Olares

Appearance

Configuration Guideline for Model ​

Every application chart for LLM should include a modelConfig.yaml file in the root directory. This file provides the essential information required by an LLM.

Here's an example of what a modelConfig.yaml file might look like:

modelConfig.yaml Example
Yaml
source_url: https://huggingface.co/TheBloke/Yarn-Mistral-7B-128k-GGUF/resolve/main/yarn-mistral-7b-128k.Q4_K_M.gguf
id: yarnmistral7b
object: model
name: Yarn Mistral 7B Q4
version: '1.0'
description: Yarn Mistral 7B is a language model for long context and supports a 128k token context window.
format: gguf
settings:
  ctx_len: 4096
  prompt_template: |-
    {prompt}
parameters:
  temperature: 0.7
  top_p: 0.95
  stream: true
  max_tokens: 4096
  stop: []
  frequency_penalty: 0
  presence_penalty: 0
metadata:
  author: NousResearch, The Bloke
  tags:
  - 7B
  - Finetuned
  size: 4370000000
engine: nitro

source_url ​

  • Type: string

The model download source. It can be an external url or a local filepath.

id ​

  • Type: string

The model identifier, which can be referenced in the API endpoints.

object ​

  • Type: string
  • Default: model

The type of the object.

name ​

  • Type: string

Human-readable name that is used for UI.

version ​

  • Type: string

The version of the model.

description ​

  • Type: string

The description of the model.

format ​

  • Type: string

The format of the model.

settings ​

The model settings.

Configuration example

Yaml
settings:
  ctx_len: 4096
  prompt_template: |-
    {prompt}

ctx_len ​

  • Type: int

The context length of the model.

prompt_template ​

  • Type: string

The prompt template of the model, which is used to generate the prompt part of the model input.

parameters ​

Parameters of the model.

Configuration example

parameters:
  temperature: 0.7
  top_p: 0.95
  stream: true
  max_tokens: 4096
  stop: []
  frequency_penalty: 0
  presence_penalty: 0

temperature ​

  • Type: float

The temperature parameter when the model generates text.

top_p ​

  • Type: float

The top-p parameter when the model generates text, which controls the probability distribution range of the output.

stream ​

  • Type: bool

Indicates whether the model generates text in a streaming manner.

max_tokens ​

  • Type: int

The maximum number of tokens generated by the model.

stop ​

  • Type: array

List of stop words.

frequency_penalty ​

  • Type: int

Frequency penalty parameter, used to adjust the frequency of vocabulary in the generated text.

presence_penalty ​

  • Type: int

Presence penalty parameter, used to adjust the probability of the presence of vocabulary in the generated text.

metadata ​

Metadata of the model.

Configuration example

metadata:
  author: NousResearch, The Bloke
  tags:
  - 7B
  - Finetuned
  size: 4370000000

author ​

  • Type: string

The author name of the model.

tags ​

  • Type: array

List of tags, used to describe the attributes or features of the model.

size ​

  • Type: int

The size of the model.

author ​

  • Type: string

The author name of the model.

engine ​

  • Type: string

The model engine used.

source_url ​

  • Type: string

The model download source. It can be an external url or a local filepath.

id ​

  • Type: string

The model identifier, which can be referenced in the API endpoints.

object ​

  • Type: string
  • Default: model

The type of the object.

name ​

  • Type: string

Human-readable name that is used for UI.

version ​

  • Type: string

The version of the model.

description ​

  • Type: string

Description of the model

###= format

  • Type: string

The format of the model.

settings ​

The model settings.

Configuration example:

Yaml
settings:
  ctx_len: 4096
  prompt_template: |-
    {prompt}

ctx_len ​

  • Type: int

The context length of the model.

prompt_template ​

  • Type: string

The prompt template of the model, which is used to generate the prompt part of the model input.

parameters ​

Model parameters.

Configuration example:

Yaml
parameters:
  temperature: 0.7
  top_p: 0.95
  stream: true
  max_tokens: 4096
  stop: []
  frequency_penalty: 0
  presence_penalty: 0

temperature ​

  • Type: float

The emotional temperature parameter when the model generates text.

top_p ​

  • Type: float

The top-p parameter when the model generates text, which controls the probability distribution range of the output.

stream ​

  • Type: bool

Indicates whether the model generates text in a streaming manner.

max_tokens ​

  • Type: int

The maximum number of tokens generated by the model.

stop ​

  • Type: array

List of stop words.

frequency_penalty ​

  • Type: int

Frequency penalty parameter, used to adjust the frequency of vocabulary in the generated text.

presence_penalty ​

  • Type: int

Presence penalty parameter, used to adjust the probability of the presence of vocabulary in the generated text.

metadata ​

Metadata of the model.

Configuration example

Yaml
metadata:
  author: NousResearch, The Bloke
  tags:
  - 7B
  - Finetuned
  size: 4370000000

author ​

  • Type: string

The name of the model author.

tags ​

  • Type: array

List of tags, used to describe the attributes or features of the model.

size ​

  • Type: int

The size of the model.

engine ​

  • Type: string

The model engine.