iCdZddlmZddlZddlmZmZmZmZm Z m Z ddl m Z ddl mZmZddlmZddlmZdd lmZer dd lmZdd lmZGd d eeZy)aG Vertex AI-specific RAG Ingestion implementation. Vertex AI RAG Engine handles embedding and chunking internally when files are uploaded, so this implementation skips the embedding step and directly uploads files to RAG corpora. Based on: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/model-reference/rag-api-v1 ) annotationsN) TYPE_CHECKINGAnyDictListOptionalTuple)verbose_logger)get_async_httpx_clienthttpxSpecialProvider)get_vertex_base_url) VertexBase)BaseRAGIngestion)Router)RAGIngestOptionsceZdZdZ d d dZ d dZ d dZ d ddZ d ddZ ddZ dd Z y)VertexAIRAGIngestiona Vertex AI RAG Engine ingestion implementation. Key differences from base: - Embedding is handled by Vertex AI RAG Engine when files are uploaded - Files are uploaded using the RAG API (import or upload) - Chunking is done by Vertex AI RAG Engine (supports custom chunking config) - Supports Google Cloud Storage (GCS) and Google Drive sources - Supports custom parsing configurations (layout parser, LLM parser) Nctj|||tj|t|j}|j ||_|j|xsd|_|j||_ y)N)ingest_optionsrouterz us-central1) r__init__rdictvector_store_configsafe_get_vertex_ai_project project_idget_vertex_ai_locationlocationsafe_get_vertex_ai_credentialsvertex_credentials)selfrrlitellm_paramss z/Users/manta/Documents/Projects/TheRoad-I1/.venv/lib/python3.12/site-packages/litellm/rag/ingestion/vertex_ai_ingestion.pyrzVertexAIRAGIngestion.__init__)st !!$~fUD!d66799.I33NCT} "&"E"En"Uc Kyw)z Vertex AI RAG Engine handles embedding internally - skip this step. Returns: None (Vertex AI embeds when files are uploaded to RAG corpus) N)r chunkss r"embedzVertexAIRAGIngestion.embed9s scPK|js td|jjd}|sB|j |j xsd|jjdd{}d}|r!|r|r|j ||||d{}||fS7-7 w)a  Store content in Vertex AI RAG corpus. Vertex AI workflow: 1. Create RAG corpus (if not provided) 2. Upload file using RAG API (Vertex AI handles chunking/embedding) Args: file_content: Raw file bytes filename: Name of the file content_type: MIME type chunks: Ignored - Vertex AI handles chunking embeddings: Ignored - Vertex AI handles embedding Returns: Tuple of (rag_corpus_id, file_id) zVvertex_project is required for Vertex AI RAG ingestion. Set it in vector_store config.vector_store_idzlitellm-rag-corpus description) display_namer*N) rag_corpus_idfilename file_content content_type)r ValueErrorrget_create_rag_corpus ingest_name_upload_file_to_corpus)r r.r-r/r& embeddingsr,result_file_ids r"storezVertexAIRAGIngestion.storeFs21  00445FG "&"9"9!--E1E 4488G#:#M  H#'#>#>+!)) $?$Nn,,s$A2B&4B"5$B&B$ B&$B&cK|j|j|jd\}}|js||_t|j}|d|jd|jd}d|i}|r||d<|j j d}|r||d <|j j d } | rd |vri|d <d d | ii|d d <tjd|tjdtj|dttjddi} | j||d|ddd{} | jdvr/d| j } tj"| t%| | j} tjdtj| d| j dr#| j dij dd }nE| j dd }tjd!||j'||"d{}tjd#||S77!w)$a Create a Vertex AI RAG corpus. Args: display_name: Display name for the corpus description: Optional description Returns: RAG corpus ID (format: projects/{project}/locations/{location}/ragCorpora/{corpus_id}) vertex_ai credentialsrcustom_llm_providerz/v1beta1/projects/z /locations/z /ragCorpora displayNamer*vector_db_configvectorDbConfigembedding_modelvertexPredictionEndpointendpointragEmbeddingModelConfigzCreating RAG corpus: Request body: indenttimeoutN@ llm_providerparamsBearer application/json Authorizationz Content-TypejsonheadersNzFailed to create RAG corpus: zCreate corpus response: doneresponsenamezPolling operation: )operation_name access_tokenzCreated RAG corpus: )_ensure_access_tokenrrr rrr1r debugrRdumpsr r RAGpost status_codetexterror Exception_poll_operation)r r+r*r\rbase_urlurl request_bodyr>r@clientrX error_msg response_data corpus_namer[s r"r2z'VertexAIRAGIngestion._create_rag_corpusys $(#<#<// +$=$  j(DO't}}5j( DMM?+ O  <(  *5L ' 33778JK -=L) *22667HI |313 -.*Z,IIL) *+D E 4SE:;~djja.P-QRS'-11t$    #*<.!9 2%     z 17 GI   +I& &  &tzz-'J&K L   V $'++J;??KK+..vr:N  #6~6F!G H $ 4 4-)!5!K 3K=ABC 6s%EI*I%C2I*I(I*(I*cKddl}t|j}|d|}ttj ddi}t |D]} |j|dd|i d{} | jd k7r/d | j} tj| t| | j} | jd rMd | vr| d } td| | jdijdd}|r|cStd| tjd| dzd||j|d{ td|d77w)a Poll a long-running operation until it completes. Args: operation_name: The operation name (e.g., "operations/123456") access_token: Access token for authentication max_retries: Maximum number of polling attempts retry_delay: Delay between polling attempts in seconds Returns: The corpus name from the completed operation Raises: Exception: If operation fails or times out rN /v1beta1/rHrIrJrPrM)rSrUzFailed to poll operation: rWrdzOperation failed: rXrYrZz&No corpus name in operation response: z Operation not done yet, attempt /zOperation timed out after z attempts)asyncior rr r r`ranger1rbrcr rdrerRr^sleep)r r[r\ max_retries retry_delayrrrgrhrjattemptrXrkoperation_datardrms r"rfz$VertexAIRAGIngestion._poll_operations, &t}}5 )N#34'-11t$  [)G#ZZ#w|n%=(H##s*8H $$Y/ **%]]_N!!&)n,*73E#&8$@AA-00R@DDVRP &&#@@PQ  27Q;-q N -- , , ,C*F4[MKLLE@ -s%A(E(*E$+C"E( E&E(&E(cK|j|j|jd\}}t|j}|d|d}dd|ii} |j j d} | r| | dd<|j} | rgt| trW| j d } | j d } | s| r1d | vri| d <d d iii| d d<| d dd d }| r| |d <| r| |d <tjd|tjdtj| ddtj| df|||xsdfd}ttj ddi}|j#||d|ddd{}|j$dvr/d|j&}tj(|t+| |j}|j d ij d!d"}|s|j d!d"}tjd#||S7#t*$r"}tj,d$|Yd}~y%d}~wwxYww)&aA Upload a file to Vertex AI RAG corpus using multipart upload. Args: rag_corpus_id: RAG corpus resource name filename: Name of the file file_content: File content bytes content_type: MIME type Returns: File ID or resource name r9r:z/upload/v1beta1/z/ragFiles:uploadrag_filer+file_descriptionr* chunk_size chunk_overlapupload_rag_file_configrag_file_chunking_configfixed_length_chunkingrag_file_transformation_configzUploading file to RAG corpus: z Metadata: rErFNrNzapplication/octet-stream)metadatafilerHgr@rJrM multipart)rPzX-Goog-Upload-Protocol)filesrSrTzFailed to upload file: ragFilerYrZzUpload complete. File ID: z!Could not parse upload response: uploaded)r]rrr rrr1chunking_strategy isinstancerr r^rRr_r r r`rarbrcrdrewarning)r r,r-r.r/r\_rgrhrr*rr|r}chunking_configrrjrXrkrlfile_ides r"r4z+VertexAIRAGIngestion._upload_file_to_corpus"s(33// +4  a't}}5 *}o=MN $ ..223EF 2=HZ  /!22 ,=t!D*..|OL1 7DOO4=cUCDz$**Xa*H)IJKtzz(35GH: : (-11u%    #*<.!9*5%     z 11(--AI   +I& & $MMOM#'' 26::62FG'++FB7  #=gY!G HN/ 0   " "%Fqc#J K s>F I H/ AIAH1.I1 I:IIIIcK|j|j|jd\}}t|j}|d|d}ddd|iii}|j }|rIt |tr9|jd} |jd } | s| r| xsd | xsd d |dd <|jjd} | r| |dd<tjd|tjdtj|dttj ddi} | j#||d|ddd{} | j$dvr/d| j&}tj(|t+|| j}|jdd}tjd ||S7}w)!a Import files from Google Cloud Storage into RAG corpus. Args: rag_corpus_id: RAG corpus resource name gcs_uris: List of GCS URIs (e.g., ["gs://bucket/file.pdf"]) Returns: Operation name for tracking import progress r9r:roz/ragFiles:importimportRagFilesConfig gcsSourceurisr|r}irU) chunkSize chunkOverlapragFileChunkingConfigmax_embedding_requests_per_minmaxEmbeddingRequestsPerMinzImporting files from GCS: rDrErFrHrIrJrMrNrOrQNrTzFailed to import files: rYrZzImport operation started: )r]rrr rrrrr1rr r^rRr_r r r`rarbrcrdre)r r,gcs_urisr\rrgrhrirr|r}max_embedding_qpmrjrXrkrlr[s r"_import_files_from_gcsz+VertexAIRAGIngestion._import_files_from_gcss  33// +4  a't}}5 ) 6FG #[682D$E( !22 ,=t!D*..|G)N)rz'RAGIngestOptions'rzOptional['Router'])r& List[str]returnOptional[List[List[float]]]) r.zOptional[bytes]r- Optional[str]r/rr&rr5rrz#Tuple[Optional[str], Optional[str]])r+strr*rrr)g@) r[rr\rruintrvfloatrr) r,rr-rr.bytesr/rrr)r,rrrrr) __name__ __module__ __qualname____doc__rr'r7r2rfr4rr%r#r"rrsS &*V*V#V   % 1-%1- 1-$ 1-  1- 0 1- -1-l&*``#` `L EMEMEM EM  EM EMNlll l $ l l\NNN Nr#r)r __future__rrRtypingrrrrrr litellm._loggingr &litellm.llms.custom_httpx.http_handlerr r #litellm.llms.vertex_ai.common_utilsr &litellm.llms.vertex_ai.vertex_llm_baser$litellm.rag.ingestion.base_ingestionrlitellmrlitellm.types.ragrrr%r#r"rsF# BB+D=A2A+ZAr#