MD2204/multi_modality

1o

2��wi��@s`dd
lm
Z

ddlmZ
ddlmZ
ddlmZ
ddlm	Z	
defdd�Z3d	ed4efdd�Zd
S)�)
�RetrievalQA)
�PromptTemplate)
�
BaseRetriever)
�
BaseChatModel)
�RETRIEVER_K�returnc

Cs|jd
ddi
d�S)zEGet retriever from vector store with higher K for multimodal context.�5similarity�
k�)�search_type�
search_kwargsN)
�as_retriever)
�vectorstore�r�GC:\Users\MoEdBouk\Desktop\AI ENGENEER PATH\multimodality_RAG\src\rag.py�
get_retrievers

�r�llm�	retrievercCs0d
}t|ddgd�}t
j|d|
dd|i
d�}|S)6z$Create a multimodal-aware RAG chain.aBYou are a highly capable AI Assistant. Use ONLY the provided context to answer the user's question. Do not use any external knowledge about the DeepSeek paper or Reinforcement Learning results.7 8CRITICAL INSTRUCTIONS:9- Some context blocks are marked as [Visual Data Transcription]. These contain data extracted from images, charts, and tables. Treat these as the absolute source of truth.10- If the user asks for specific values, read the row and column of the Markdown tables exactly.11- Numeric Format Check: Ensure the numbers you report match the scale in the context (e.g., if the context shows thousands like 6500, do not report decimals like 0.65).12- If the answer is not in the context, say "I don't know based on the provided data." Do not guess or use memory.13 14    Context:15    {context}16 17    Question: {question}18    Answer:�context�question)�templateZinput_variables�stuffT�prompt)rZ19chain_typerZreturn_source_documentsZchain_type_kwargsN)rrZfrom_chain_type)rr�prompt_template�PROMPT�qa_chainrrr�create_rag_chains

�




�rN)Zlangchain_classic.chainsr�langchain_core.promptsr�langchain_core.retrieversrZlangchain_core.language_modelsr�20src.configrrrrrrr�<module>
s