Temporary fix：for o1-xx model need to covert systemMessage to aiMessage. #850
Conversation
| // eslint-disable-next-line @typescript-eslint/no-explicit-any | ||
| const chatModel = (ChainManager.chain as any).last.bound; | ||
| const modelName = chatModel.modelName || chatModel.model; | ||
| const isO1Model = modelName.startsWith("o1"); |
There was a problem hiding this comment.
Add a TODO: hack for o1 models, to be removed when they support system prompt
|
Added my comments. Just merged your other PR, this one needs a rebase. |
src/LLMProviders/chainManager.ts
Outdated
| @@ -317,12 +317,26 @@ export default class ChainManager { | |||
| this.validateChatModel(); | |||
| this.validateChainInitialization(); | |||
|
|
|||
| // eslint-disable-next-line @typescript-eslint/no-explicit-any | |||
| const chatModel = (ChainManager.chain as any).last.bound; | |||
There was a problem hiding this comment.
What's the difference between using this vs. "ChatModelManager.getChatModel()"?
There was a problem hiding this comment.
@logancyang At first, I plan to use ChatModelManager.getChatModel(), but I copied your code for consistency.
|
We can install the latest langchain openai and o1 models support streaming now: langchain-ai/langchainjs#7229 |
|
Merging now. With one caveat: Copilot Plus still doesn't support o1 models because its chain runner has more system messages. Right now I have no intention of enabling o1 models in Copilot Plus because
For the long term, o1 may support system message, let's see. |
related: #828