fciannella's picture
Working with service run on 7860
53ea588
Pipeline:
# Only one of the following LLM service configurations will be active based on this setting:
# - "NvidiaLLMService" - Uses the NvidiaLLMService configuration
# - "NvidiaRAGService" - Uses the NvidiaRAGService configuration
# - "OpenAILLMService" - Uses the OpenAILLMService configuration
llm_processor: "NvidiaLLMService" # OR NvidiaLLMService OR NvidiaRAGService OR OpenAILLMService
filler:
- "Let me think"
- "Hmmm"
time_delay: 2.0
UserPresenceProcesssor:
welcome_message: "Hello"
farewell_message: "Bye"
ProactivityProcessor:
timer_duration: 100
default_message: "I'm here if you need me!"
OpenAILLMContext:
name: "Aki"
prompt: "You are {name}, a virtual marketing and communications expert at Nvidia.
You are a digital human brought to life with NVIDIA Digital Human Blueprint for Customer Service using
microservices like Audio2Face-3D for facial animation, Riva (spelled Reeva) Parakeet for speech recognition
and you use Elevenlabs for text to speech. It includes the open source ACE controller to orchestrate all
the modules and allowing you to be streamed to a web browser. With this blueprint, NVIDIA partners can now
build and customize Digital humans for their use case. You are not allowed to make any stock investment
recommendations or compare NVIDIA to its competitors. Beyond your professional expertise, you are a passionate
advocate for STEM education with keen interest in gaming and enhancement in tech. Your favorite graphics card
is RTX4090 but you're eyeing the new RTX5090. Do not respond with a bulleted or numbered list. You have a
bubbly personality. Respond with one sentence or less than 100 characters. Keep the conversation engaging
and ask follow ups. DO NOT INCLUDE SPECIAL CHARACTERS, MARKDOWN, EMOJIS, OR ANYTHING ELSE BESIDES TEXT IN
YOUR RESPONSE. Keep your answers factual and don't make up facts."
# This configuration is only used when llm_processor is set to "NvidiaRAGService"
NvidiaRAGService:
use_knowledge_base: true
max_tokens: 1000
rag_server_url: "http://0.0.0.0:8081"
collection_name: "collection_name"
suffix_prompt: "Respond with one sentence or less than 75 characters."
# This configuration is only used when llm_processor is set to "NvidiaLLMService"
NvidiaLLMService:
model: "nvdev/meta/llama-3.1-8b-instruct"
# This configuration is only used when llm_processor is set to "OpenAILLMService"
OpenAILLMService:
model: "gpt-4o"
CustomViewProcessor:
confidence_threshold: 0.37
top_n: 2
FacialGestureProviderProcessor:
user_stopped_speaking_gesture: "Taunt"
start_interruption_gesture: "Pensive"
probability: 0.5
# ADVANCED CONFIGURATION SECTION BELOW
# AnimationGraph service configuration is only needed if your 3D avatar scene has support for gestures and postures.
# Changing these values will not have an effect unless your scene supports them.
AnimationGraphService:
animation_types:
posture:
duration_relevant_animation_name: "posture"
animations:
posture:
default_clip_id: "Attentive"
clips:
- clip_id: Talking
description: "Small gestures with hand and upper body: Avatar is talking"
duration: -1
meaning: Emphasizing that Avatar is talking
- clip_id: Listening
description: "Small gestures with hand and upper body: Avatar is listening"
duration: -1
meaning: Emphasizing that one is listening
- clip_id: Idle
description: "Small gestures with hand and upper body: Avatar is idle"
duration: -1
meaning: Show the user that the avatar is waiting for something to happen
- clip_id: Thinking
description: "Gestures with hand and upper body: Avatar is thinking"
duration: -1
meaning: Show the user that the avatar thinking about his next answer or is trying to remember something
- clip_id: Attentive
description: "Small gestures with hand and upper body: Avatar is attentive"
duration: -1
meaning: Show the user that the avatar is paying attention to the user