`。`/` がない場合はプライマリプロバイダーが使われます(同一プロバイダー内フォールバック)。失敗したプロバイダーはクールダウン期間に入り、連続タイムアウトを防ぎます。
|
||||||
|
|
||||||
### 基本設定
|
### 基本設定
|
||||||
|
|
||||||
1. **設定ファイルの作成:**
|
1. **設定ファイルの作成:**
|
||||||
|
|
|
||||||
30
README.md
30
README.md
|
|
@ -689,6 +689,36 @@ PicoClaw routes providers by protocol family:
|
||||||
|
|
||||||
This keeps the runtime lightweight while making new OpenAI-compatible backends mostly a config operation (`api_base` + `api_key`).
|
This keeps the runtime lightweight while making new OpenAI-compatible backends mostly a config operation (`api_base` + `api_key`).
|
||||||
|
|
||||||
|
### Model Fallbacks
|
||||||
|
|
||||||
|
`model_fallbacks` defines a fallback chain that activates when the primary model returns a retriable error (rate limit, server error, etc.). Fallbacks can target **different providers** — the provider is lazily created and cached on first use.
|
||||||
|
|
||||||
|
```json
|
||||||
|
{
|
||||||
|
"agents": {
|
||||||
|
"defaults": {
|
||||||
|
"provider": "vllm",
|
||||||
|
"model": "MiniMax-M1-80k",
|
||||||
|
"model_fallbacks": ["openai/gpt-4o", "anthropic/claude-sonnet-4-5-20250929"]
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"providers": {
|
||||||
|
"vllm": {
|
||||||
|
"api_key": "minimax-key",
|
||||||
|
"api_base": "https://api.minimax.io/v1"
|
||||||
|
},
|
||||||
|
"openai": {
|
||||||
|
"auth_method": "oauth"
|
||||||
|
},
|
||||||
|
"anthropic": {
|
||||||
|
"api_key": "sk-ant-xxx"
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
The format is `provider/model`. If no `/` is present, the primary provider is assumed (same-provider fallback). Failed providers enter a cooldown period to avoid repeated timeouts.
|
||||||
|
|
||||||
<details>
|
<details>
|
||||||
<summary><b>Zhipu</b></summary>
|
<summary><b>Zhipu</b></summary>
|
||||||
|
|
||||||
|
|
|
||||||
|
|
@ -3,10 +3,13 @@
|
||||||
"defaults": {
|
"defaults": {
|
||||||
"workspace": "~/.picoclaw/workspace",
|
"workspace": "~/.picoclaw/workspace",
|
||||||
"restrict_to_workspace": true,
|
"restrict_to_workspace": true,
|
||||||
|
"provider": "zhipu",
|
||||||
"model": "glm-4.7",
|
"model": "glm-4.7",
|
||||||
|
"model_fallbacks": ["openai/gpt-4o"],
|
||||||
"max_tokens": 8192,
|
"max_tokens": 8192,
|
||||||
"temperature": 0.7,
|
"temperature": 0.7,
|
||||||
"max_tool_iterations": 20
|
"max_tool_iterations": 20,
|
||||||
|
"task_reminder_interval": 5
|
||||||
}
|
}
|
||||||
},
|
},
|
||||||
"channels": {
|
"channels": {
|
||||||
|
|
|
||||||
Loading…
Add table
Reference in a new issue