Tomas Valenta
646 lines
15 KiB
TypeScript
646 lines
15 KiB
TypeScript
import { Creds } from 'hooks/useModelProviderArgs'
|
|
|
|
import { PromptFragment } from './prompt'
|
|
import { TemplateID, templates } from './template'
|
|
|
|
export enum ModelProvider {
|
|
OpenAI = 'OpenAI',
|
|
Replicate = 'Replicate',
|
|
Anthropic = 'Anthropic',
|
|
HuggingFace = 'HuggingFace',
|
|
Banana = 'Banana',
|
|
AzureOpenAI = 'AzureOpenAI'
|
|
}
|
|
|
|
export interface ModelConfig extends Model {
|
|
args: ModelArgs
|
|
prompt: PromptFragment[]
|
|
}
|
|
|
|
export type ArgValue = string | number
|
|
|
|
export interface ModelArgs {
|
|
[arg: string]: ArgValue | undefined
|
|
}
|
|
|
|
export interface ModelTemplateArg {
|
|
label?: string
|
|
editable?: boolean
|
|
type: 'string' | 'number'
|
|
// If this property is defined it is used as a default value.
|
|
value?: ArgValue
|
|
min?: number
|
|
max?: number
|
|
step?: number
|
|
placeholder?: string
|
|
optional?: boolean
|
|
}
|
|
|
|
export interface ProviderTemplateCreds {
|
|
[key: string]: Omit<ModelTemplateArg, 'editable' | 'value'>
|
|
}
|
|
|
|
export interface Model {
|
|
provider: ModelProvider
|
|
name: string
|
|
}
|
|
|
|
export interface ModelTemplate extends Model {
|
|
/**
|
|
* Args should be exactly the same as the args to the langchain's model class in Python.
|
|
*/
|
|
args?: { [arg: string]: ModelTemplateArg }
|
|
}
|
|
|
|
export interface ProviderTemplate {
|
|
// These creds are merged with the model args then send to the API.
|
|
creds?: ProviderTemplateCreds
|
|
models: Omit<ModelTemplate, 'provider'>[]
|
|
link?: string
|
|
}
|
|
|
|
export const providerTemplates: {
|
|
[provider in keyof typeof ModelProvider]: ProviderTemplate
|
|
} = {
|
|
[ModelProvider.Anthropic]: {
|
|
link: 'https://anthropic.com',
|
|
creds: {
|
|
anthropic_api_key: {
|
|
label: 'Anthropic API Key',
|
|
type: 'string',
|
|
},
|
|
},
|
|
models: [
|
|
{
|
|
name: 'Claude v1.3',
|
|
args: {
|
|
model: {
|
|
value: 'claude-v1.3',
|
|
type: 'string',
|
|
},
|
|
max_tokens_to_sample: {
|
|
type: 'number',
|
|
editable: true,
|
|
label: 'Max tokens to sample',
|
|
value: 2048,
|
|
step: 1,
|
|
min: 1,
|
|
},
|
|
temperature: {
|
|
label: 'Temperature',
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0.4,
|
|
min: 0.01,
|
|
max: 1,
|
|
step: 0.01,
|
|
},
|
|
top_p: {
|
|
label: 'top-p',
|
|
editable: true,
|
|
type: 'number',
|
|
value: -1,
|
|
min: 0.01,
|
|
max: 1,
|
|
step: 0.01,
|
|
},
|
|
top_k: {
|
|
label: 'top-k',
|
|
editable: true,
|
|
type: 'number',
|
|
value: -1,
|
|
min: 0,
|
|
step: 1,
|
|
},
|
|
},
|
|
},
|
|
{
|
|
name: 'Claude Instant v1',
|
|
args: {
|
|
model: {
|
|
value: 'claude-instant-v1',
|
|
type: 'string',
|
|
},
|
|
max_tokens_to_sample: {
|
|
type: 'number',
|
|
editable: true,
|
|
label: 'Max tokens to sample',
|
|
value: 2048,
|
|
step: 1,
|
|
min: 1,
|
|
},
|
|
temperature: {
|
|
label: 'Temperature',
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0.4,
|
|
min: 0.01,
|
|
max: 1,
|
|
step: 0.01,
|
|
},
|
|
top_p: {
|
|
label: 'top-p',
|
|
editable: true,
|
|
type: 'number',
|
|
value: -1,
|
|
min: 0.01,
|
|
max: 1,
|
|
step: 0.01,
|
|
},
|
|
top_k: {
|
|
label: 'top-k',
|
|
editable: true,
|
|
type: 'number',
|
|
value: -1,
|
|
min: 0,
|
|
step: 1,
|
|
},
|
|
},
|
|
},
|
|
],
|
|
},
|
|
[ModelProvider.HuggingFace]: {
|
|
link: 'https://huggingface.co',
|
|
creds: {
|
|
huggingfacehub_api_token: {
|
|
label: 'Hugging Face API Key',
|
|
type: 'string',
|
|
},
|
|
},
|
|
models: [
|
|
{
|
|
name: 'Inference API',
|
|
args: {
|
|
repo_id: {
|
|
editable: true,
|
|
type: 'string',
|
|
label: 'Repo ID',
|
|
placeholder: 'owner-name/repo-name',
|
|
},
|
|
max_length: {
|
|
type: 'number',
|
|
editable: true,
|
|
step: 1,
|
|
min: 1,
|
|
value: 250,
|
|
optional: true
|
|
},
|
|
temperature: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0,
|
|
max: 100,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
top_p: {
|
|
editable: true,
|
|
type: 'number',
|
|
min: 0,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
top_k: {
|
|
editable: true,
|
|
type: 'number',
|
|
min: 0,
|
|
step: 1,
|
|
optional: true
|
|
},
|
|
repetition_penalty: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0,
|
|
max: 100,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
max_time: {
|
|
editable: true,
|
|
type: 'number',
|
|
min: 0,
|
|
max: 120,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
},
|
|
},
|
|
{
|
|
name: 'Inference Endpoints',
|
|
args: {
|
|
endpoint_url: {
|
|
editable: true,
|
|
type: 'string',
|
|
label: 'Endpoint URL',
|
|
placeholder: '...endpoints.huggingface.cloud',
|
|
},
|
|
max_length: {
|
|
type: 'number',
|
|
editable: true,
|
|
step: 1,
|
|
min: 1,
|
|
value: 250,
|
|
optional: true
|
|
},
|
|
temperature: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0.01,
|
|
max: 100,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
top_p: {
|
|
editable: true,
|
|
type: 'number',
|
|
min: 0,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
top_k: {
|
|
editable: true,
|
|
type: 'number',
|
|
min: 0,
|
|
step: 1,
|
|
optional: true
|
|
},
|
|
repetition_penalty: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0,
|
|
max: 100,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
max_time: {
|
|
editable: true,
|
|
type: 'number',
|
|
min: 0,
|
|
max: 120,
|
|
step: 0.01,
|
|
optional: true
|
|
},
|
|
},
|
|
},
|
|
],
|
|
},
|
|
[ModelProvider.Replicate]: {
|
|
link: 'https://replicate.com',
|
|
creds: {
|
|
replicate_api_token: {
|
|
label: 'Replicate API Key',
|
|
type: 'string',
|
|
},
|
|
},
|
|
models: [
|
|
{
|
|
name: 'Hosted model',
|
|
args: {
|
|
model: {
|
|
placeholder: 'owner-name/model-name:version',
|
|
editable: true,
|
|
type: 'string',
|
|
label: 'Model'
|
|
},
|
|
max_length: {
|
|
type: 'number',
|
|
editable: true,
|
|
value: 2500,
|
|
step: 1,
|
|
min: 1,
|
|
},
|
|
temperature: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0.5,
|
|
min: 0,
|
|
step: 0.01,
|
|
},
|
|
top_p: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0.9,
|
|
min: 0,
|
|
max: 1,
|
|
step: 0.01,
|
|
},
|
|
top_k: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0,
|
|
min: 0,
|
|
step: 0.01,
|
|
},
|
|
repetition_penalty: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0,
|
|
step: 0.01,
|
|
},
|
|
},
|
|
},
|
|
],
|
|
},
|
|
[ModelProvider.Banana]: {
|
|
link: 'https://banana.dev',
|
|
creds: {
|
|
banana_api_key: {
|
|
label: 'Banana API Key',
|
|
type: 'string',
|
|
},
|
|
},
|
|
models: [
|
|
{
|
|
name: 'Hosted model',
|
|
args: {
|
|
model_key: {
|
|
placeholder: 'xxxx-xxxx-xxxx-xxxx',
|
|
editable: true,
|
|
type: 'string',
|
|
label: 'Model Key'
|
|
},
|
|
max_length: {
|
|
type: 'number',
|
|
editable: true,
|
|
value: 2500,
|
|
step: 1,
|
|
min: 1,
|
|
},
|
|
temperature: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0.5,
|
|
min: 0,
|
|
step: 0.01,
|
|
},
|
|
top_p: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0.9,
|
|
min: 0,
|
|
max: 1,
|
|
step: 0.01,
|
|
},
|
|
top_k: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 0,
|
|
min: 0,
|
|
step: 0.01,
|
|
},
|
|
repetition_penalty: {
|
|
editable: true,
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0,
|
|
step: 0.01,
|
|
},
|
|
},
|
|
},
|
|
],
|
|
},
|
|
[ModelProvider.OpenAI]: {
|
|
link: 'https://openai.com',
|
|
creds: {
|
|
openai_api_key: {
|
|
label: 'OpenAI API Key',
|
|
type: 'string',
|
|
},
|
|
},
|
|
models: [
|
|
{
|
|
name: 'GPT 3.5 Turbo',
|
|
args: {
|
|
model_name: {
|
|
type: 'string',
|
|
value: 'gpt-3.5-turbo',
|
|
},
|
|
max_tokens: {
|
|
label: 'Max tokens',
|
|
type: 'number',
|
|
value: 2048,
|
|
editable: true,
|
|
},
|
|
temperature: {
|
|
type: 'number',
|
|
label: 'Temperature',
|
|
value: 0.0,
|
|
min: 0,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
presence_penalty: {
|
|
type: 'number',
|
|
label: 'Presence penalty',
|
|
value: 0.0,
|
|
min: -2,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
frequency_penalty: {
|
|
type: 'number',
|
|
label: 'Frequency penalty',
|
|
value: 0.0,
|
|
min: -2,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
top_p: {
|
|
label: 'Top-p',
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0.0,
|
|
max: 1,
|
|
step: 0.01,
|
|
editable: true,
|
|
},
|
|
},
|
|
},
|
|
{
|
|
name: 'GPT 4',
|
|
args: {
|
|
model_name: {
|
|
type: 'string',
|
|
value: 'gpt-4',
|
|
},
|
|
max_tokens: {
|
|
label: 'Max tokens',
|
|
type: 'number',
|
|
value: 2048,
|
|
editable: true,
|
|
},
|
|
temperature: {
|
|
type: 'number',
|
|
label: 'Temperature',
|
|
value: 0,
|
|
min: 0.0,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
presence_penalty: {
|
|
type: 'number',
|
|
label: 'Presence penalty',
|
|
value: 0.0,
|
|
min: -2,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
frequency_penalty: {
|
|
type: 'number',
|
|
label: 'Frequency penalty',
|
|
value: 0.0,
|
|
min: -2,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
top_p: {
|
|
label: 'Top-p',
|
|
type: 'number',
|
|
value: 1.0,
|
|
min: 0,
|
|
max: 1,
|
|
step: 0.01,
|
|
editable: true,
|
|
},
|
|
},
|
|
|
|
},
|
|
],
|
|
},
|
|
[ModelProvider.AzureOpenAI]: {
|
|
link: 'https://azure.microsoft.com/en-us/products/cognitive-services/openai-service',
|
|
creds: {
|
|
openai_api_key: {
|
|
label: 'Azure OpenAI API Key',
|
|
type: 'string',
|
|
},
|
|
openai_api_version: {
|
|
label: 'Azure OpenAI API version',
|
|
type: 'string',
|
|
},
|
|
openai_api_base: {
|
|
label: 'Azure OpenAI API base',
|
|
type: 'string',
|
|
},
|
|
},
|
|
models: [
|
|
{
|
|
name: 'Hosted model',
|
|
args: {
|
|
deployment_name: {
|
|
label: 'Deployment name',
|
|
type: 'string',
|
|
placeholder: 'gpt-3.5-turbo',
|
|
editable: true,
|
|
},
|
|
model_name: {
|
|
label: 'Model name',
|
|
type: 'string',
|
|
value: 'gpt-3.5-turbo',
|
|
editable: true,
|
|
},
|
|
max_tokens: {
|
|
label: 'Max tokens',
|
|
type: 'number',
|
|
value: 2048,
|
|
editable: true,
|
|
},
|
|
temperature: {
|
|
type: 'number',
|
|
label: 'Temperature',
|
|
value: 0.0,
|
|
min: 0,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
presence_penalty: {
|
|
type: 'number',
|
|
label: 'Presence penalty',
|
|
value: 0.0,
|
|
min: -2,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
frequency_penalty: {
|
|
type: 'number',
|
|
label: 'Frequency penalty',
|
|
value: 0.0,
|
|
min: -2,
|
|
step: 0.01,
|
|
max: 2,
|
|
editable: true,
|
|
},
|
|
top_p: {
|
|
label: 'Top-p',
|
|
type: 'number',
|
|
value: 1,
|
|
min: 0.0,
|
|
max: 1,
|
|
step: 0.01,
|
|
editable: true,
|
|
},
|
|
},
|
|
}
|
|
],
|
|
}
|
|
}
|
|
|
|
export function getModelArgs(
|
|
modelConfig: Pick<ModelConfig, 'provider' | 'name' | 'args'>,
|
|
creds: Creds,
|
|
): ModelArgs {
|
|
const template = providerTemplates[modelConfig.provider]?.models.find(m => m.name === modelConfig.name)
|
|
if (!template) throw new Error(`Cannot find model template ${modelConfig.provider}/${modelConfig.name}`)
|
|
|
|
const defaultArgs = Object
|
|
.entries(template.args || {})
|
|
.reduce<ModelArgs>((prev, [key, info]) => {
|
|
if (info.value !== undefined) {
|
|
prev[key] = info.value
|
|
}
|
|
return prev
|
|
}, {})
|
|
|
|
return {
|
|
...defaultArgs,
|
|
...creds[modelConfig.provider]?.creds,
|
|
...modelConfig.args,
|
|
}
|
|
}
|
|
|
|
export const defaultModelProvider = ModelProvider.OpenAI
|
|
export const defaultModelName = providerTemplates[defaultModelProvider].models[0].name
|
|
|
|
export function getMissingCreds(provider: ModelProvider, creds: Creds) {
|
|
return Object
|
|
.entries(providerTemplates[provider]?.creds || {})
|
|
.filter(([key]) => creds[provider]?.creds?.[key] === undefined)
|
|
}
|
|
|
|
export function getDefaultModelConfig(templateID: TemplateID): ModelConfig {
|
|
const prompt = templates[templateID].prompt
|
|
return {
|
|
provider: defaultModelProvider,
|
|
name: defaultModelName,
|
|
args: {},
|
|
prompt,
|
|
}
|
|
}
|
|
|
|
export function isModelEqual(m1: Model, m2: Model) {
|
|
return m1.provider === m2.provider && m1.name === m2.name
|
|
}
|