cmem-plugin-llm 0.6.0

Interact with Large Language Models.

cmem-plugin-llm

Interact with Large Language Models.

This is a plugin for eccenca Corporate Memory. You can install it with the cmemc command line clients like this:

cmemc admin workspace python install cmem-plugin-llm

Create Embedding

This plugin contain the following capabilitities:

• create embeddings: it allows to create embeddings from an arbitrary data point or from specified data point paths (properties/columns). After being processed each data point receive two additional paths the _embedding path and the _embedding_source path.
  • The _embedding path contain the generated embedding.
  • The _embedding_source path contain the paths utilized for generate the embeddings in text format.

Parameters

• url: openAI compatible endpoint, default https://api.openai.com/v1
• model: embedding model, default text-embedding-3-small
• api_key: api key of the endpoint, default blank
• timout_single_request: the request timenout in milliseconds, default 10000
• entries_processing_buffer: number of processed entries per request, default 1000
• embedding_paths: specify which paths should be used for embedding generation, default all
• embedding_output_text: output path that will contain the embedding text, default _embedding_source
• embedding_output_path: output path that will contain the generated embedding, default _embedding

Execute Instruction

This plugin contain the following capabilitities:

• execute instruction: it allows to execute an LLM instruction over a given list of entities. After being processed each entity receive one additional path, the _instruction_output.
  • The _instruction_output path contains the output of the executed instruction over the entitiy.

Parameters

• url: openAI compatible endpoint, default https://api.openai.com/v1
• model: embedding model, default gpt-o
• api_key: api key of the endpoint, default blank
• timout_single_request: the request timenout in milliseconds, default 10000
• instruction_template: the instruction template default instruct template: Write a paragraph about this entity: ${entity}
• prompt_template: the prompt template default prompt template:

              [{
                  "role": "developer",
                  "content": "You are a helpful assistant."
              },
              {
                  "role": "user",
                  "content": "${instruct}"
              }]
  

• instruct_output_path: output path that will contain the output of the executed instruction, default _instruction_output