pig-llm

Unified multi-provider LLM API for Python.

Features

• 🔌 Multi-provider support: OpenAI, Anthropic, Google, and more
• 🎯 Unified interface: Same API across all providers
• 🔄 Streaming support: Real-time token streaming
• 🛡️ Error handling: Automatic retries and fallbacks
• 📊 Usage tracking: Token counting and cost estimation

Installation

pip install pig-llm

Quick Start

from pig_llm import LLM

# Initialize with API key
llm = LLM(provider="openai", api_key="sk-...")

# Simple completion
response = llm.complete("What is the meaning of life?")
print(response.content)

# Streaming
for chunk in llm.stream("Tell me a story"):
    print(chunk.content, end="", flush=True)

# With system message
response = llm.complete(
    "Translate to Spanish",
    system="You are a helpful translator",
)

Supported Providers

Core Providers

• OpenAI - GPT-4, GPT-3.5, etc.
• Anthropic - Claude 3, Claude 2
• Google - Gemini Pro, Gemini Ultra
• Azure OpenAI - Azure-hosted OpenAI models

Additional Providers

• Groq - Ultra-fast LLM inference
• Mistral - Mistral AI models
• OpenRouter - Access to multiple models
• Amazon Bedrock - AWS-hosted foundation models
• xAI (Grok) - xAI's Grok models
• Cerebras - Fastest inference speeds
• Cohere - Command models for enterprise
• Perplexity - Search-augmented LLMs
• DeepSeek - Chinese LLM with strong coding
• Together AI - Open-source model hosting

Configuration

from pig_llm import LLM, Config

config = Config(
    provider="openai",
    model="gpt-4",
    temperature=0.7,
    max_tokens=1000,
    timeout=30,
)

llm = LLM(config=config)

Provider-Specific Examples

Amazon Bedrock

# Uses AWS credentials from environment
llm = LLM(provider="bedrock", api_key="us-east-1")  # region as api_key
response = llm.complete("Hello", model="anthropic.claude-3-sonnet-20240229-v1:0")

xAI (Grok)

llm = LLM(provider="xai", api_key="xai-...")
response = llm.complete("What's happening?", model="grok-beta")

Cerebras

llm = LLM(provider="cerebras", api_key="csk-...")
response = llm.complete("Fast inference!", model="llama3.1-8b")

Cohere

llm = LLM(provider="cohere", api_key="...")
response = llm.complete("Hello", model="command-r-plus")

Perplexity

llm = LLM(provider="perplexity", api_key="pplx-...")
response = llm.complete("What's the latest news?", model="llama-3.1-sonar-large-128k-online")
# Citations available in response.metadata["citations"]

DeepSeek

llm = LLM(provider="deepseek", api_key="...")
response = llm.complete("写一段Python代码", model="deepseek-chat")

Together AI

llm = LLM(provider="together", api_key="...")
response = llm.complete("Hello", model="meta-llama/Llama-3-70b-chat-hf")

License

MIT