LLM Providers¶

faker-news works with any OpenAI-compatible LLM API. This guide provides detailed setup instructions for popular providers.

OpenAI¶

Setup¶

Using Setup WizardUsing Environment VariablesUsing Python Config

faker-news setup
# Select "OpenAI" when prompted

export OPENAI_API_KEY="sk-..."
export OPENAI_BASE_URL="https://api.openai.com/v1"  # Optional

from faker_news import LLMClientConfig

llm_config = LLMClientConfig(
    api_key="sk-...",
    base_url="https://api.openai.com/v1",
    model_headlines="gpt-4o-mini",
    model_writing="gpt-4o"
)

Recommended Models¶

Purpose	Model	Speed	Quality	Cost
Headlines	`gpt-4o-mini`	⚡⚡⚡	⭐⭐⭐	💰
Headlines	`gpt-3.5-turbo`	⚡⚡⚡	⭐⭐	💰
Articles	`gpt-4o-mini`	⚡⚡⚡	⭐⭐⭐	💰
Articles (Quality)	`gpt-4o`	⚡⚡	⭐⭐⭐⭐⭐	💰💰💰
Articles (Quality)	`gpt-4-turbo`	⚡⚡	⭐⭐⭐⭐	💰💰💰

Example Configuration¶

from faker import Faker
from faker_news import NewsProvider, LLMClientConfig

# Balanced setup (recommended)
llm_config = LLMClientConfig(
    model_headlines="gpt-4o-mini",  # Fast and cheap
    model_writing="gpt-4o-mini"     # Good quality
)

# Quality-focused setup
llm_config = LLMClientConfig(
    model_headlines="gpt-4o-mini",  # Headlines don't need premium
    model_writing="gpt-4o"          # Best quality for articles
)

fake = Faker()
fake.add_provider(NewsProvider(fake, llm_config=llm_config))

Getting an API Key¶

Sign up at platform.openai.com
Navigate to API keys
Create a new secret key
Copy and store securely

Alibaba DashScope (Qwen)¶

Setup¶

Using Setup WizardUsing Environment VariablesUsing Python Config

faker-news setup
# Select "Alibaba DashScope/Qwen" when prompted

export DASHSCOPE_API_KEY="sk-..."
export DASHSCOPE_BASE_URL="https://dashscope-intl.aliyuncs.com/compatible-mode/v1"  # Optional

from faker_news import LLMClientConfig

llm_config = LLMClientConfig(
    api_key="sk-...",
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
    model_headlines="qwen-flash",
    model_writing="qwen-plus"
)

Recommended Models¶

Purpose	Model	Speed	Quality	Cost
Headlines	`qwen-flash`	⚡⚡⚡	⭐⭐⭐	💰
Headlines	`qwen-turbo`	⚡⚡	⭐⭐⭐⭐	💰💰
Articles	`qwen-flash`	⚡⚡⚡	⭐⭐⭐	💰
Articles (Quality)	`qwen-plus`	⚡⚡	⭐⭐⭐⭐	💰💰
Articles (Quality)	`qwen-max`	⚡	⭐⭐⭐⭐⭐	💰💰💰

Example Configuration¶

from faker import Faker
from faker_news import NewsProvider, LLMClientConfig

# Balanced setup
llm_config = LLMClientConfig(
    model_headlines="qwen-flash",  # Fast and cheap
    model_writing="qwen-plus"      # Better quality
)

fake = Faker()
fake.add_provider(NewsProvider(fake, llm_config=llm_config))

Getting an API Key¶

Sign up at Alibaba Cloud Model Studio
Enable DashScope API
Create an API key
Use international endpoint for global access

Azure OpenAI¶

Setup¶

from faker_news import LLMClientConfig

llm_config = LLMClientConfig(
    api_key="your-azure-api-key",
    base_url="https://{your-resource-name}.openai.azure.com/openai/deployments/{deployment-name}",
    model_headlines="gpt-4o-mini",  # Your deployment name
    model_writing="gpt-4"           # Your deployment name
)

Example Configuration¶

from faker import Faker
from faker_news import NewsProvider, LLMClientConfig

# Azure OpenAI setup
llm_config = LLMClientConfig(
    api_key="abc123...",
    base_url="https://my-openai.openai.azure.com/openai/deployments/gpt-4o-mini",
    model_headlines="gpt-4o-mini",
    model_writing="gpt-4"
)

fake = Faker()
fake.add_provider(NewsProvider(fake, llm_config=llm_config))

Deployment Names

Use your Azure deployment names as the model names. These may differ from OpenAI's model names.

Getting Started with Azure OpenAI¶

Create an Azure account
Create an Azure OpenAI resource
Deploy models (e.g., gpt-4o-mini, gpt-4)
Get API key and endpoint from Azure portal

Other OpenAI-Compatible APIs¶

faker-news works with any API that implements the OpenAI chat completions interface.

Generic Setup¶

from faker_news import LLMClientConfig

llm_config = LLMClientConfig(
    api_key="your-api-key",
    base_url="https://your-api-endpoint.com/v1",
    model_headlines="your-fast-model",
    model_writing="your-quality-model"
)

Compatible Providers¶

These providers have OpenAI-compatible APIs:

Anthropic (via compatibility layer)
Groq - Ultra-fast inference
Together.ai - Multiple open models
Perplexity - Search-augmented models
Anyscale - Llama and Mistral models
DeepSeek - Chinese LLM provider
Local models (via LM Studio, Ollama, vLLM, etc.)

Example: Local LLM with Ollama¶

from faker_news import LLMClientConfig

# Run Ollama locally (ollama serve)
llm_config = LLMClientConfig(
    api_key="not-needed",  # Ollama doesn't require API key
    base_url="http://localhost:11434/v1",
    model_headlines="llama3.1",
    model_writing="llama3.1"
)

Example: Groq¶

from faker_news import LLMClientConfig

llm_config = LLMClientConfig(
    api_key="gsk_...",
    base_url="https://api.groq.com/openai/v1",
    model_headlines="llama-3.1-70b-versatile",
    model_writing="llama-3.1-70b-versatile"
)

Choosing a Provider¶

By Use Case¶

Use Case	Recommended Provider	Why
Production	OpenAI	Best reliability and quality
Development	OpenAI (gpt-4o-mini)	Good balance of cost/quality
High-volume	Alibaba Qwen	Cost-effective for bulk
Local/Offline	Ollama + local models	No API costs, privacy
Fast inference	Groq	Ultra-fast responses
Budget-conscious	Alibaba Qwen	Lowest cost per token

By Region¶

Region	Recommended Provider
North America	OpenAI, Azure OpenAI
Europe	OpenAI, Azure OpenAI
Asia-Pacific	Alibaba DashScope
China	Alibaba DashScope
Global	OpenAI

By Requirements¶

Need best quality? → OpenAI GPT-4o

Need lowest cost? → Alibaba Qwen Flash

Need fastest speed? → Groq with Llama

Need privacy? → Local models (Ollama)

Need reliability? → OpenAI or Azure OpenAI

Cost Optimization¶

Model Selection Strategy¶

from faker_news import LLMClientConfig

# Cost-optimized: use cheapest models
llm_config = LLMClientConfig(
    model_headlines="gpt-4o-mini",  # Cheap for simple headlines
    model_writing="gpt-4o-mini"     # Still good quality
)

# Quality-optimized: use better model for articles only
llm_config = LLMClientConfig(
    model_headlines="gpt-4o-mini",  # Cheap for headlines
    model_writing="gpt-4o"          # Premium for articles
)

Batch Size Tuning¶

Larger batches = fewer API calls = lower cost:

from faker_news import NewsProvider

# Cost-optimized setup
provider = NewsProvider(
    fake,
    llm_config=llm_config,
    headline_batch=200,  # Large batches
    article_batch=50
)

Caching Strategy¶

Maximize cache reuse to minimize API calls:

from faker import Faker
from faker_news import NewsProvider

fake = Faker()
fake.add_provider(NewsProvider(fake))

# Preload once, use many times
fake.news_preload_headlines(1000)

# Reuse content
headlines = [fake.news_headline() for _ in range(500)]
fake.news_reset("reuse")
headlines = [fake.news_headline() for _ in range(500)]  # No new API calls

Troubleshooting¶

API Key Not Working¶

Verify key format matches provider
Check environment variables: echo $OPENAI_API_KEY

Test with curl:

curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"

Connection Errors¶

Verify base URL is correct
Check network connectivity
Verify API endpoint is accessible
Check for firewall/proxy issues

Rate Limiting¶

If you hit rate limits:

Reduce batch sizes
Add delays between generations
Upgrade to higher tier
Switch to provider with higher limits

Model Not Found¶

If you get "model not found" errors:

Verify model name is correct
Check if you have access to that model
For Azure: use deployment name, not model name

Next Steps¶

Configuration - Customize batch sizes and settings
API Reference - Complete API documentation