Skip to main content
Open In ColabOpen on GitHub

RunPod LLM

Get started with RunPod LLMs.

Overview​

This guide covers how to use the LangChain RunPod LLM class to interact with text generation models hosted on RunPod Serverless.

Setup​

  1. Install the package: 
    pip install -qU langchain-runpod
  2. Deploy an LLM Endpoint: Follow the setup steps in the RunPod Provider Guide to deploy a compatible text generation endpoint on RunPod Serverless and get its Endpoint ID.
  3. Set Environment Variables: Make sure RUNPOD_API_KEY and RUNPOD_ENDPOINT_ID are set.
import getpass
import os

# Make sure environment variables are set (or pass them directly to RunPod)
if "RUNPOD_API_KEY" not in os.environ:
    os.environ["RUNPOD_API_KEY"] = getpass.getpass("Enter your RunPod API Key: ")
if "RUNPOD_ENDPOINT_ID" not in os.environ:
    os.environ["RUNPOD_ENDPOINT_ID"] = input("Enter your RunPod Endpoint ID: ")

Instantiation​

Initialize the RunPod class. You can pass model-specific parameters via model_kwargs and configure polling behavior.

from langchain_runpod import RunPod

llm = RunPod(
    # runpod_endpoint_id can be passed here if not set in env
    model_kwargs={
        "max_new_tokens": 256,
        "temperature": 0.6,
        "top_k": 50,
        # Add other parameters supported by your endpoint handler
    },
    # Optional: Adjust polling
    # poll_interval=0.3,
    # max_polling_attempts=100
)

Invocation​

Use the standard LangChain .invoke() and .ainvoke() methods to call the model. Streaming is also supported via .stream() and .astream() (simulated by polling the RunPod /stream endpoint).

prompt = "Write a tagline for an ice cream shop on the moon."

# Invoke (Sync)
try:
    response = llm.invoke(prompt)
    print("--- Sync Invoke Response ---")
    print(response)
except Exception as e:
    print(
        f"Error invoking LLM: {e}. Ensure endpoint ID/API key are correct and endpoint is active/compatible."
    )
# Stream (Sync, simulated via polling /stream)
print("\n--- Sync Stream Response ---")
try:
    for chunk in llm.stream(prompt):
        print(chunk, end="", flush=True)
    print()  # Newline
except Exception as e:
    print(
        f"\nError streaming LLM: {e}. Ensure endpoint handler supports streaming output format."
    )

Async Usage​

# AInvoke (Async)
try:
    async_response = await llm.ainvoke(prompt)
    print("--- Async Invoke Response ---")
    print(async_response)
except Exception as e:
    print(f"Error invoking LLM asynchronously: {e}.")
# AStream (Async)
print("\n--- Async Stream Response ---")
try:
    async for chunk in llm.astream(prompt):
        print(chunk, end="", flush=True)
    print()  # Newline
except Exception as e:
    print(
        f"\nError streaming LLM asynchronously: {e}. Ensure endpoint handler supports streaming output format."
    )

Chaining​

The LLM integrates seamlessly with LangChain Expression Language (LCEL) chains.

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import PromptTemplate

# Assumes 'llm' variable is instantiated from the 'Instantiation' cell
prompt_template = PromptTemplate.from_template("Tell me a joke about {topic}")
parser = StrOutputParser()

chain = prompt_template | llm | parser

try:
    chain_response = chain.invoke({"topic": "bears"})
    print("--- Chain Response ---")
    print(chain_response)
except Exception as e:
    print(f"Error running chain: {e}")

# Async chain
try:
    async_chain_response = await chain.ainvoke({"topic": "robots"})
    print("--- Async Chain Response ---")
    print(async_chain_response)
except Exception as e:
    print(f"Error running async chain: {e}")

Endpoint Considerations​

  • Input: The endpoint handler should expect the prompt string within {"input": {"prompt": "...", ...}}.
  • Output: The handler should return the generated text within the "output" key of the final status response (e.g., {"output": "Generated text..."} or {"output": {"text": "..."}}).
  • Streaming: For simulated streaming via the /stream endpoint, the handler must populate the "stream" key in the status response with a list of chunk dictionaries, like [{"output": "token1"}, {"output": "token2"}].

API reference​

For detailed documentation of the RunPod LLM class, parameters, and methods, refer to the source code or the generated API reference (if available).

Link to source code: https://github.com/runpod/langchain-runpod/blob/main/langchain_runpod/llms.py