Models

Model Hub

Deploy and manage AI models for workflow automation.

---

Overview

Model Hub provides centralized AI model management:

• Open-source models — Access Hugging Face and custom models
• External providers — Connect OpenAI, Anthropic, Google
• Fine-tuned models — Train models on your data
• Model versioning — Track and deploy model versions

---

Model Types

Open-Source Models

Access models from Hugging Face and other sources:

Category	Examples
Text generation	Llama 2, Mistral, Falcon
Embeddings	BGE, E5, Instructor
Classification	BERT, RoBERTa
Image	Stable Diffusion, SDXL

External Models

Connect to cloud AI providers:

Provider	Models
OpenAI	GPT-4, GPT-4o, GPT-3.5, DALL-E, Whisper
Anthropic	Claude 3 Opus, Sonnet, Haiku
Google	Gemini 1.5 Pro, Gemini 1.5 Flash
Azure	Azure OpenAI deployments

Fine-Tuned Models

Custom models trained on your data:

Type	Use Case
Classification	Custom categories
Extraction	Domain-specific entities
Generation	Brand voice, style
Embeddings	Domain-specific similarity

---

Adding Models

Add External Provider

1. Navigate to Models → External Models
2. Click Add Provider
3. Configure:

Provider: OpenAI
API Key: sk-...
Organization ID: org-... (optional)
Base URL: https://api.openai.com (default)
Models:
  - gpt-4
  - gpt-4o
  - gpt-3.5-turbo
  - text-embedding-3-small

Add Open-Source Model

1. Navigate to Models → Open-Source
2. Click Add Model
3. Configure:

Model: mistralai/Mistral-7B-Instruct-v0.2
Source: huggingface
Deployment:
  type: serverless | dedicated
  gpu: A100
  replicas: 1

Import Custom Model

1. Navigate to Models → Custom
2. Click Import Model
3. Provide:
   • Model files (safetensors, GGUF, etc.)
   • Model configuration
   • Tokenizer files
   • Deployment settings

---

Fine-Tuning

Create Fine-Tune Job

1. Navigate to Models → Fine-Tuning
2. Click Create Fine-Tune
3. Configure:

Fine-Tune Job:
  name: "Invoice Classifier v2"
  base_model: gpt-3.5-turbo
  training_data:
    source: dataset
    dataset_id: invoice-training-data
  validation_data:
    source: dataset
    dataset_id: invoice-validation-data
  hyperparameters:
    epochs: 3
    batch_size: 4
    learning_rate_multiplier: 1.0

Prepare Training Data

Format training data:

{
  "messages": [
    {"role": "system", "content": "You classify invoices."},
    {"role": "user", "content": "Invoice from Acme Corp for office supplies"},
    {"role": "assistant", "content": "Category: Office Supplies\nVendor Type: Supplier"}
  ]
}

Monitor Training

Track fine-tuning progress:

Metric	Description
Training loss	Model learning progress
Validation loss	Generalization quality
Epochs completed	Training iterations
Estimated time	Time remaining

Deploy Fine-Tuned Model

After training completes:

1. Review training metrics
2. Test with sample inputs
3. Click Deploy
4. Select deployment target

---

Model Configuration

Model Settings

Configure model behavior:

Model Settings:
  model_id: gpt-4
  parameters:
    temperature: 0.7
    max_tokens: 1000
    top_p: 1.0
    frequency_penalty: 0.0
    presence_penalty: 0.0
  timeout: 30s
  retry:
    count: 3
    delay: 1s

Rate Limits

Set usage limits:

Rate Limits:
  requests_per_minute: 100
  tokens_per_minute: 100000
  requests_per_day: 10000
  alert_threshold: 80%

Cost Management

Track and control costs:

Cost Management:
  budget:
    monthly: $1000
    alert_at: 80%
  tracking:
    by_workflow: true
    by_user: true
  limits:
    max_request_cost: $1

---

Model Endpoints

Create Endpoint

Expose models as API endpoints:

Endpoint: invoice-classifier
Model: ft:gpt-3.5-turbo:acme:invoice-v2
Authentication: api_key
Rate limit: 100/minute
URL: https://api.kore.ai/models/invoice-classifier/v1

Use Endpoint

curl -X POST https://api.kore.ai/models/invoice-classifier/v1/complete \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Classify this invoice: ...",
    "max_tokens": 100
  }'

---

Integrations

Hugging Face

Connect your Hugging Face account:

Hugging Face Integration:
  api_key: hf_...
  organization: your-org
  sync_models: true
  auto_import:
    - pattern: "your-org/*"

Weights & Biases

Track experiments with W&B:

W&B Integration:
  api_key: ...
  project: ai-for-process
  entity: your-team
  log:
    - training_metrics
    - model_artifacts
    - predictions

S3 / Cloud Storage

Store models in cloud storage:

Storage Integration:
  provider: s3
  bucket: ai-models
  prefix: ai-for-process/
  credentials:
    access_key: ...
    secret_key: ...

---

Supported Models

Text Generation

Model	Provider	Context
GPT-4	OpenAI	128K
GPT-4o	OpenAI	128K
Claude 3 Opus	Anthropic	200K
Claude 3 Sonnet	Anthropic	200K
Gemini 1.5 Pro	Google	1M
Llama 2 70B	Open-source	4K
Mistral 7B	Open-source	32K

Embeddings

Model	Provider	Dimensions
text-embedding-3-small	OpenAI	1536
text-embedding-3-large	OpenAI	3072
BGE-large	Open-source	1024
E5-large-v2	Open-source	1024

Image

Model	Provider	Type
DALL-E 3	OpenAI	Generation
Stable Diffusion XL	Open-source	Generation
GPT-4 Vision	OpenAI	Understanding

Audio

Model	Provider	Type
Whisper	OpenAI	Transcription
Whisper-large-v3	Open-source	Transcription

---

Related

• Prompts
• Workflow Nodes
• Getting Started