|
| 1 | +# LiteLLM with Swarms |
| 2 | + |
| 3 | +LiteLLM provides a unified interface for 100+ LLM providers. Swarms uses LiteLLM to support multiple providers through a single API. |
| 4 | + |
| 5 | +## Quick Start |
| 6 | + |
| 7 | +```python |
| 8 | +from swarms import Agent |
| 9 | + |
| 10 | +# Use any LiteLLM-supported model |
| 11 | +agent = Agent( |
| 12 | + model_name="gpt-4o-mini", # Change this to any provider |
| 13 | + max_loops=1, |
| 14 | +) |
| 15 | + |
| 16 | +response = agent.run("Hello, world!") |
| 17 | +``` |
| 18 | + |
| 19 | +## Supported Providers |
| 20 | + |
| 21 | +Switch providers by changing `model_name`: |
| 22 | + |
| 23 | +```python |
| 24 | +# OpenAI |
| 25 | +Agent(model_name="gpt-4o") |
| 26 | +Agent(model_name="gpt-4o-mini") |
| 27 | +Agent(model_name="gpt-3.5-turbo") |
| 28 | + |
| 29 | +# Anthropic Claude |
| 30 | +Agent(model_name="claude-3-5-sonnet-20241022") |
| 31 | +Agent(model_name="claude-3-opus") |
| 32 | + |
| 33 | +# Google Gemini |
| 34 | +Agent(model_name="gemini/gemini-pro") |
| 35 | +Agent(model_name="gemini/gemini-1.5-pro") |
| 36 | + |
| 37 | +# Azure OpenAI |
| 38 | +Agent(model_name="azure/gpt-4") |
| 39 | + |
| 40 | +# Ollama (local) |
| 41 | +Agent(model_name="ollama/llama2") |
| 42 | +Agent(model_name="ollama/mistral") |
| 43 | + |
| 44 | +# Cohere |
| 45 | +Agent(model_name="command-r") |
| 46 | +Agent(model_name="command-r-plus") |
| 47 | + |
| 48 | +# DeepSeek |
| 49 | +Agent(model_name="deepseek/deepseek-chat") |
| 50 | +Agent(model_name="deepseek/deepseek-r1") |
| 51 | + |
| 52 | +# Groq |
| 53 | +Agent(model_name="groq/llama-3.1-70b-versatile") |
| 54 | + |
| 55 | +# OpenRouter |
| 56 | +Agent(model_name="openrouter/google/palm-2-chat-bison") |
| 57 | + |
| 58 | +# X.AI |
| 59 | +Agent(model_name="xai/grok-beta") |
| 60 | +``` |
| 61 | + |
| 62 | +## Using LiteLLM Wrapper Directly |
| 63 | + |
| 64 | +```python |
| 65 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 66 | + |
| 67 | +llm = LiteLLM( |
| 68 | + model_name="gpt-4o", |
| 69 | + temperature=0.7, |
| 70 | + max_tokens=2000, |
| 71 | + verbose=True, |
| 72 | +) |
| 73 | + |
| 74 | +response = llm.run("What is machine learning?") |
| 75 | +``` |
| 76 | + |
| 77 | +## Features |
| 78 | + |
| 79 | +### 1. Vision (Image Input) |
| 80 | + |
| 81 | +```python |
| 82 | +from swarms import Agent |
| 83 | + |
| 84 | +agent = Agent(model_name="gpt-4o", max_loops=1) |
| 85 | + |
| 86 | +# Supports: file path, URL, or base64 |
| 87 | +response = agent.run( |
| 88 | + "Describe this image", |
| 89 | + img="path/to/image.jpg" # or URL or base64 |
| 90 | +) |
| 91 | +``` |
| 92 | + |
| 93 | +### 2. Tool/Function Calling |
| 94 | + |
| 95 | +```python |
| 96 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 97 | + |
| 98 | +tools = [ |
| 99 | + { |
| 100 | + "type": "function", |
| 101 | + "function": { |
| 102 | + "name": "get_weather", |
| 103 | + "description": "Get weather for a location", |
| 104 | + "parameters": { |
| 105 | + "type": "object", |
| 106 | + "properties": { |
| 107 | + "location": {"type": "string"} |
| 108 | + }, |
| 109 | + "required": ["location"] |
| 110 | + } |
| 111 | + } |
| 112 | + } |
| 113 | +] |
| 114 | + |
| 115 | +llm = LiteLLM( |
| 116 | + model_name="gpt-4o", |
| 117 | + tools_list_dictionary=tools, |
| 118 | + tool_choice="auto", |
| 119 | +) |
| 120 | + |
| 121 | +response = llm.run("What's the weather in San Francisco?") |
| 122 | +``` |
| 123 | + |
| 124 | +### 3. Reasoning Models |
| 125 | + |
| 126 | +```python |
| 127 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 128 | + |
| 129 | +llm = LiteLLM( |
| 130 | + model_name="openai/o1-preview", |
| 131 | + reasoning_enabled=True, |
| 132 | + max_tokens=4000, |
| 133 | +) |
| 134 | + |
| 135 | +response = llm.run("Solve this complex math problem...") |
| 136 | +``` |
| 137 | + |
| 138 | +### 4. Streaming |
| 139 | + |
| 140 | +```python |
| 141 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 142 | + |
| 143 | +llm = LiteLLM(model_name="gpt-4o", stream=True) |
| 144 | + |
| 145 | +for chunk in llm.run("Tell me a story"): |
| 146 | + print(chunk, end="", flush=True) |
| 147 | +``` |
| 148 | + |
| 149 | +### 5. Audio Input |
| 150 | + |
| 151 | +```python |
| 152 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 153 | + |
| 154 | +llm = LiteLLM( |
| 155 | + model_name="gpt-4o", |
| 156 | + audio="path/to/audio.wav", |
| 157 | +) |
| 158 | + |
| 159 | +response = llm.run("Transcribe this audio") |
| 160 | +``` |
| 161 | + |
| 162 | +### 6. Advanced Configuration |
| 163 | + |
| 164 | +```python |
| 165 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 166 | + |
| 167 | +llm = LiteLLM( |
| 168 | + model_name="gpt-4o", |
| 169 | + system_prompt="You are a helpful assistant.", |
| 170 | + temperature=0.7, |
| 171 | + max_tokens=4000, |
| 172 | + stream=False, |
| 173 | + verbose=True, |
| 174 | + retries=3, |
| 175 | + caching=False, |
| 176 | + top_p=1.0, |
| 177 | +) |
| 178 | + |
| 179 | +response = llm.run("Explain neural networks") |
| 180 | +``` |
| 181 | + |
| 182 | +## Provider Setup |
| 183 | + |
| 184 | +### Azure OpenAI |
| 185 | + |
| 186 | +```python |
| 187 | +import os |
| 188 | +os.environ["AZURE_API_KEY"] = "your-key" |
| 189 | +os.environ["AZURE_API_BASE"] = "https://your-resource.openai.azure.com/" |
| 190 | +os.environ["AZURE_API_VERSION"] = "2024-02-15-preview" |
| 191 | + |
| 192 | +agent = Agent(model_name="azure/gpt-4", max_loops=1) |
| 193 | +``` |
| 194 | + |
| 195 | +### Anthropic Claude |
| 196 | + |
| 197 | +```python |
| 198 | +import os |
| 199 | +os.environ["ANTHROPIC_API_KEY"] = "your-key" |
| 200 | + |
| 201 | +agent = Agent(model_name="claude-3-5-sonnet-20241022", max_loops=1) |
| 202 | +``` |
| 203 | + |
| 204 | +### Google Gemini |
| 205 | + |
| 206 | +```python |
| 207 | +import os |
| 208 | +os.environ["GEMINI_API_KEY"] = "your-key" |
| 209 | + |
| 210 | +agent = Agent(model_name="gemini/gemini-pro", max_loops=1) |
| 211 | +``` |
| 212 | + |
| 213 | +### Ollama (Local) |
| 214 | + |
| 215 | +```python |
| 216 | +# No API key needed - ensure Ollama is running |
| 217 | +agent = Agent(model_name="ollama/llama2", max_loops=1) |
| 218 | +``` |
| 219 | + |
| 220 | +## Complete Examples |
| 221 | + |
| 222 | +### Multi-Provider Comparison |
| 223 | + |
| 224 | +```python |
| 225 | +from swarms import Agent |
| 226 | + |
| 227 | +models = ["gpt-4o-mini", "claude-3-5-sonnet-20241022", "gemini/gemini-pro"] |
| 228 | +task = "Explain quantum computing in one paragraph." |
| 229 | + |
| 230 | +for model_name in models: |
| 231 | + print(f"\n=== {model_name} ===") |
| 232 | + agent = Agent(model_name=model_name, max_loops=1) |
| 233 | + response = agent.run(task) |
| 234 | + print(response[:200]) |
| 235 | +``` |
| 236 | + |
| 237 | +### Vision Analysis |
| 238 | + |
| 239 | +```python |
| 240 | +from swarms import Agent |
| 241 | + |
| 242 | +agent = Agent(model_name="gpt-4o", max_loops=1) |
| 243 | + |
| 244 | +response = agent.run( |
| 245 | + "Analyze this image and describe what you see.", |
| 246 | + img="https://example.com/image.jpg" |
| 247 | +) |
| 248 | +print(response) |
| 249 | +``` |
| 250 | + |
| 251 | +### Streaming Response |
| 252 | + |
| 253 | +```python |
| 254 | +from swarms.utils.litellm_wrapper import LiteLLM |
| 255 | + |
| 256 | +llm = LiteLLM(model_name="gpt-4o", stream=True) |
| 257 | + |
| 258 | +print("Response: ", end="") |
| 259 | +for chunk in llm.run("Write a short poem about AI"): |
| 260 | + print(chunk, end="", flush=True) |
| 261 | +``` |
| 262 | + |
| 263 | +## Resources |
| 264 | + |
| 265 | +- **LiteLLM Docs**: https://docs.litellm.ai/ |
| 266 | +- **Providers**: https://docs.litellm.ai/docs/providers |
| 267 | +- **Swarms Wrapper**: `swarms/utils/litellm_wrapper.py` |
0 commit comments