Workflow updated with the use of LLMs (using Amazon Bedrock) #3
Conversation
|
|
||
| from langchain.prompts import PromptTemplate | ||
| from langchain.llms.bedrock import Bedrock | ||
| from langchain_community.chat_models import BedrockChat |
There was a problem hiding this comment.
The official package for accessing bedrock models is now langchain-aws
There was a problem hiding this comment.
Fixed in commit e611804
| 'claude': 'anthropic.claude-3-haiku-20240307-v1:0', | ||
| } | ||
|
|
||
| logging.getLogger().setLevel(os.environ.get('LOG_LEVEL', 'WARNING').upper()) |
There was a problem hiding this comment.
Have you considered using Powertools for Lambda? They offer a cool structured logging convenience.
There was a problem hiding this comment.
Fixed in commit e611804
| augmented_json_format_str = json.dumps(json_format) | ||
|
|
||
| logging.info(f'Extract data prompt') | ||
| logging.info(extract_data_prompt.format(json_format=augmented_json_format_str, |
There was a problem hiding this comment.
Have you considered using a few shot prompting template?
There was a problem hiding this comment.
Added in commit e611804
|
|
||
| extract_data_prompt = ChatPromptTemplate.from_messages(messages_data) | ||
|
|
||
| chain_extract_data = extract_data_prompt | llm_data | StrOutputParser() |
There was a problem hiding this comment.
Have you considered using structured output or tools to do info extraction?
There was a problem hiding this comment.
Added in commit e611804
There was a problem hiding this comment.
Hi david, I think overall my original comments were neatly addressed and this looks a lot more "curent" in terms of how to use langchain, very good work.
I left some comments, they are all minor, so I will go ahead and approve, take in the comments and, if you want apply them, otherwise, feel free to merge.
architecture.pptx
Outdated
There was a problem hiding this comment.
Maybe this could be a pdf, seems more inclusive, not everyone has ppt
There was a problem hiding this comment.
The PPT is required by the solutions team but I'm including a PDF version anyway.
There was a problem hiding this comment.
Added PDF diagram in 0aa1ba7
| item = event | ||
|
|
||
| logger.info('Item:') | ||
| logger.info(item) |
There was a problem hiding this comment.
We are already logging the entire event on line 124. Maybe you could reconsider one of them.
|
|
||
| #Attemp to categorize item | ||
| text = item['text'] | ||
| logger.info(f'Text: {text}') |
| logger.info(f'Text: {text}') | ||
|
|
||
| text = demoji.replace(text, "") | ||
| item['text_clean'] = text |
|
|
||
| try: | ||
|
|
||
| #meta_topics_str = ','.join(META_TOPICS) |
| text: str | ||
| ) -> ExtractedInformation: | ||
|
|
||
| bedrock_llm = ChatBedrock( |
There was a problem hiding this comment.
Have you considered extracting this instantiation to root level so the client is created only once per lifetime of the lambda function?
There was a problem hiding this comment.
Cant extract from the function since the parameters are different for each instance of the client.
|
|
||
| claude_information_extraction_prompt_template = INFORMATION_EXTRACTION_PROMPT_SELECTOR.get_prompt(MODEL_ID) | ||
|
|
||
| print("The prompt template") |
There was a problem hiding this comment.
Is there a reason for using prints when you have a logger instance? It might also be an opportunity to have some logs as INFO and others as DEBUG.
|
|
||
| text_insight = text_information_extraction(META_SENTIMENTS_STR, item['text_clean']) | ||
| logger.info(f'Text insights:') | ||
| logger.info(text_insight) |
There was a problem hiding this comment.
You seem to be double-logging some things, for example line 119 logs the same as this line 150.
| }) | ||
|
|
||
| print("Information extraction object") | ||
| print(type(information_extraction_obj)) |
There was a problem hiding this comment.
Is this necessary? We know the type is ExtractedInformation since we are using structured outputs
| region = args.region | ||
|
|
||
| sqs = boto3.client('sqs') | ||
| translate = boto3.client(service_name='translate', region_name=region, use_ssl=True) |
There was a problem hiding this comment.
This is optional only for sending a test load to the solution and the translation part is only used if the user decides to test for posts in Spanish
|
Solution upgraded to support the use of LLMs to make information extraction more efficient: Description of changes:
|
Issue #, if available:
Description of changes:
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.