AB-Law / AB-Law/QuietStories

[Feature]: Implement Provider Strategy Pattern for Multi-Model LLM Support

オープン
#53 コメント 0 件 リアクション 1 件 担当者 1 名 @Copilot に割り当て済み GitHub で見る
bug enhancement python
主要言語
Python
スター
0
フォーク
0
PR マージ指標
30日以内にマージされた PR はありません

説明

### Is your feature request related to a problem? Please describe.

Provider strategy pattern created for different LLM model families (GPT-5, GPT-4o, LM Studio, Ollama) with automatic routing based on model prefixes.
Scenario generation failing due to JSON parsing errors when using GPT-5 models. Character background isn't created correctly on gpt-5 models.

The scenario generation system fails when using GPT-5 models (gpt-5-nano, gpt-5-mini) due to malformed JSON responses that cannot be parsed by the fallback JSON extraction logic.
Error Details:
fallback
Root Cause: GPT-5 models generate JSON responses with formatting issues (unterminated strings, extra commas, malformed structures) that the current simple extraction logic cannot handle.

### Describe the solution you'd like

1. Provider Capabilities Matrix (backend/providers/capabilities.py)
Model family detection and capability mapping
API family routing (responses vs chat_completions vs openai_compatible)
2. Strategy Pattern Implementation
OpenAIGPT5Strategy: GPT-5 Responses API integration
OpenAIGPT4oStrategy: GPT-4o Chat Completions via LangChain
OpenAICompatibleStrategy: LM Studio/Ollama support
3. Provider Routing (backend/providers/openai.py)
Automatic strategy selection based on model name prefix
Unified ProviderResponse interface

Testing - Test suite needs to be updated for new architecture

### Describe alternatives you've considered

_No response_

### Additional context

_No response_

### Would you like to contribute this feature?

- [ ] Yes, I would like to implement this feature

コントリビューションガイド

コントリビューションガイドを開く

調査の方向性

The issue describes a strategy pattern for LLM providers. Start by examining backend/providers/capabilities.py and backend/providers/openai.py to understand the current structure. The goal is to create new strategy classes (OpenAIGPT5Strategy, OpenAIGPT4oStrategy, OpenAICompatibleStrategy) that handle different API families and JSON response formats. Testing will need updates to cover the new architecture.

索引モデルが issue の本文から書いたものです。

評価

技術スタック
ollama
領域
ai, backend
issue の種類
機能追加
難易度
5/5
見積もり時間
1週間以上
活発さ
停滞
明瞭さ
おおむね明確
初心者へのやさしさ
25/100

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。