-
Checked other resources
Commit to Help
Example Codeclass AnswerWithJustification(pydantic.v1.BaseModel):
"""An answer to the user question along with justification for the answer."""
answer: str
justification: str
sample_prompt = "What weighs more a pound of bricks or a pound of feathers"
...
# you fill in this part
...
azure_openai_result = azure_openai_configured_structured_llm.invoke(sample_prompt)
print(f"{azure_openai_result=}")
bedrock_claude_result = bedrock_claude_configured_structured_llm.invoke(sample_prompt)
print(f"{bedrock_claude_result=})DescriptionI'd like an example of how, using the new langchain.chat_models.base.init_chat_model function, I can make a chain that has an LLM configured to use structured output and which lets me configure at runtime whether to use System InfoSystem Information
Package Information
|
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 3 replies
-
|
To switch between from langchain.chat_models import init_chat_model
import pydantic
class AnswerWithJustification(pydantic.v1.BaseModel):
"""An answer to the user question along with justification for the answer."""
answer: str
justification: str
sample_prompt = "What weighs more a pound of bricks or a pound of feathers"
# Initialize AzureChatOpenAI model
azure_openai_configured_structured_llm = init_chat_model(
"gpt-4o",
model_provider="AzureChatOpenAI",
temperature=0.7,
azure_deployment="your_deployment_name", # Replace with your actual deployment name
azure_endpoint="https://your-endpoint.azure.com/" # Replace with your actual endpoint
)
# Initialize BedrockChat model
bedrock_claude_configured_structured_llm = init_chat_model(
"amazon.some-model",
model_provider="BedrockChat",
temperature=0.7
)
# Use the models
azure_openai_result = azure_openai_configured_structured_llm.invoke(sample_prompt)
print(f"{azure_openai_result=}")
bedrock_claude_result = bedrock_claude_configured_structured_llm.invoke(sample_prompt)
print(f"{bedrock_claude_result=}")In this example:
This setup allows you to dynamically choose between |
Beta Was this translation helpful? Give feedback.
-
|
@dosu a few comments
When using |
Beta Was this translation helpful? Give feedback.
-
|
To address your comments and ensure the correct configuration for both
from langchain.chat_models import init_chat_model
import pydantic
class AnswerWithJustification(pydantic.v1.BaseModel):
"""An answer to the user question along with justification for the answer."""
answer: str
justification: str
sample_prompt = "What weighs more a pound of bricks or a pound of feathers"
# Initialize AzureChatOpenAI model
azure_openai_configured_structured_llm = init_chat_model(
"gpt-4o",
model_provider="azure_openai",
deployment_name="your_deployment_name", # Replace with your actual deployment name
azure_endpoint="https://your-endpoint.azure.com/" # Replace with your actual endpoint
)
# Initialize BedrockChat model
bedrock_claude_configured_structured_llm = init_chat_model(
"amazon.some-model",
model_provider="bedrock",
model_kwargs={'temperature': 0.7}
)
# Use the models
azure_openai_result = azure_openai_configured_structured_llm.invoke(sample_prompt)
print(f"{azure_openai_result=}")
bedrock_claude_result = bedrock_claude_configured_structured_llm.invoke(sample_prompt)
print(f"{bedrock_claude_result=}")This setup ensures: |
Beta Was this translation helpful? Give feedback.
-
|
What I'm after is something that can switch between azure and bedrock structured output LLMs using nothing but configuration. I figured out how to do that on my own after some experimentation. Here's a passing pytest for me. def test_init_chat_model_basic_example():
from langchain.chat_models import init_chat_model
import pydantic
class AnswerWithJustification(pydantic.v1.BaseModel):
"""An answer to the user question along with justification for the answer."""
answer: str
justification: str
sample_prompt = "What weighs more a pound of bricks or a pound of feathers"
azure_openai_gpt_35_turbo_deployment_name = "<YOUR DEPLOYMENT NAME HERE>"
bedrock_claude_3_haiku_model_id = "anthropic.claude-3-haiku-20240307-v1:0"
configurable_model = init_chat_model()
azure_configuration = configurable_model.with_config(
configurable={
"model_provider": "azure_openai",
"model": azure_openai_gpt_35_turbo_deployment_name,
"temperature": 0.7,
"deployment_name": azure_openai_gpt_35_turbo_deployment_name,
}
)
structured_azure_configuration = azure_configuration.with_structured_output(
AnswerWithJustification
)
bedrock_configuration = configurable_model.with_config(
configurable={
"model": bedrock_claude_3_haiku_model_id,
"model_provider": "bedrock",
"model_kwargs": {"temperature": 0},
"model_id": bedrock_claude_3_haiku_model_id,
}
)
structured_bedrock_configuration = bedrock_configuration.with_structured_output(
AnswerWithJustification
)
azure_result = structured_azure_configuration.invoke(sample_prompt)
bedrock_result = structured_bedrock_configuration.invoke(sample_prompt)
assert isinstance(azure_result, AnswerWithJustification)
assert isinstance(bedrock_result, AnswerWithJustification)
print(f"{azure_result=}")
# azure_result=AnswerWithJustification(answer='A pound of bricks and a pound of feathers weigh the same.',
# justification='Both a pound of bricks and a pound of feathers have the same weight of one pound. The weight is
# determined by the pound unit, which is the same for both bricks and feathers.')
print(f"{bedrock_result=}")
# bedrock_result=AnswerWithJustification(answer='A pound of bricks and a pound of feathers weigh the same
# amount.', justification='By definition, a pound is a unit of mass or weight. One pound of any material,
# whether it is bricks or feathers, will weigh the same. This is because the pound is a standard unit of
# measurement that represents a fixed amount of mass, regardless of the material. So a pound of bricks and a
# pound of feathers would both weigh exactly one pound.') |
Beta Was this translation helpful? Give feedback.
What I'm after is something that can switch between azure and bedrock structured output LLMs using nothing but configuration. I figured out how to do that on my own after some experimentation. Here's a passing pytest for me.