This document details the AI services available in the ComradeAI, emphasizing their functionalities, strengths, and specific parameters.
Since version 0.18.7 a simplified usage of Services can be used. Take a look at AvailableServices.md.
In ComradeAI, the choice and configuration of an AI agent are crucial. This is managed through the agent
variable, which are then used as part of the routingStrategy
, and requestAgentConfig
dictionary.
The agent
variable specifies which AI agent you intend to use. This should be set to the agent’s identifier within the ComradeAI framework.
The routingStrategy
is a crucial component in directing your requests to the correct AI agent. It includes the agent
variable as part of its configuration, especially for ‘auto’ and ‘direct’ strategies.
agent = "gpt-3.5-turbo-1106"
routingStrategy = RoutingStrategy("direct", agent)
# Means you know which agent you need and there is no need for Mycelium to auto-define it.
requestAgentConfig = {
"max_tokens": 100,
"temperature": 0.7
# Add other parameters as needed
}
The service is incapsulated by “OpenAI_DALLE3” agent.
Generates high-quality, creative images from textual descriptions.
size
(string): Image resolution, choices: “1024x1024”, “1792x1024”, “1024x1792”.style
(string): Image style, choices: “vivid”, “natural”.n
(integer): Number of images to generate.quality
(string): Image quality, choices: “standard”, “hd”.The service is incapsulated by “Google_GeminiProVisoin” agent.
Handles multimodal tasks including text generation, image analysis, and video processing.
model
(string): Specific model selection. Supports “gemini-pro” for text to text generation and “gemini-pro-vision” for text/image(s)/video(s) to text generation.max_output_tokens
(integer): Limits response size, usually 10-1000.temperature
(float): Controls the randomness in responses, with a range from 0.1 to 1.0.top_p
(float): Influences the diversity of generated text, range: 0.1 - 1.0.top_k
(integer): Narrows the set of tokens to be considered for responses, typically from 1 to 40.stop_sequences
(string/array): Defines specific sequences where the model will stop generating further tokens.The services are incapsulated by “OpenAI_GPT_Completions” agent.
The ComradeAI framework integrates three advanced models from OpenAI’s GPT series, each designed for specific text generation tasks and offering unique capabilities.
GPT 3.5 Turbo is designed for high-speed, cost-effective text generation, maintaining a balance between performance and accuracy.
model
: “gpt-3.5-turbo-1106”.max_tokens
(integer): Limits the output length, typically 10-4096 tokens.temperature
(float): Controls randomness in output, range: 0.0 - 1.0.top_p
(float): Influences diversity, range: 0.1 - 1.0.frequency_penalty
(float): Reduces repetition, range: 0.0 - 2.0.presence_penalty
(float): Encourages new concepts, range: 0.0 - 2.0.stop
(string/array): Specifies tokens at which the model should stop generating further content.GPT 4 excels in understanding and generating both human language and code, making it versatile for a wide range of applications.
model
(string): “gpt-4-1106-preview”.max_tokens
(integer): Limits output length, up to 4096 tokens.temperature
(float): Adjusts creativity, range: 0.0 - 1.0.top_p
(float): Controls diversity, range: 0.1 - 1.0.frequency_penalty
(float): Decreases repetition, range: 0.0 - 2.0.presence_penalty
(float): Encourages novelty, range: 0.0 - 2.0.stop
(string/array): Specifies tokens at which the model should stop generating further content.GPT 4 Vision is a multimodal model capable of processing and generating both text and images, bridging the gap between visual and textual data.
model
(string): “gpt-4-vision-preview”.max_tokens
(integer): Limits output length, up to 4096 tokens.temperature
(float): Adjusts creativity, range: 0.0 - 1.0.top_p
(float): Controls diversity, range: 0.1 - 1.0.frequency_penalty
(float): Decreases repetition, range: 0.0 - 2.0.presence_penalty
(float): Encourages novelty, range: 0.0 - 2.0.stop
(string/array): Specifies tokens at which the model should stop generating further content.The service is incapsulated by “Meta_LLaMa2” agent.
Efficient in diverse text processing tasks, offering unique capabilities.
temperature
(float): Adjusts creativity level, range: 0.0 - 1.0.max_tokens
(integer): Total token count limit, typically up to 4096.max_response_words
(integer): Maximum word count for responses. Default is 100 words.The service is incapsulated by “YandexGPT2-FULL” agent.
Specialized in generating and processing text in Russian, with a nuanced understanding of the language.
temperature
(float): Controls creativity and style, range: 0.0 - 1.0.maxtokens
(integer): Total token count limit, usually up to 8000.The service is incapsulated by “groot” agent.
Used to verify successful connectivity to the Mycelium network.