LangChain supports integration with Groq chat models. Groq specializes in fast AI inference.
To get started, you'll first need to install the langchain-groq package:
%pip install -qU langchain-groq
Request an API key and set it as an environment variable:
export GROQ_API_KEY=<YOUR API KEY>
Alternatively, you may configure the API key when you initialize ChatGroq.
Here's an example of it in action:
from langchain_core.prompts import ChatPromptTemplate
from langchain_groq import ChatGroq
chat = ChatGroq(
temperature=0,
model="llama3-70b-8192",
# api_key="" # Optional if not set as an environment variable
)
system = "You are a helpful assistant."
human = "{text}"
prompt = ChatPromptTemplate.from_messages([("system", system), ("human", human)])
chain = prompt | chat
chain.invoke({"text": "Explain the importance of low latency for LLMs."})
API Reference:ChatPromptTemplate | ChatGroq
AIMessage(content="Low latency is crucial for Large Language Models (LLMs) because it directly impacts the user experience, model performance, and overall efficiency. Here are some reasons why low latency is essential for LLMs:\n\n1. **Real-time Interaction**: LLMs are often used in applications that require real-time interaction, such as chatbots, virtual assistants, and language translation. Low latency ensures that the model responds quickly to user input, providing a seamless and engaging experience.\n2. **Conversational Flow**: In conversational AI, latency can disrupt the natural flow of conversation. Low latency helps maintain a smooth conversation, allowing users to respond quickly and naturally, without feeling like they're waiting for the model to catch up.\n3. **Model Performance**: High latency can lead to increased error rates, as the model may struggle to keep up with the input pace. Low latency enables the model to process information more efficiently, resulting in better accuracy and performance.\n4. **Scalability**: As the number of users and requests increases, low latency becomes even more critical. It allows the model to handle a higher volume of requests without sacrificing performance, making it more scalable and efficient.\n5. **Resource Utilization**: Low latency can reduce the computational resources required to process requests. By minimizing latency, you can optimize resource allocation, reduce costs, and improve overall system efficiency.\n6. **User Experience**: High latency can lead to frustration, abandonment, and a poor user experience. Low latency ensures that users receive timely responses, which is essential for building trust and satisfaction.\n7. **Competitive Advantage**: In applications like customer service or language translation, low latency can be a key differentiator. It can provide a competitive advantage by offering a faster and more responsive experience, setting your application apart from others.\n8. **Edge Computing**: With the increasing adoption of edge computing, low latency is critical for processing data closer to the user. This reduces latency even further, enabling real-time processing and analysis of data.\n9. **Real-time Analytics**: Low latency enables real-time analytics and insights, which are essential for applications like sentiment analysis, trend detection, and anomaly detection.\n10. **Future-Proofing**: As LLMs continue to evolve and become more complex, low latency will become even more critical. By prioritizing low latency now, you'll be better prepared to handle the demands of future LLM applications.\n\nIn summary, low latency is vital for LLMs because it ensures a seamless user experience, improves model performance, and enables efficient resource utilization. By prioritizing low latency, you can build more effective, scalable, and efficient LLM applications that meet the demands of real-time interaction and processing.", response_metadata={'token_usage': {'completion_tokens': 541, 'prompt_tokens': 33, 'total_tokens': 574, 'completion_time': 1.499777658, 'prompt_time': 0.008344704, 'queue_time': None, 'total_time': 1.508122362}, 'model_name': 'llama3-70b-8192', 'system_fingerprint': 'fp_87cbfbbc4d', 'finish_reason': 'stop', 'logprobs': None}, id='run-49dad960-ace8-4cd7-90b3-2db99ecbfa44-0')
You can view the available models here.
Groq chat models support tool calling to generate output matching a specific schema. The model may choose to call multiple tools or the same tool multiple times if appropriate.
Here's an example:
from typing import Optional
from langchain_core.tools import tool
@tool
def get_current_weather(location: str, unit: Optional[str]):
"""Get the current weather in a given location"""
return "Cloudy with a chance of rain."
tool_model = chat.bind_tools([get_current_weather], tool_choice="auto")
res = tool_model.invoke("What is the weather like in San Francisco and Tokyo?")
res.tool_calls
API Reference:tool
[{'name': 'get_current_weather',
'args': {'location': 'San Francisco', 'unit': 'Celsius'},
'id': 'call_pydj'},
{'name': 'get_current_weather',
'args': {'location': 'Tokyo', 'unit': 'Celsius'},
'id': 'call_jgq3'}]
.with_structured_output()
You can also use the convenience .with_structured_output() method to coerce ChatGroq into returning a structured output.Here is an example:
from langchain_core.pydantic_v1 import BaseModel, Field
class Joke(BaseModel):
"""Joke to tell user."""
setup: str = Field(description="The setup of the joke")
punchline: str = Field(description="The punchline to the joke")
rating: Optional[int] = Field(description="How funny the joke is, from 1 to 10")
structured_llm = chat.with_structured_output(Joke)
structured_llm.invoke("Tell me a joke about cats")
Joke(setup='Why did the cat join a band?', punchline='Because it wanted to be the purr-cussionist!', rating=None)
Behind the scenes, this takes advantage of the above tool calling functionality.
Async
chat = ChatGroq(temperature=0, model="llama3-70b-8192")
prompt = ChatPromptTemplate.from_messages([("human", "Write a Limerick about {topic}")])
chain = prompt | chat
await chain.ainvoke({"topic": "The Sun"})
AIMessage(content='Here is a limerick about the sun:\n\nThere once was a sun in the sky,\nWhose warmth and light caught the eye,\nIt shone bright and bold,\nWith a fiery gold,\nAnd brought life to all, as it flew by.', response_metadata={'token_usage': {'completion_tokens': 51, 'prompt_tokens': 18, 'total_tokens': 69, 'completion_time': 0.144614022, 'prompt_time': 0.00585394, 'queue_time': None, 'total_time': 0.150467962}, 'model_name': 'llama3-70b-8192', 'system_fingerprint': 'fp_2f30b0b571', 'finish_reason': 'stop', 'logprobs': None}, id='run-e42340ba-f0ad-4b54-af61-8308d8ec8256-0')
Streaming
chat = ChatGroq(temperature=0, model="llama3-70b-8192")
prompt = ChatPromptTemplate.from_messages([("human", "Write a haiku about {topic}")])
chain = prompt | chat
for chunk in chain.stream({"topic": "The Moon"}):
print(chunk.content, end="", flush=True)
Silvery glow bright
Luna's gentle light shines down
Midnight's gentle queen
Passing custom parameters
You can pass other Groq-specific parameters using the model_kwargs argument on initialization. Here's an example of enabling JSON mode:
chat = ChatGroq(
model="llama3-70b-8192", model_kwargs={"response_format": {"type": "json_object"}}
)
system = """
You are a helpful assistant.
Always respond with a JSON object with two string keys: "response" and "followup_question".
"""
human = "{question}"
prompt = ChatPromptTemplate.from_messages([("system", system), ("human", human)])
chain = prompt | chat
chain.invoke({"question": "what bear is best?"})
AIMessage(content='{ "response": "That\'s a tough question! There are eight species of bears found in the world, and each one is unique and amazing in its own way. However, if I had to pick one, I\'d say the giant panda is a popular favorite among many people. Who can resist those adorable black and white markings?", "followup_question": "Would you like to know more about the giant panda\'s habitat and diet?" }', response_metadata={'token_usage': {'completion_tokens': 89, 'prompt_tokens': 50, 'total_tokens': 139, 'completion_time': 0.249032839, 'prompt_time': 0.011134497, 'queue_time': None, 'total_time': 0.260167336}, 'model_name': 'llama3-70b-8192', 'system_fingerprint': 'fp_2f30b0b571', 'finish_reason': 'stop', 'logprobs': None}, id='run-558ce67e-8c63-43fe-a48f-6ecf181bc922-0')
