OpenCode
68a9071cf1
Local branch had diverged from origin/master (sibling commits on the same base). Rewrote local history linearly on top of origin/master, folding in all local content: exa/hot-platform discovery hunters, leaks ledger, nightly sweep orchestrator, updated .gitignore and skill docs, plus the latest hunt output and vault state. Remote-only files (vault keys, channel scripts) were restored rather than dropped, so the resulting tree is a full union of both sides.
1 line
75 KiB
JSON
1 line
75 KiB
JSON
{"repo": "aiandcivilization/localrag", "files": {"config/test_config.yaml": {"0c96c6986e8be56bda16742ea2419fee557efefd": {"t": 1785746823.4376814, "n": 12812, "h": "4d089c9dac5062ab", "c": "#config path\nconfig_path: \"config/config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" # Provider of the vector database (Qdrant or Faiss)\npersist_directory: \"data/vector_stores/qdrant_semantic_test_2\" #\"data/vector_stores/qdrant_semantic_bge_jinai\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"data/politique\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\ncollection_name: \"qdrant_vectorstore\" # Name of the vector collection\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n\n# Chunking parameters (only for recursive splitting)\nsplitting_method: \"constant\" # Method to split the documents into chunks (semantic or \"recursive\")\nchunk_size: 5000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 65 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 7 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (semantic + lexical, merging the results)\ndeep_search: false # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content, expert opinions on political, geopolitical and economic issues\",\n \"emails\": \"Emails, chat messages, and other text-based communication\",\n \"jobs\": \"job descriptions, or job offers\",\n \"food\": \"Food,food recipes, and cooking instructions\",\n \"linkedin\": \"LinkedIn posts, technical articles, data science and AI content\",\n \"prompts\": \"refined prompts made by the user\",\n \"sheets\": \"Google sheets, excel files, and other tabular data\",\n \"charts\": \"Charts, graphs, and diagrams containing curves, lines, tendencies showing economic, political, or social data\",\n}\n\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"Snowflake/snowflake-arctic-embed-l-v2.0\" #\"jinaai/jina-embeddings-v3\" #\"Snowflake/snowflake-arctic-embed-l-v2.0\" #\"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\ndense_embedding_size: 1024 #Size of the dense embedding model\nsparse_embedding_model: \"Qdrant/bm42-all-minilm-l6-v2-attentions\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 30522 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 8000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"groq\" #\"github\" #\"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gemini-2.0-flash-exp\" #\"meta-Llama-3.1-405B-Instruct\" #\"llama3-405b\" #gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\"\nmodels_dict: {\"gemini-2.0-flash-exp\":\"google\",\n \"deepseek-r1-distill-llama-70b\":\"groq\",\n \"llama-3.3-70b-versatile\":\"groq\",\n \"gpt-4o\": \"github\",\n \"meta-Llama-3.1-405B-Instruct\": \"github\",\n \"gpt-4o-mini\": \"github\",\n \"o1-mini\": \"github\",\n \"o1-preview\": \"github\",\n \"llama3-405b\": \"sambanova\",\n \"llama-3.1-8b-instant\": \"groq\",\n \"llama-3.2-3b-preview\": \"groq\",\n \"gemini-1.5-pro\":\"google\",\n \"gemini-1.5-flash\":\"google\",\n \"gemini-exp-1206\":\"google\",\n \"DeepSeek-R1\":\"github\",\n \"gemini-2.0-flash-thinking-exp-01-21\":\"google\"\n \n}\n\n#Prompting parameters\nfr_rag_prompt_path: \"prompts/base_rag_prompt_fr.txt\" #Path to the file containing the RAG prompt\nen_rag_prompt_path: \"prompts/base_rag_prompt_en.txt\" #Path to the file containing the RAG prompt\n\n#vllm parameters\nvllm_model_name: \"gemini-exp-1206\" #VLLM model to use for the generation\nvllm_provider: \"gemini\" # Provider of the VLLM model\n\n#LLM parameters\ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\nllm_token_target: 0 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions_dict: {\n \"employer contact writing\": \"Writing professional communication like cover letters and outreach messages to potential employers\",\n \"email support inquiry\": \"User asks a question (whatever the subject), asking explicitely to use emails as information source.\",\n \"job search assistance\": \"Providing guidance and help with finding job opportunities, including search strategies and job board navigation\",\n \"document content question\": \"Addressing general queries about document content, or any particular request that is not covered properly by the other actions\",\n \"graph creation request\": \"User asks explicitelly to create charts, graphs, and diagrams\",\n \"meeting notes summarization\": \"Converting meeting recordings or raw notes into concise, structured summaries\",\n #\"previous answer correction\": \"User asks to correct a previous answer, modifying an already created content, shortening or expanding it, or changing the content or wanting to add more information that was not included in the previous answer (e.g. user starts his prompt with 'no, more details about...','make this shorter', 'no i search for...') the object of the modification is not explicitely present in his query.\",\n \"prompt engineering request\": \"User asks the system to generate a detailed prompt for articial image creation, refining the prompt until it meets the user's needs\",\n \"political question\": \"User asks a question about political, geopolitical, or economic issues, or about a personnality, or ask the assistant to do a task related to political thinking\",\n \"sheet or table info extraction\":\"User asks to extract information from a table or a sheet, like google sheets, excel files, or any tabular data\",\n}\n\n#Agentic RAG parameters\nquery_breaker_model: \"o1-mini\"\nquery_breaker_provider: \"github\"\n\n#Intent classifier parameters\nquery_classification_model: \"gemini-2.0-flash-exp\"\nquery_classification_provider: \"google\"\n\n#Transcription parameters\ntranscription_chunk_size: 1800 #corresponds to 30 mins of audio per chunks (size is limited to 25Mb /request)"}, "20dce8f6668d5a8a57345290dbf8f02a04f84a4f": {"t": 1785767887.3692775, "n": 12812, "h": "4d089c9dac5062ab", "c": "#config path\nconfig_path: \"config/config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" # Provider of the vector database (Qdrant or Faiss)\npersist_directory: \"data/vector_stores/qdrant_semantic_test_2\" #\"data/vector_stores/qdrant_semantic_bge_jinai\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"data/politique\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\ncollection_name: \"qdrant_vectorstore\" # Name of the vector collection\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n\n# Chunking parameters (only for recursive splitting)\nsplitting_method: \"constant\" # Method to split the documents into chunks (semantic or \"recursive\")\nchunk_size: 5000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 65 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 7 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (semantic + lexical, merging the results)\ndeep_search: false # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content, expert opinions on political, geopolitical and economic issues\",\n \"emails\": \"Emails, chat messages, and other text-based communication\",\n \"jobs\": \"job descriptions, or job offers\",\n \"food\": \"Food,food recipes, and cooking instructions\",\n \"linkedin\": \"LinkedIn posts, technical articles, data science and AI content\",\n \"prompts\": \"refined prompts made by the user\",\n \"sheets\": \"Google sheets, excel files, and other tabular data\",\n \"charts\": \"Charts, graphs, and diagrams containing curves, lines, tendencies showing economic, political, or social data\",\n}\n\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"Snowflake/snowflake-arctic-embed-l-v2.0\" #\"jinaai/jina-embeddings-v3\" #\"Snowflake/snowflake-arctic-embed-l-v2.0\" #\"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\ndense_embedding_size: 1024 #Size of the dense embedding model\nsparse_embedding_model: \"Qdrant/bm42-all-minilm-l6-v2-attentions\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 30522 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 8000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"groq\" #\"github\" #\"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gemini-2.0-flash-exp\" #\"meta-Llama-3.1-405B-Instruct\" #\"llama3-405b\" #gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\"\nmodels_dict: {\"gemini-2.0-flash-exp\":\"google\",\n \"deepseek-r1-distill-llama-70b\":\"groq\",\n \"llama-3.3-70b-versatile\":\"groq\",\n \"gpt-4o\": \"github\",\n \"meta-Llama-3.1-405B-Instruct\": \"github\",\n \"gpt-4o-mini\": \"github\",\n \"o1-mini\": \"github\",\n \"o1-preview\": \"github\",\n \"llama3-405b\": \"sambanova\",\n \"llama-3.1-8b-instant\": \"groq\",\n \"llama-3.2-3b-preview\": \"groq\",\n \"gemini-1.5-pro\":\"google\",\n \"gemini-1.5-flash\":\"google\",\n \"gemini-exp-1206\":\"google\",\n \"DeepSeek-R1\":\"github\",\n \"gemini-2.0-flash-thinking-exp-01-21\":\"google\"\n \n}\n\n#Prompting parameters\nfr_rag_prompt_path: \"prompts/base_rag_prompt_fr.txt\" #Path to the file containing the RAG prompt\nen_rag_prompt_path: \"prompts/base_rag_prompt_en.txt\" #Path to the file containing the RAG prompt\n\n#vllm parameters\nvllm_model_name: \"gemini-exp-1206\" #VLLM model to use for the generation\nvllm_provider: \"gemini\" # Provider of the VLLM model\n\n#LLM parameters\ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\nllm_token_target: 0 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions_dict: {\n \"employer contact writing\": \"Writing professional communication like cover letters and outreach messages to potential employers\",\n \"email support inquiry\": \"User asks a question (whatever the subject), asking explicitely to use emails as information source.\",\n \"job search assistance\": \"Providing guidance and help with finding job opportunities, including search strategies and job board navigation\",\n \"document content question\": \"Addressing general queries about document content, or any particular request that is not covered properly by the other actions\",\n \"graph creation request\": \"User asks explicitelly to create charts, graphs, and diagrams\",\n \"meeting notes summarization\": \"Converting meeting recordings or raw notes into concise, structured summaries\",\n #\"previous answer correction\": \"User asks to correct a previous answer, modifying an already created content, shortening or expanding it, or changing the content or wanting to add more information that was not included in the previous answer (e.g. user starts his prompt with 'no, more details about...','make this shorter', 'no i search for...') the object of the modification is not explicitely present in his query.\",\n \"prompt engineering request\": \"User asks the system to generate a detailed prompt for articial image creation, refining the prompt until it meets the user's needs\",\n \"political question\": \"User asks a question about political, geopolitical, or economic issues, or about a personnality, or ask the assistant to do a task related to political thinking\",\n \"sheet or table info extraction\":\"User asks to extract information from a table or a sheet, like google sheets, excel files, or any tabular data\",\n}\n\n#Agentic RAG parameters\nquery_breaker_model: \"o1-mini\"\nquery_breaker_provider: \"github\"\n\n#Intent classifier parameters\nquery_classification_model: \"gemini-2.0-flash-exp\"\nquery_classification_provider: \"google\"\n\n#Transcription parameters\ntranscription_chunk_size: 1800 #corresponds to 30 mins of audio per chunks (size is limited to 25Mb /request)"}, "e59d9ea3c148276cc9c2670e9c0dbbe3f5436302": {"t": 1785767888.5479221, "n": 8890, "h": "12820d6288f38482", "c": "#config path\nconfig_path: \"config/test_config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" #'Chroma'\npersist_directory: \"data/vector_stores/qdrant_semantic_snowflake\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\"\nsplitting_method: \"semantic\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"data/emails\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n# Chunking parameters (only for recursive splitting)\nchunk_size: 1000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 85 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 10 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nauto_hybrid_search: false # Enable or disable the ability for the model to find itself the keyword\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (merge retrievers + routing if activated)\ndeep_search: true # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content\",\n \"emails\": \"Emails\",\n \"jobs\": \"Jobs\",\n}\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"Snowflake/snowflake-arctic-embed-l-v2.0\" #\"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\nsparse_embedding_model: \"naver/splade-v3\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 1024 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 4000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\" \ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\nllm_token_target: 7000 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions: [\"rediger un texte pour une offre\",\"question sur des emails\",\"rechercher des offres d'emploi\",\"question normale sur des documents\",\"write instagram description\"]"}, "2db8efec808884558f95b443017747da18cd2427": {"t": 1785767889.7961922, "n": 8846, "h": "b950be63e7bc0007", "c": "#config path\nconfig_path: \"config/test_config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" #'Chroma'\npersist_directory: \"data/vector_stores/qdrant_semantic_bge_test\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\"\nsplitting_method: \"semantic\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"data/emails\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n# Chunking parameters (only for recursive splitting)\nchunk_size: 1000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 85 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 10 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nauto_hybrid_search: false # Enable or disable the ability for the model to find itself the keyword\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (merge retrievers + routing if activated)\ndeep_search: true # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content\",\n \"emails\": \"Emails\",\n \"jobs\": \"Jobs\",\n}\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\nsparse_embedding_model: \"naver/splade-v3\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 1024 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 4000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\" \ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\nllm_token_target: 7000 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions: [\"rediger un texte pour une offre\",\"question sur des emails\",\"rechercher des offres d'emploi\",\"question normale sur des documents\",\"write instagram description\"]"}, "4b13b0d22cfe321e8735c0e2db9afd9275c66583": {"t": 1785767891.0016973, "n": 8902, "h": "d1c547aec0c0abb3", "c": "#config path\nconfig_path: \"config/test_config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" #'Chroma'\npersist_directory: \"data/vector_stores/qdrant_semantic_bge_test\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\"\nsplitting_method: \"semantic\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"data/emails\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\nuse_server: \"False\"\nserver_url: \"http://localhost:6333\"\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n# Chunking parameters (only for recursive splitting)\nchunk_size: 1000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 85 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 10 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nauto_hybrid_search: false # Enable or disable the ability for the model to find itself the keyword\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (merge retrievers + routing if activated)\ndeep_search: true # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content\",\n \"emails\": \"Emails\",\n \"jobs\": \"Jobs\",\n}\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\nsparse_embedding_model: \"naver/splade-v3\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 1024 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 4000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\" \ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\nllm_token_target: 7000 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions: [\"rediger un texte pour une offre\",\"question sur des emails\",\"rechercher des offres d'emploi\",\"question normale sur des documents\",\"write instagram description\"]"}, "8a5541d6ddf2bf9319d3c0c9318688b0aa5667c3": {"t": 1785767892.1773565, "n": 8846, "h": "b950be63e7bc0007", "c": "#config path\nconfig_path: \"config/test_config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" #'Chroma'\npersist_directory: \"data/vector_stores/qdrant_semantic_bge_test\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\"\nsplitting_method: \"semantic\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"data/emails\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n# Chunking parameters (only for recursive splitting)\nchunk_size: 1000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 85 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 10 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nauto_hybrid_search: false # Enable or disable the ability for the model to find itself the keyword\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (merge retrievers + routing if activated)\ndeep_search: true # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content\",\n \"emails\": \"Emails\",\n \"jobs\": \"Jobs\",\n}\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\nsparse_embedding_model: \"naver/splade-v3\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 1024 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 4000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\" \ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\nllm_token_target: 7000 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions: [\"rediger un texte pour une offre\",\"question sur des emails\",\"rechercher des offres d'emploi\",\"question normale sur des documents\",\"write instagram description\"]"}}, "config/movie_studio_config.yaml": {"0c96c6986e8be56bda16742ea2419fee557efefd": {"t": 1786068509.9492974, "n": 12482, "h": "cdf348ce0e67d5c2", "c": "#config path\nconfig_path: \"config/config.yaml\"\n\n# Database initialization parameters\nvectordb_provider: \"Qdrant\" # Provider of the vector database (Qdrant or Faiss)\npersist_directory: \"data/vector_stores/qdrant_movie_studio\" #\"data/vector_stores/qdrant_semantic_bge_jinai\" #Directory to store the index (give the name you want, just dont change \"data/vector_stores/\" # Method to split the documents into chunks (semantic or \"recursive\")\npath: \"subprojects/movie_studio/data_movie_studio\" # Path of the root folder of the documents you want to put in the index \nprocess_log_file: \"processed_docs.log\"\ncollection_name: \"qdrant_vectorstore\" # Name of the vector collection\n\n\n\n#Knowledge graph index parameters\nbuild_knowledge_graph: false # Enable or disable the building of the knowledge graph when initializing the database\nnb_nodes: 1 # Number of nodes to retrieve from the knowledge graph when asking a query (analogous to the number of chunks)\nallow_kg_retrieval: false # Enable or disable the retrieval of info from the knowledge graph during the generation\nkg_target_context: 1500\nmin_relations_pct: 0.1\nmin_descriptions_pct: 0.2\nmin_communities_pct: 0.2\n\n\n# Chunking parameters (only for recursive splitting)\nsplitting_method: \"constant\" # Method to split the documents into chunks (semantic or \"recursive\")\nchunk_size: 5000 # Chunk size for the documents in ters of number of characters (only for recursive splitting)\nchunk_overlap: 0 # Overlap between chunks in terms of number of characters (only for recursive splitting)\nsemantic_threshold: 65 # Threshold for the semantic similarity between two chunks #75 avant\nchunking_embedding_model: \"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #BAAI/bge-m3\" # Embedding model to use for the chunking\n\n# Database cloning parameters\nclone_database: false # Enable or disable the cloning of chunks of an existing index when initializing a database\nclone_persist: \"./data/vector_stores/chroma_semantic_mapnet_v2\" # Directory from which to clone the index\nclone_embedding_model: \"all-mpnet-base-v2\" # Embedding of the index from which to clone\n\n# Search parameters\nnb_chunks: 7 # Number of chunks to retrieve for the database (not taken into account is using reranker target)\nsearch_type: \"similarity\" # Type of search to use for the documents (similarity or exact)\nhybrid_search: false # Enable or disable the hybrid search (keyword + similarity)\nlength_threshold: 40 # Threshold for the length of the chunks (number of words)\nuse_multi_query: false # Enable or disable the multi-query\nadvanced_hybrid_search: true # Enable or disable the advanced hybrid search (semantic + lexical, merging the results)\ndeep_search: false # Enable or disable the advanced RAG answer (Multi step answer generation)\nalpha: 0.5 # Alpha parameter for the Advanced Hybrid Search\nenable_routing: false # Enable or disable the routing of the queries to lexical/semantic search\nsuggestions_enabled: false # Enable or disable the query suggestions\nuse_history: false # Enable or disable the history of the chat to be used for answering the queries\nchat_history: 'no_chat' # Chat history to use for the queries\ndata_sources: {\n \"politique\": \"Political Content, expert opinions on political, geopolitical and economic issues\",\n \"emails\": \"Emails, chat messages, and other text-based communication\",\n \"jobs\": \"job descriptions, or job offers\",\n \"food\": \"Food,food recipes, and cooking instructions\",\n \"linkedin\": \"LinkedIn posts, technical articles, data science and AI content\",\n \"prompts\": \"refined prompts made by the user\",\n}\n\n \n# Search filters\nuse_autocut: true # Enable or disable the autocut\nautocut_beta: 2.5 # Beta parameter for the autocut (number of std that the downward movement is allowed)\nfilter_on_length: true # Filter the chunks based on the length (number of words)\nenable_source_filter: true # Enable or disable the source filter\nword_filter: alice # Filter the results based on the keyword\nsource_filter: [] # Filter the results based on the source\nsource_filter_type: \"$eq\" # Type of source filter (include or exclude)\nfield_filter: [] # Field to filter on (source or length)\nfield_filter_type: \"$eq\" # Type of field filter (include or exclude)\n\n\n# Embedding models parameters\nembedding_model: \"jinaai/jina-embeddings-v3\" #\"Snowflake/snowflake-arctic-embed-l-v2.0\" #\"jinaai/jina-embeddings-v3\" #\"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" # \"BAAI/bge-m3\" #\"jinaai/jina-embeddings-v3\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"BAAI/bge-m3\" #\"all-mpnet-base-v2\" #\"all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"sentence_transformers/all-MiniLM-L6-v2\" #\"BAAI/bge-m3\" #\"ollama/bge-m3\" #\"BAAI/bge-m3\" #Embedding model to use for vectorstore creation, retrieval and semantic chunking\ndense_embedding_size: 1024 #Size of the dense embedding model\nsparse_embedding_model: \"Qdrant/bm42-all-minilm-l6-v2-attentions\" #Embedding model to use for sparse retrieval\nsparse_embedding_size: 30522 #Size of the sparse embedding model\n\n\n# Reranker parameters\nreranker_token_target: 8000 #We give documents to the reranker until this number of tokens is reached (sum of the tokens of the documents)\ntoken_compression: false # Enable or disable the token compressor\nnb_rerank: 5 # Number of documents to rerank (input 10 documents, output 5 documents)\nuse_reranker: true # Enable or disable the reranker\nreranker_model: \"jinaai/jina-reranker-v2-base-multilingual\" #\"BAAI/bge-reranker-large\" #\"jinaai/jina-reranker-v2-base-multilingual\" # Reranker model to us\n\n# Auto-merging parameters\nauto_merging: false # Enable or disable the automatic merging of the chunks\nauto_merging_threshold: 7 # Threshold for the merging of the chunks (number of concomitant sources)\n\n\n# Generation parameters\nllm_provider: \"google\" #\"github\" #\"github\" #\"groq\" #\"sambanova\" #\"sambanova\" #\"groq\" # Provider of the LLM model ('cerebras or ollama or groq') don't forget to change the provider if you change the llm. Pay attention the model name can vary depending on the provider !\nmodel_name: \"gemini-2.0-pro-exp-02-05\" #\"meta-Llama-3.1-405B-Instruct\" #\"llama3-405b\" #gpt-4o\" #llama-3.1-70b-versatile\" #\"llama-3.2-90b-text-preview\" #\"llama3-405b\" # \"llama-3.2-90b-text-preview\" #\"llama3-405b\" #LLM used for inference the name can be in the following list, pay attention that the llm name should correspond to the llm provider: \"hermes3\" (ollama);\"llama-3.1-8b-instant (groq)\"\nmodels_dict: {\"gemini-2.0-flash\":\"google\",\n \"gemini-2.0-pro-exp-02-05\":\"google\",\n \"gemini-2.0-flash-thinking-exp-01-21\":\"google\",\n \"gemini-2.0-flash-lite-preview-02-05\":\"google\",\n \"o3-mini\":\"github\",\n \"deepseek-r1-distill-llama-70b\":\"groq\",\n \"llama-3.3-70b-versatile\":\"groq\",\n \"gpt-4o\": \"github\",\n \"meta-Llama-3.1-405B-Instruct\": \"github\",\n \"gpt-4o-mini\": \"github\",\n \"o1-mini\": \"github\",\n \"o1-preview\": \"github\",\n \"llama3-405b\": \"sambanova\",\n \"llama-3.1-8b-instant\": \"groq\",\n \"llama-3.2-3b-preview\": \"groq\",\n \"DeepSeek-R1\":\"github\",\n }\n\n#Prompting parameters\nfr_rag_prompt_path: \"prompts/base_rag_prompt_fr.txt\" #Path to the file containing the RAG prompt\nen_rag_prompt_path: \"prompts/base_rag_prompt_en.txt\" #Path to the file containing the RAG prompt\n\n#vllm parameters\nvllm_model_name: \"gemini-2.5-pro-exp-03-25\" #VLLM model to use for the generation\nvllm_provider: \"gemini\" # Provider of the VLLM model\nvision_model_dict: { \n \"gemini-2.5-pro-exp-03-25\":\"gemini\",\n \"gemini-2.0-pro-exp-02-05\":\"gemini\",\n \"gpt-4o-mini\": \"github\",\n \"gpt-4o\": \"github\",\n \"gemini-2.0-flash\":\"gemini\",\n \"gemini-2.0-flash-thinking-exp-01-21\":\"gemini\",\n \"phi-4-multimodal-instruct\": \"github\",\n}\n\n#LLM parameters\ncot_enabled: false # Enable or disable the Chain-Of-Thoughts (COT) feature for the LLM answer\nstream: false # Enable or disable the streaming of the generation (if all the output text is generated and then printed in the interface or if the tokens are generated and printed one by one)\ntemperature: 1 # Temperature for the generation\ntop_k: 45 # Top-k for the generation\ntop_p: 0.95 # Top-p for the generation\nmax_tokens: 30000 # Maximum number of tokens for the generation\nllm_token_target: 0 # We give documents to the llm until this number of tokens is reached (sum of the tokens of the documents) , 0 means we don't use this parameter, if autocut is enabled, this parameter is not used !!\nsave_answer: false # Enable or disable the saving of the answer\nprompt_language: \"en\" # Language used for all the prompts (and consequently the LLM)\n\n# EVAL parameters\ntop_k: 5 # Evaluation is valid if the ground truth is in the top k chunks retrieved (you cannot modify this parameter for now)\nevaluate_generation: false # Enable or disable the evaluation of the generation (only evaluate the retrieval if disabled)\nanswer_relevancy_llm: \"gemma2\" # LLM model to use for generating the artificial queries\nanswer_relevancy_provider: \"ollama\" # Provider of the LLM model for the artificial queries\nanswer_relevancy_embedding: \"BAAI/bge-m3\" # Embedding model to use for the artificial queries\nmetrics_list: [\"latency\",\"answer_relevancy\",\"hallucination_score\"] # List of metrics to evaluate, if a metric is not in the list, it is not computed\n\n#Entity/Relation Extraction parameters\nentity_model_name: \"knowledgator/gliner-poly-small-v1.0\" #Entity extraction model, can be in: \"knowledgator/gliner-poly-base-v1.0\" #\"knowledgator/gliner-bi-large-v1.0\" #\"gliner-community/gliner_large-v2.5\" (check on Huggingface for the latest models)\nrelation_model_name: \"knowledgator/gliner-multitask-large-v0.5\" #Relation extraction model, can be in: \"knowledgator/gliner-multitask-large-v0.5\" (check on Huggingface for the latest models)\nallowed_entities_path: \"src/allowed_entities.json\" #Path to the file containing the allowed entity types (you can modify this file)\nallowed_relations_path: \"src/allowed_relations.json\" #Path to the file containing the allowed relations types (you can modify this file)\nallowed_precise_relations_path: \"src/allowed_detailled_relations.json\" #Path to the file containing the allowed precise relations types (you can modify this file)\nentity_detection_threshold: 0.5 #Threshold for the entity detection (be careful when changing this parameter)\nrelation_extraction_threshold: 0.5 #Threshold for the relation extraction (be careful when changing this parameter)\ndisambiguate_threshold: 50 #Threshold for the disambiguation of the entities (be careful when changing this parameter)\n\n#Community summarisation / entity description parameters\ndescription_model_name: \"gemma2:2b\" #LLM model to use for generating the entity descriptions (smaller LLMs are faster but less accurate)\ndescription_llm_provider: \"ollama\" # Provider of the LLM model for the entity descriptions\n\n# Possible actions\nactions_dict: {\n \"artificial voice creation\": \"User asks the system to generate a specific artificial voice or audio for a chartacter.\",\n \"prompt engineering request\": \"User asks the system to generate a detailed textual prompt for an image creation model.\",\n \"adapt image description to template\": \"User asks the system to adapt a given prompt/image_description to a specific prompt template.\",\n \"make images lifelike\": \"User requests to transform images into a more realistic representation, or blockbuster movie style.\",\n \"image2video\": \"User requests to generate a description of a video animation based on an uploaded image, including action and camera descriptions.\",\n \"image2sound\": \"User requests to generate a short prompt or description for sound effects that would match an uploaded image.\",\n \"remove watermark\": \"User requests to remove watermarks or logos from something\",\n \"instagram reel captioning\": \"User wants to add subtitles/captions to an uploaded video file.\"\n}\n\n#Agentic RAG parameters\nquery_breaker_model: \"o1-mini\"\nquery_breaker_provider: \"github\"\n\n#Intent classifier parameters\nquery_classification_model: \"qwen-2.5-32b\" # \"llama-3.3-70b-versatile\"\nquery_classification_provider: \"groq\"\n\n\n# Watermark removal settings\nwatermark_removal_output_dir: \"C:/Users/Yanis/Documents/AI Art/watermarked_removed\"\nwatermark_margin_percentage: 10\nwatermark_detection_threshold: 0.2\n\n# Captioned video output settings\ncaptioned_videos_output_dir: \"output/captioned_videos\""}}}} |