Skip to content

Next Gen UI Embedded Llama Stack Server Inference

This module is part of the Next Gen UI Agent project.

Module Category Module Status

Support for LLM inference using Embedded Llama Stack server.

Provides

  • LlamaStackEmbeddedAsyncAgentInference to use LLM hosted in embedded Llama Stack server, started from provided Llama Stack config file.
  • init_inference_from_env method to init Llama Stack inference (remote or embedded) based on environment variables
    • INFERENCE_MODEL - LLM model to use - inference is not created if undefined (default value can be provided as method parameter)
    • LLAMA_STACK_HOST - remote LlamaStack host - if defined then it is used with LLAMA_STACK_PORT to create remote LlamaStack inference
    • LLAMA_STACK_PORT - remote LlamaStack port - optional, defaults to 5001
    • LLAMA_STACK_URL - remote LlamaStack url - if LLAMA_STACK_HOST is not defined, but this url is defined, then it is used to create remote LlamaStack inference
    • LLAMA_STACK_CONFIG_FILE - path to embedded LlamaStack server config file, used only if no remote LlamaStack is configured (default value can be provided as method parameter)
  • examples/llamastack-ollama.yaml example of the LlamaStack config file to use LLM from Ollama running on localhost (with model also taken from INFERENCE_MODEL env variable).

Installation

pip install -U next_gen_ui_llama_stack_embedded

Example

Instantiation of LlamaStackEmbeddedAsyncAgentInference

from next_gen_ui_llama_stack_embedded import LlamaStackEmbeddedAsyncAgentInference

config_file = "example/llamastack-ollama.yaml"
model = "llama3.2:latest"

inference = LlamaStackEmbeddedAsyncAgentInference(config_file, model)

# init UI Agent using inference

Inference initialization from environment variables

from next_gen_ui_llama_stack_embedded import init_inference_from_env

# default model used if env variable is not defined
INFERENCE_MODEL_DEFAULT = "granite3.3:2b"

inference = init_inference_from_env(default_model=INFERENCE_MODEL_DEFAULT)

if (inference):
    # init UI Agent using inference
else:
    print("Inference not initialized because not configured in env variables")