r2rpy/py/core/configs/lm_studio.toml

[app]
# LLM used for internal operations, like deriving conversation names
fast_llm = "lm_studio/llama-3.2-3b-instruct"

# LLM used for user-facing output, like RAG replies
quality_llm = "lm_studio/llama-3.2-3b-instruct"

# LLM used for ingesting visual inputs
vlm = "lm_studio/llama3.2-vision" # TODO - Replace with viable candidate

# LLM used for transcription
audio_lm = "lm_studio/llama-3.2-3b-instruct" # TODO - Replace with viable candidate

[embedding]
provider = "litellm"
base_model = "lm_studio/text-embedding-nomic-embed-text-v1.5"
base_dimension = nan
batch_size = 128
concurrent_request_limit = 2

[completion_embedding]
# Generally this should be the same as the embedding config, but advanced users may want to run with a different provider to reduce latency
provider = "litellm"
base_model = "lm_studio/text-embedding-nomic-embed-text-v1.5"
base_dimension = nan
batch_size = 128
concurrent_request_limit = 2

[agent]
tools = ["search_file_knowledge"]

[completion]
provider = "litellm"
concurrent_request_limit = 1

  [completion.generation_config]
  temperature = 0.1
  top_p = 1
  max_tokens_to_sample = 1_024
  stream = false