Devcrocod/langchain4j #476
Conversation
langchain4j.json
Outdated
| ], | ||
| "link": "https://github.com/langchain4j/langchain4j", | ||
| "init": [ | ||
| "when (\"$model\") {\n\"bedrock\" -> USE { dependencies(\"dev.langchain4j:langchain4j-bedrock:$v\") }\n\"anthropic\" -> USE { dependencies(\"dev.langchain4j:langchain4j-anthropic:$v\") }\n\"azure\" -> USE { dependencies(\"dev.langchain4j:langchain4j-azure-open-ai:$v\") }\n\"googleai\" -> USE { dependencies(\"dev.langchain4j:langchain4j-google-ai-gemini:$v\") }\n\"vertexai\" -> USE { dependencies(\"dev.langchain4j:langchain4j-vertex-ai-gemini:$v\") }\n\"huggingface\" -> USE { dependencies(\"dev.langchain4j:langchain4j-hugging-face:$v\") }\n\"ollama\" -> USE { dependencies(\"dev.langchain4j:langchain4j-ollama:$v\") }\n\"openai\" -> USE { dependencies(\"dev.langchain4j:langchain4j-open-ai:$v\") }\n\"openaiofficial\" -> USE { dependencies(\"dev.langchain4j:langchain4j-open-ai-official:$v\") }\nelse -> error(\"Not recognized dependency '$model'. Use follow models: `bedrock`, `anthropic`, `azure`, `googleai`, `vertexai`, `huggingface`, `ollama`, `openai`, `openaiofficial`\")\n}\nUSE { import(\"dev.langchain4j.model.$model.*\") }" |
There was a problem hiding this comment.
Probably define a map?
val modelToArtifact = mapOf ("bedrock" to "bedrock", "openaiofficial" to "open-ai-official", ...)
In this case there'll be just one USE call.
There was a problem hiding this comment.
But even better to maintain these models in a separate integration library
There was a problem hiding this comment.
Do you mean creating a separate descriptor for each artifact?
I’d prefer not to do that, as it would significantly increase the number of descriptors
There was a problem hiding this comment.
No, I mean putting this logic into an artifact (a single one). It could be placed in our internal maven repo. In this case you may add tests for the integration, add additional features easily and so on, so it's a more reliable approach. Descriptor then will stay the same but without this init code.
Integration classes accept options:
https://github.com/Kotlin/dataframe/blob/79900825169933d8b644c709a3d6601a29bf9fa6/dataframe-jupyter/src/main/kotlin/org/jetbrains/kotlinx/dataframe/jupyter/Integration.kt#L64
So you may access model name from there.
Lets-plot integration underwent the same transformation
There was a problem hiding this comment.
Alright, I understand.
Nevertheless, I’m not going to do it))
Unfortunately, at the moment I can’t guarantee long-term support for langchain4j, so I don’t think it’s worth creating a new entity, even if it simplifies integration with jupyter. However, if a stable version is released and Kotlin support becomes reliable, then we can easily add such an artifact — and the descriptor for users will remain the same
Description
add langchain4j descriptor
parameter
modeldefines a specific dependency:Type of change
Checklist for new descriptors (remove if not applicable):
descriptionfield accurately describes the library and that thelinkpoints to the current repository.propertiesfield contains all versions of Maven dependecies used in thedependenciessection.dependenciessection use version variables correctly.rendererssection has appropriate rendering rules.