seperating model settings for different prompts.

[arvin8]

It appears that there are 2 different types of prompts used. prompts used in extraction tasks and prompts used in conceptualization tasks.
While extraction prompts might benefit from a lower temperature setting, conceptualization prompts might be better with a higher temperature setting. You might want to consider seperating the settings for different type of prompts.

[gzone0111]

That is a correct observation. You can currently achieve this by using two separate instances of the LLMGenerator interface with different configurations.

For example:

from atlas_rag.llm_generator.generation_config import GenerationConfig
# 1. Configure for Triple Extraction (Default/Low Temperature)
# Implicitly uses default config for LLM generation
triple_generator = LLMGenerator(client=client, model_name="Qwen/Qwen2.5-7B-Instruct")

kg_extractor = KnowledgeGraphExtractor(model=triple_generator, config=kg_extraction_config)

# ... Function call for triple extraction ...
# 2. Configure for Concept Generation (Higher Temperature)
# Create a specific config with higher temperature/creativity
gen_config = GenerationConfig(temperature=0.7, max_tokens=32768)

concept_generator = LLMGenerator( client=client, model_name="Qwen/Qwen2.5-7B-Instruct", default_config=gen_config )

# Swap the model generator in the extractor
kg_extractor.model = concept_generator

# ... Subsequent function call for concept extraction ...

Note: Please be aware that swapping the model does not update the output file configuration. The output directory and filename are determined during the first run_extraction call, so the concept extraction results will continue using the initial path unless you manually re-initialize the extraction config.